HK40114579A - Method and apparatus for encoding and decoding one or more portions of an image and storage medium - Google Patents

Description

Methods and apparatus for encoding and decoding one or more portions of an image, as well as storage media.

(This application is a divisional application of the application filed on November 29, 2019, with application number 2019800852695 and invention title "Adaptive Loop Filter with Nonlinear Limiting (ALF)".)

Technical Field

This invention relates to the encoding or decoding of blocks of video components. Embodiments of the invention reveal specific, but not exclusive, uses in controlling filters used to filter samples of such components. In particular, but not exclusively, is the control of adaptive loop filters.

Background Technology

Video coding comprises image coding (an image is equivalent to a single frame of video). In video coding, coding tools such as quantization of transform coefficients or motion compensation (often done using interpolation filters) frequently introduce distortion biases/effects (distortions that appear regular or at least non-random in a given context). To compensate for these biases/artifacts and improve (or at least maintain) coding efficiency, certain coding tools are used, called post-filters or in-loop filters. Deblocking filters (DBF), sample adaptive shift (SAO) filters, or adaptive loop filters (ALF) are some examples of such coding tools. Applying in-loop filters within the coding loop increases the image quality they provide to the current frame, thereby increasing the coding efficiency of the next frame encoded based on the current frame. For example, quantization of DCT coefficients is efficient for video compression, but it often introduces block artifacts (biases) at the boundaries of compressed sample blocks. Deblocking filters reduce the undesirable effects caused by these artifacts. Deblocking frames within the coding loop (using DBF) (before the frame is used as a reference frame for motion compensation of another frame) significantly increases the coding efficiency of motion compensation compared to deblocking frames outside the coding loop (e.g., just before the display frame).

This invention specifically relates to an adaptive loop filter (ALF), which also functions as an in-loop filter to reduce unwanted compression artifacts in decoded frames/images. The ALF has been studied by the Video Coding Experts Group/Moving Picture Experts Group (VCEG/MPEG) standardization group and is being considered for use in the Universal Video Coding (VVC) standard, for example, in version 3 of the VVC test model software (VTM-3.0 or VVC draft version 3).

Summary of the Invention

The purpose of embodiments of the present invention is to solve one or more problems or disadvantages of the aforementioned encoding or decoding of blocks of video components.

While ALF is an efficient encoding tool, its linear filtering is a suboptimal solution for improving image quality. This invention enables the use of nonlinear filtering within ALF to improve its efficiency and/or performance.

According to aspects of the invention, devices/apparatus, methods, programs, computer-readable storage media, and signals are provided as set forth in the appended claims. Other features of the invention will be apparent from the dependent claims and the description.

According to a first aspect of the invention, a method is provided for controlling an adaptive loop filter for one or more image portions of an image, the method comprising controlling filtering of a first sample based on one or more neighboring sample values of a first sample value of the image portion, wherein the control uses a nonlinear function having one or more neighboring sample values as variables. Suitably, the variables of the nonlinear function include two or more neighboring sample values. Suitably, the variables of the nonlinear function also include the first sample value and a first variable depending on the position of the one or more neighboring sample values. Suitably, the first variable depends on the position of the two or more neighboring sample values. Suitably, the output of the nonlinear function is used as the input (or input parameter) of the adaptive loop filter.

According to a second aspect of the invention, a method is provided for controlling a filter for one or more image portions of an image, the method comprising controlling filtering of the first sample based on one or more neighboring sample values of a first sample value of the image portion, wherein the control uses a nonlinear function having a plurality of variables including the first sample value, one or more neighboring sample values, and a first variable, the first variable depending on the position of the one or more neighboring sample values. Suitably, the plurality of variables includes two or more neighboring sample values. Suitably, the first variable depends on the position of the two or more neighboring sample values. Suitably, the filter is an adaptive loop filter.

According to a third aspect of the invention, a method is provided for controlling a filter for one or more image portions of an image, the method comprising controlling filtering of the first sample based on one or more neighboring sample values of a first sample value of the image portion, wherein the control uses one or more limiting functions having one or more control parameters based on the first sample value, two or more neighboring sample values, and a limiting parameter. Suitably, the one or more control parameters are based on the first sample value, two or more neighboring sample values, and a limiting parameter. Suitably, the filter is an adaptive loop filter. Suitably, each of the one or more limiting functions is one of the following: max(-b,min(b,d)), min(b,max(-b,d)), max(c-b,min(c+b,n)), or min(c+b,max(c-b,n)); and c is the first sample value, n is the neighboring sample value, d = n-c, and b is the limiting parameter.

According to a fourth aspect of the invention, a method is provided for controlling an adaptive loop filter for one or more image portions of an image, the method comprising controlling filtering of a first sample based on one or more neighboring sample values of a first sample value of the image portion, wherein the control uses a nonlinear combination of the first sample value and one or more neighboring sample values as input parameters to the adaptive loop filter. Suitably, the adaptive loop filter is as specified in VTM3.0.

According to a fifth aspect of the invention, a method is provided for controlling an adaptive loop filter for one or more image portions of an image, the method comprising controlling filtering of a first sample based on two or more neighboring sample values of a first sample value of the image portion, wherein the control uses a nonlinear combination of the two or more neighboring sample values as input parameters to the adaptive loop filter. Suitably, the adaptive loop filter is as specified in VTM3.0.

For the fourth and fifth aspects of the invention, the following features may be provided according to embodiments thereof. Suitably, the nonlinear combination is part of a nonlinear function. Suitably, the input parameters of the adaptive loop filter also include a first variable that depends on the positions of one or more neighboring sample values. Suitably, the first variable depends on the positions of two or more neighboring sample values. Suitably, the nonlinear combination is a first sample value, one (or two or more) neighboring sample values, and the first variable.

According to a sixth aspect of the invention, a method is provided for processing one or more portions of an image, the image portions having associated chroma samples and luminance samples, wherein the method includes determining at least one of the following based on information obtained from a bitstream or a first sample value of the image portion and one or more neighboring sample values: using or not using a filter controlled by the method according to the first, second, third, fourth, or fifth aspect; enabling or disabling the use of the filter; or filtering parameters used with the filter when filtering the first sample value. Suitably, the information available from the bitstream includes flags or indexes. Suitably, the information available from the bitstream includes one or more of the following: information for identifying the filter; flags for indicating use or non-use; flags for indicating enable or disable; information about a first variable used with the filter; or information about a limiting parameter for specifying a range of values. Suitably, the information about the first variable is used to specify a value or identify a first function as the first variable. Suitably, the first variable depends on the position of one or more neighboring sample values (or can vary based on the position of one or more neighboring sample values). Suitably, the first variable depends on the position of two or more neighboring sample values.

According to a seventh aspect of the invention, a method for encoding one or more images is provided, the method comprising controlling a filter according to a first aspect, a second aspect, a third aspect, a fourth aspect, or a fifth aspect, or processing a portion of the image according to a sixth aspect. Suitably, the method further comprises: receiving an image; encoding the received image and generating a bitstream; and processing the encoded image, wherein the processing includes control according to the first aspect, the second aspect, the third aspect, the fourth aspect, or the fifth aspect, or processing according to the sixth aspect. Suitably, when subordinate to the sixth aspect, the method further comprises providing said information in the bitstream. Suitably, the method further comprises: selecting a nonlinear function or one or more limiting functions from a plurality of available functions; using the selected function when processing the encoded image; and providing information in the bitstream for identifying the selected function.

According to an eighth aspect of the invention, a method for decoding one or more images is provided, the method comprising controlling a filter according to a first aspect, a second aspect, a third aspect, a fourth aspect, or a fifth aspect, or processing according to a sixth aspect, for one or more portions of the image. Suitably, the method further comprises: receiving a bit stream; decoding information from the received bit stream to obtain an image; and processing the obtained image, wherein the processing includes control according to the first aspect, the second aspect, the third aspect, the fourth aspect, or the fifth aspect, or processing according to the sixth aspect. Suitably, when subordinate to the sixth aspect, the method further comprises obtaining the information from the bit stream. Suitably, the method further comprises: obtaining information from the bit stream for identifying a nonlinear function or one or more limiting functions from a plurality of available functions; and using the identified function when processing the obtained image.

According to a ninth aspect of the invention, an apparatus is provided for controlling a filter for one or more portions of an image, the apparatus comprising a controller configured to perform a method according to a first aspect, a second aspect, a third aspect, a fourth aspect, a fifth aspect, or a sixth aspect.

According to a tenth aspect of the invention, an apparatus for encoding an image is provided, the apparatus including a control device according to a ninth aspect. Suitably, the apparatus is configured to perform the method according to a seventh aspect.

According to an eleventh aspect of the invention, an apparatus for decoding an image is provided, the apparatus including a control device according to a ninth aspect. Suitably, the apparatus is configured to perform the method according to an eighth aspect.

According to a twelfth aspect of the invention, a method is provided for controlling an adaptive loop filter for one or more image portions of an image, the method comprising controlling filtering of a first sample based on a plurality of neighboring sample values of a first sample value of the image portion, wherein the control comprises using a nonlinear function having one or more variables based on one or more of the neighboring sample values. It should be understood that neighboring samples (values) are not limited to adjacent samples (values), but also include samples (values) around or near the first sample (value). Suitably, the control comprises using one or more neighboring sample values from one or more other neighboring samples not used for determination or as variables in the nonlinear function as input parameters to the adaptive loop filter. Suitably, the neighboring samples whose values are used for determination or as variables in one or more (or the) nonlinear function are arranged in a shape such that they intersect with the first sample at an intersection point; or form a parallelogram. Suitable, the neighboring samples whose values are used to determine or serve as variables in one or more (or the) nonlinear functions are arranged in the following shape: when the values of 8 neighboring samples are used to determine or serve as variables in the nonlinear function, an intersection with the first sample at the intersection point, consisting of 5 samples high and 5 samples wide, or a hollow parallelogram with sides of 3 samples long; or when the values of 12 neighboring samples are used to determine or serve as variables in one or more (or the) nonlinear functions, a parallelogram with sides of 3 samples long. Suitable, the intersection with the first sample at the intersection point (and/or center) is one of the following: a "+" intersection of vertical and horizontal lines; or an "X" intersection of diagonal lines. Suitable, the (hollow) parallelogram is one of the following: a square; a rectangle; or a rhombus shape. Suitable, the (hollow) parallelogram surrounds or encloses the first sample located at the center. Optionally, the neighboring samples whose values are used to determine or represent variables in one or more (or the) nonlinear functions are arranged in the following shape at the center of the first sample: a vertical line "|"; a horizontal line "-"; a diagonal/slash "\" from the top left to the bottom right; or a diagonal/slash "/" from the top right to the bottom left. Optionally, the neighboring samples whose values are used to determine or represent variables in one or more (or the) nonlinear functions are arranged in the shape of a (hollow) polygon. Appropriately, the (hollow) polygon surrounds or encloses the first sample located at the center. Optionally, the neighboring samples whose values are used to determine or represent variables in one or more (or the) nonlinear functions are arranged in any combination of the above shapes. Appropriately, the filter variable is shared among two or more neighboring samples. Appropriately, the arrangement shape of the adjacent samples has symmetry about the center. Appropriately, the cross shape, parallelogram shape, or polygon shape has symmetry about the center. Optionally, the arrangement shape of the neighboring samples does not have symmetry about the center. Appropriately, the cross shape or parallelogram shape does not have symmetry about the center. Suitablely, the first sample and the neighboring samples among the plurality of neighboring sample values are arranged in the following shape: intersecting with the first sample at an intersection point; or a parallelogram or a hollow parallelogram. Suitablely, the first sample value and the neighboring samples among the plurality of neighboring sample values are arranged in the shape of a parallelogram: when the plurality of neighboring sample values consist of sample values of 24 neighboring samples, each side has a length of 4 samples; or when the plurality of neighboring sample values consist of sample values of 12 neighboring samples, each side has a length of 3 samples. Suitablely, intersecting with the first sample at an intersection point (and/or center) is one of the following: a "+" intersection of vertical and horizontal lines; or an "X" intersection of diagonal lines. Suitablely, the (hollow) parallelogram is one of the following: a square; a rectangle; or a rhombus shape. Suitablely, the (hollow) parallelogram surrounds or encloses the first sample located at the center. Optionally, the first sample and neighboring samples among the plurality of neighboring sample values are arranged with the first sample centered in the following shapes: a vertical line "|"; a horizontal line "-"; a diagonal/slash "\" from the top left to the bottom right; or a diagonal/slash "/" from the top right to the bottom left. Optionally, the first sample and neighboring samples among the plurality of neighboring sample values are arranged in the shape of a (hollow) polygon. Suitablely, the (hollow) polygon surrounds or encloses the first sample located at the center. Optionally, the first sample and neighboring samples among the plurality of neighboring sample values are arranged in any combination of the aforementioned shapes. Suitablely, filter variables are shared among two or more neighboring samples. Suitablely, the arrangement shape of adjacent samples has symmetry about the center. Suitablely, the cross shape, parallelogram shape, or polygon shape has symmetry about the center. Optionally, the arrangement shape of neighboring samples does not have symmetry about the center. Suitablely, the cross shape, parallelogram shape, or polygon shape does not have symmetry about the center. Suitablely, the first sample and the neighboring samples are luminance component samples. Optionally, the first sample and the neighboring samples are chromaticity component samples. Appropriately, the variables of the (or one or more) nonlinear functions also include the first sample value, and the differences between the first sample value and the respective neighboring sample values among the one or more neighboring sample values to which the (or one or more) nonlinear functions are applied. Appropriately, the variables of the (or one or more) nonlinear functions also include the first sample value and one or more filter variables depending on the position of the one or more neighboring samples, wherein the respective filter variables are the same for two or more neighboring samples; and the (or one or more) nonlinear functions are applied to the sum of the following differences: the difference is two or more differences between the first sample value and the respective neighboring sample values of two or more neighboring sample values among two or more neighboring samples having the same filter variables. Appropriately, the output of the (or one or more) nonlinear functions is used as the input (or input parameter) of an adaptive loop filter. Appropriately, the nonlinear function includes one or more limiting functions, and each of the one or more limiting functions is one of the following: max(-b,min(b,d)), min(b,max(-b,d)), max(c-b,min(c+b,n)), min(c+b,max(c-b,n)), max(-b,min(b,d1+d2)), min(b,max(-b,d1+d2)), max(2*c-b,min(2*c+b,n1+n2)) or min(2*c+b,max(2*c-b,n1+n2)), where c is the first sample value, n or n1 or n2 are the neighboring sample values, d = n-c, d1 = n1-c, d2 = n2-c, and b is the limiting parameter.

According to a thirteenth aspect of the invention, a method is provided for processing one or more portions of an image, the image portions having associated chroma samples and luminance samples, wherein the method includes determining at least one of the following based on information obtained from a bitstream or a first sample value and one or more neighboring sample values: using or not using a filter controlled by the method of the twelfth aspect; enabling or disabling the use of said filter; or filtering parameters or filter variables used with said filter when filtering the first sample value. Suitably, the information obtained from the bitstream includes a flag provided for one of the luminance or chroma components, the flag indicating at least one of the following for that component: using or not using a filter controlled by the method of the twelfth aspect; or enabling or disabling the use of said filter. Suitably, a flag is provided in an adaptive parameter set. Suitably, the information obtained from the bitstream includes a flag provided for one or more image portions, the flag indicating at least one of the following for one or more image portions: using or not using a filter controlled by the method of the twelfth aspect; or enabling or disabling the use of said filter. Suitably, a flag is provided in an adaptive parameter set.

According to a fourteenth aspect of the invention, a method for encoding one or more images is provided, the method comprising controlling a filter according to the method of the twelfth aspect, or processing a portion of the image according to the method of the thirteenth aspect. Suitably, the method further comprises: receiving an image; encoding the received image and generating a bitstream; and processing the encoded image, wherein the processing includes controlling the filter according to the method of the twelfth aspect or processing according to the method of the thirteenth aspect. Suitably, the method (when subordinate to the thirteenth aspect) further comprises providing the information in the bitstream. Suitably, the nonlinear function includes one or more limiting functions; the variables of the nonlinear function (or the one or more) further include one or more filter variables depending on the location of one or more neighboring samples whose values are used as variables in the nonlinear function (or the one or more); providing the information includes providing one or more limiting parameters in the bitstream; and when the filter variable for a location is zero, the limiting parameters used with the limiting function applied to the neighboring sample values of neighboring samples at that location are not provided in the bitstream. Appropriately, the method also includes: selecting a nonlinear function or one or more limiting functions from a plurality of available functions; using the selected function when processing the coded image; and providing information in the bit stream for identifying the selected function.

According to a fifteenth aspect of the invention, a method for decoding one or more images is provided, the method comprising controlling a filter according to the method of the twelfth aspect, or processing a portion of the image according to the method of the thirteenth aspect. Suitably, the method further comprises: receiving a bit stream; decoding information from the received bit stream to obtain an image; and processing the obtained image, wherein the processing includes controlling a filter according to the method of the twelfth aspect, or processing according to the method of the thirteenth aspect. Suitably, the method (when subordinate to the thirteenth aspect) further comprises obtaining the information from the bit stream. Suitably, the nonlinear function includes one or more limiting functions; the variables of the nonlinear function (or the one or more) further include one or more filter variables depending on the location of one or more neighboring samples whose values are used as variables in the nonlinear function (or the one or more); and the limiting function is not applied to neighboring sample values at a location when the filter variable used at a location is zero. Suitably, the method further comprises: obtaining information from the bit stream for identifying the nonlinear function or the one or more limiting functions from a plurality of available functions; and using the identified function when processing the obtained image.

According to a sixteenth aspect of the invention, an apparatus is provided, the apparatus comprising: a controller configured to perform the methods of the twelfth or thirteenth aspect; an encoder configured to perform the methods of the fourteenth aspect; or a decoder configured to perform the methods of the fifteenth aspect.

According to a seventeenth aspect of the invention, there is provided an apparatus for controlling a filter for one or more portions of an image, the apparatus including a controller configured to perform the method of the twelfth or thirteenth aspect.

According to the eighteenth aspect of the invention, an apparatus for encoding an image is provided, the apparatus including the control means of the seventeenth aspect. Suitably, the apparatus is configured to perform the method of the fourteenth aspect.

According to a nineteenth aspect of the invention, an apparatus for decoding an image is provided, the apparatus including the control device of the seventeenth aspect. Suitably, the apparatus is configured to perform the method of the fifteenth aspect.

According to the twentieth aspect of the invention, a program is provided that, when run on a computer or processor, causes the computer or processor to execute the method according to the first, second, third, fourth, fifth, sixth, seventh, eighth, twelfth, thirteenth, fourteenth, or fifteenth aspect.

According to a twenty-first aspect of the present invention, a computer-readable storage medium is provided for storing a computer program according to the twentieth aspect.

According to a twenty-second aspect of the invention, a signal is provided carrying an information dataset for an image encoded using the method according to the seventh or fourteenth aspect and represented by a bit stream, the image comprising a set of reconstructable samples, each reconstructable sample having a sample value, wherein the information dataset includes control data for controlling filtering of a first reconstructable sample based on the sample values of neighboring samples of the first reconstructable sample.

With respect to the above aspects of the invention, the following features may be provided according to embodiments thereof. Suitably, the nonlinear function includes one or more of a limiting function, a sawtooth kernel function, a triangular kernel function, or a Gaussian kernel function. Appropriately, the nonlinear function includes one or more limiting functions, and each of the one or more limiting functions is one of the following: max(-b,min(b,d)), min(b,max(-b,d)), max(c-b,min(c+b,n)), min(c+b,max(c-b,n)), max(-b,min(b,d1+d2)), min(b,max(-b,d1+d2)), max(2*c-b,min(2*c+b,n1+n2)) or min(2*c+b,max(2*c-b,n1+n2)), where c is the first sample value, n or n1 or n2 are the neighboring sample values, d = n-c, d1 = n1-c, d2 = n2-c, and b is the limiting parameter. Appropriately, the variables are the independent variables of the (mathematical) nonlinear function, and the output of the nonlinear function is its dependent variable. Appropriately, the position of a neighboring sample is defined as its relative distance or relative displacement to the position of the first sample. Appropriately, the position of a neighboring sample is defined as its position in the (raster) scan sequence (appropriately, its position relative to the first sample in the (raster) scan sequence). Appropriately, the adaptive loop filter uses a nonlinear function for filtering. Appropriately, the output of the nonlinear function is used as the input (or input parameter) of the adaptive loop filter. Appropriately, applying the adaptive loop filter to one or more such inputs (parameters) involves using a linear function with one or more inputs (parameters) as its variables. Appropriately, the adaptive loop filter is as specified in VTM3.0.

Another aspect of the invention relates to a program that, when executed by a computer or processor, causes the computer or processor to perform any of the methods described above. This program may be provided separately, or it may be carried on, through, or within a carrier medium. The carrier medium may be non-transitory, such as a storage medium, particularly a computer-readable storage medium. The carrier medium may also be transient, such as a signal or other transmission medium. This signal may be transmitted via any suitable network, including the Internet.

Another aspect of the invention relates to a camera that includes a device according to any of the foregoing device aspects. According to yet another aspect of the invention, a mobile device is provided that includes a device according to any of the foregoing device aspects and/or a camera embodying the aforementioned camera aspects.

Any feature in one aspect of the invention may be applied to other aspects of the invention in any suitable combination. In particular, a method aspect may be applied to an apparatus aspect, and vice versa. Furthermore, features implemented in hardware may be implemented in software, and vice versa. Any references herein to software and hardware features should be interpreted accordingly. Any apparatus feature described herein may also be provided as a method feature, and vice versa. As used herein, device plus functional features may alternatively be represented according to their respective architectures, such as a suitably programmed processor and associated memory, etc. It should also be understood that specific combinations of the various features described and defined in any aspect of the invention may be implemented independently and/or provided and/or used.

Attached Figure Description

Embodiments of the invention will now be described by way of example only and with reference to the following figures, wherein:

Figure 1 shows the location of ALF in a typical decoding loop of VTM-3.0;

Figure 2 is a flowchart outlining the syntax elements of ALF in VTM-3.0;

Figure 3-a is a flowchart illustrating the steps for filtering chroma components according to an embodiment of the present invention;

Figure 3-b shows the filter shape and coefficient arrangement of a chromaticity filter according to an embodiment of the present invention;

Figure 4-a is a flowchart illustrating the steps for filtering the luminance (Luma) component according to an embodiment of the present invention;

Figure 4-b shows the filter shape and four possible coefficient arrangements of the brightness filter according to an embodiment of the present invention;

Figure 5 is a flowchart outlining the modified syntax elements according to an embodiment of the present invention;

Figure 6 is a block diagram of the linear filter in ALF in VTM-3.0;

Figure 7 is a block diagram of a nonlinear filter according to an embodiment of the present invention;

Figure 8 is a block diagram of a linear filter obtained according to an embodiment of the present invention when the nonlinear portion of the filter in Figure 7 is disabled;

Figure 9 is a flowchart illustrating ALF encoding processing according to an embodiment of the present invention;

Figure 10 is a flowchart illustrating the steps of an encoding method according to an embodiment of the present invention;

Figure 11 is a flowchart illustrating the steps of a decoding method according to an embodiment of the present invention;

Figure 12 is a block diagram schematically illustrating a data communication system that can implement one or more embodiments of the present invention;

Figure 13 is a block diagram illustrating the components of a processing apparatus that can implement one or more embodiments of the present invention;

Figure 14 is a diagram illustrating a network camera system that can implement one or more embodiments of the present invention;

Figure 15 is a diagram illustrating a smartphone that can implement one or more embodiments of the present invention;

Figure 16-a provides a 7×7 diamond filter shape with a reduced number of limiting positions according to an embodiment of the present invention; and

Figure 16-b provides a 5×5 diamond filter shape with a reduced number of limiting positions according to an embodiment of the present invention.

Detailed Implementation

The embodiments of the present invention described below relate to improved image encoding and decoding.

In this specification, "notifying with a signal" can refer to inserting (providing/including/encoding in) or extracting/obtaining (decoding) information related to one or more parameters used to control the filter (e.g., using, not using, enabling or disabling modes/schemes or other filter control related information) into (providing/including/encoding in) the bit stream.

In this specification, the term "slice" is used as an example of an image portion (other examples of such an image portion would be a block or block group (which is a group/set of blocks). It should be understood that embodiments of the invention can also be implemented based on image portions (e.g., blocks or block groups) and appropriately modified parameters/values/syntax (such as the header of the image portion/block/block group (instead of the slice header), the type of the image portion/block/block group (instead of the slice type), and the statistics of the image portion/block/block group (instead of the slice statistics)) instead of slices. It should also be understood that in embodiments of the invention, an Adaptive Parameter Set (APS) or a block (group) header can also be used instead of a slice header or sequence parameter. The Set of Photons (SPS) is used to signal the ALF parameters (or information for using ALF filtering). When the APS is used to signal the ALF parameters (or information for using ALF filtering), the slice header or block group header can be used, for example, by indicating the Adaptive Set Identifier (aps_id) to indicate which APS must be used to obtain the ALF parameters (or information for using ALF filtering). It should also be understood that any of the following can be referred to as an image portion: slice, block group, block, coding tree unit (CTU)/maximum coding unit (LCU), coding tree block (CTB), coding unit (CU), prediction unit (PU), transform unit (TU), or pixel/sample block.

It should also be understood that when a filter or tool is described as "valid," the filter/tool is "enabled," "available," or "used"; when described as "invalid," the filter/tool is "disabled," "unusable," or "unused"; and "class" refers to a group, grouping, category, or classification of one or more elements. Furthermore, it should be understood that when a flag is described as "valid," it means that the flag indicates that the associated filter/tool is "valid."

Adaptive Loop Filter (ALF)

Figure 1 illustrates the location of the ALF in a typical decoding loop of VTM-3.0. In 101, image portions (e.g., slices) are decoded in units of coding tree units (CTUs: the largest coding unit in VVC, typically 128×128 samples/pixel). CTUs are broken down into rectangular blocks or coding units (CUs), which are encoded using a specific prediction scheme/mode and are typically lossy encodings of residual blocks. Due to the use of block-based encoding, block artifacts may be visible at the boundaries between coded blocks. In 102, the decoded image portions are then processed by DBF to reduce/remove these artifacts. Typically, to encode the residuals (blocks) used for block prediction, a DCT-like transform is used to transform the residual values (to compress residual energy across several coefficients), and the transformed coefficients are quantized to reduce encoding costs. This quantization typically introduces some ringing artifacts in the reconstructed blocks (i.e., blocks in the reference frame stored in frame buffer 106). In step 103, the output image portion of the DBF is then processed by the SAO filter, which helps reduce some of these artifacts with lower computational cost. In step 104, the output image portion of the SAO filter is then processed by the ALF. The ALF can further reduce artifacts such as "ringing". The ALF has higher-order error modeling capabilities, but is computationally expensive. The output image portion of the ALF is then sent to an output (e.g., a display or a communication interface for communicating with the display) 105. The output image portion of the ALF can also be placed in a frame buffer 106 (as part of a reference frame stored therein), making it available for time prediction (when using a time prediction tool). This is why the DBF, SAO filter, and ALF are referred to as "in-loop" filters. The encoder can disable some in-loop filters so that they can be bypassed at the decoding point (i.e., no filtering is performed, and the output of the step corresponding to the disabled tool is the same as its input). Furthermore, in some cases, the processed image portion is not limited to a slice, but can be a full frame containing one or more slices, with the possibility of applying filters across slice boundaries (if more than one exists) to reduce artifacts at these boundaries. For multi-component images (e.g., images in YCrCb format), DBF, SAO filters, or ALF processing are applied to each component separately and may differ (e.g., using different filtering parameters than other components).

Figure 2 provides an overview of the syntax elements present in VTM-3.0 for ALF. The Sequence Parameter Set (SPS) uses flags to indicate whether the ALF tool is valid (i.e., enabled) for a video sequence (201), and if so, the slice header indicates whether the ALF is valid for a slice (202) and provides filter parameters (203 to 207) for controlling the ALF. When the ALF is valid in 202, it is valid for at least the luma component, and the slice header further indicates whether the ALF is valid for each chroma component (203).

ALF can use more than one filter for the luminance component. More than one filter coefficient table can be used because different filter configurations can share the same filter coefficients (described later). In the syntax element scheme of the VTM-3.0 software, these different filter configurations cannot be distinguished individually, and filters with different configurations are considered the same filter. In the following description, the same reference for individual filters is used unless explicitly referenced to a different coefficient table. This sharing of individual filters in the syntax element scheme is an efficient way to reduce the number of bits allocated to filters by taking into account that the statistics of the filtered image are the same for rotated orthogonal and/or mirrored configurations: that is, a reference filter designed for a reference sample configuration is specified as being rotated and/or mirrored to filter orthogonal and/or mirrored sample configurations. The chip header contains the number (204) of coded luminance filters (one or more).

When filtering luminance samples, the ALF locally classifies (categorizes) the samples (based on the configuration of neighboring samples) into one of 25 possible classes (categories/classifications/groups) to select a filter (which is associated with/assigned to that specific class/category/classification/group) to be locally applied to those samples. The terms "local" classification and "local" application are used here because samples are processed in blocks (typically 4×4 samples, e.g., in VTM-3.0) or units of CUs. The selected luminance filter is then signaled in the bitstream, for example, using an index/information to identify it. When the number of luminance filters used during encoding processing is greater than one, the chip header also contains one or more indices for identifying/selecting these luminance filters (e.g., indices for up to 25 filters across 25 classes, each corresponding to a luminance filter used for one of those classes) (205). In the case where the number of luminance filters used in encoding processing is one, that single luminance filter is applied to all classes.

The header then contains all the filter coefficients (or filter parameters) for each luma filter (206) (i.e., a table of coefficients for the individual luma filters used during encoding), followed by the filter coefficients for the chroma filter (207). Note that in VTM-3.0, when ALF is active (i.e. enabled) for both chroma components, the two chroma components share the same chroma filter.

When ALF is active, filtering can be enabled per CTU for each ALF-active component. Within the coded bitstream, for each CTU, the coded slice data includes an entropy-coded flag (208, 209, and 210) indicating whether ALF is enabled for that CTU's component, and thus whether ALF filtering of that component's samples for that CTU must be used. This flag is encoded using Context Adaptive Binary Arithmetic Coding (CABAC).

In VTM-3.0, the coefficients of the luminance filter 206 are communicated via signal as follows:

[1] First, the “alf coefficient increment flag” is signaled, which indicates whether certain filters can be disabled, and if the “alf coefficient increment flag” is zero and if the number of brightness filters is greater than 1, the “coeff increment prediction mode flag” is signaled, indicating that the filter coefficient encoding will use prediction (to be described in more detail later).

[2] In VTM-3.0, in order to encode the luminance coefficients using (exp-)Golomb codes, we use a 3 (2 for chrominance) (exp-)Golomb configuration. The only parameter of the (exp-)Golomb code is the (exp-)Golomb exponent (often denoted as "k"). Each configuration has a (exp-)Golomb index. The parameters of the (exp-)Golomb code for the filter coefficients are signaled using a variable-length code (VLC) of the "minimum exponent" of the (exp-)Golomb code, and then for each (exp-)Golomb index, if the (exp-)Golomb exponent must be increased for that index and the following indices (starting with the "minimum exponent" for the first index), a signal is used to indicate this.

[3] Then, if the “alf coefficient increment flag” indicates that some filters are enabled, the flag is signaled for each filter to indicate whether the filter is disabled (and therefore not encoded).

[4] Then, the filter coefficients are signaled for each (not disabled) filter using (exp-)Golomb codes (for signed integers), where the (exp-)Golomb codes have (exp-)Golomb exponents taken from the table storing (exp-)Golomb parameters and associated with the (exp-)Golomb index of the filter coefficients (in a fixed table).

When the "coeff incremental prediction mode flag" is signaled, it instructs the filter coefficient encoding to use prediction. This means that the filter coefficients of the "current" filter, which has a filter index greater than or equal to 1 (i.e., the index identifying each filter), are encoded as the difference between the filter coefficients of the "current" filter and the filter coefficients of previously processed filters (e.g., using the filter coefficients of previously processed filters as filter coefficient predictors—that is, using prediction, encoded as filter coefficient residual values). The first filter (with a filter index of zero) is encoded without prediction.

The coefficients of the chroma filter 207 are signaled in a similar manner to those for the luminance filter, except that there is no "alf coefficient increment flag", no "coeff increment prediction mode flag", and two (exp-)Golomb indices instead of three.

Figure 3-a is a flowchart illustrating the steps for filtering chromaticity components according to an embodiment of the present invention, and Figure 3-b provides the filter shape and coefficient arrangement of the chromaticity filter according to an embodiment of the present invention.

In Figure 3-a, ALF filtering is applied differently between the luminance and chrominance components. Starting with a simple sample, Figure 3-a provides the main steps for filtering the chrominance component. The input image portion 301 (e.g., a block or group of blocks) contains the chrominance samples to be filtered. The input filter parameters 302 include ALF parameters described with reference to the ALF syntax elements in Figure 2 (e.g., "alf coefficient increment flag", "coeff increment prediction mode flag", filter coefficients, or any other flags or filter parameters of ALF). The chrominance filter is derived/obtained in 303 using the encoded filter coefficients of the chrominance filter.

In the variant, the chroma filter has a rhombus shape/pattern/mask/support of size 5×5 (i.e., 5 sample heights and 5 sample widths (see 306 in Figure 3-b)). The filter coefficients of the chroma filter are organized such that the chroma filter has symmetry about the center. There are 6 encoded filter coefficients for this chroma filter, with indices numbered from 0 to 5, located at the corresponding coefficient positions shown in the figure. Superscript signs ("+" or "-") are provided for the index numbers to distinguish between different coefficient positions that share the same filter coefficient, i.e., when referring to two symmetrical (in the sense that they share the same filter coefficient) adjacent positions: i ⁺ corresponds to a neighboring sample with an index that is encoded/decoded/processed/stored/accessed in (raster) scan order after the sample to be filtered (i.e., at the center of the filter shape/pattern/mask/support); and ^i- corresponds to a neighboring sample with an index that is encoded/decoded/processed/stored/accessed in (raster) scan order before the sample to be filtered. The seventh coefficient (coefficient position with index number 6) used for the center of the chroma filter shape is derived/derived from the other filter coefficients. The value of this seventh coefficient at coefficient position 6 is equal to 1 - 2.∑ _{i < 6} w _i (estimated using fixed-point computation (w _i = filter coefficient at index i)). That is, 1 minus twice (because of symmetry) the sum of all filter coefficients for coefficient positions 0-5. Therefore, the sum of all coefficients in the diamond shape, including the seventh coefficient, is 1 (1 << 7) with a left 7-bit offset, where 7 is the bit precision of the encoded filter coefficients used for fixed-point computation minus one. This is why only half of "total number of filter coefficients minus 1" is the number of filter coefficients encoded in the bitstream.

For each chromaticity component sample that is valid (i.e. applied) by ALF, the filtering in step 304 is performed as follows to obtain the output image portion 305: For each current sample of the input image portion,

If the current sample belongs to a CTU where ALF is enabled, and if the neighboring samples of the current sample are available to obtain the filter coefficients at the coefficient positions in the rhombus shape of the Chroma filter (where the current sample is located at the center of the rhombus shape) (e.g., using boundary extension on the image boundary, or using neighboring patch samples if they are available on the patch boundary); then the output filtered sample at the same position as the current sample I(x,y) at position (x,y) is equal to:

O(x,y)=((1＜＜(N-1))+∑ _(i,j) w(i,j).I(x+i,y+j))>>N, (1)

Where i & j are two integer displacements (horizontal and vertical) relative to the center of the filter shape (i.e., the position of the current sample at (x, y)), w(i, j) is the filter coefficient at displacement (i, j), N = 7 is the number of integer approximations of the fractional part of the real number (fixed-point representation) used in the representation of the filter coefficient w(i, j) at displacement (i, j), I(x+i, y+j) is the input sample value at displacement (i, j) relative to the current sample position (x, y), and O(x, y) is the output filtered sample value at position (x, y). a << N means applying an N-bit leftward displacement to the integer value a. This is equivalent to multiplying the integer by 2 to the power of N. a >> N means applying an N-bit rightward displacement to the integer value a. Here, since the result of the sum within the parentheses of equation (1) is positive in most cases, this is equivalent to dividing the integer by 2 to the power of N. For negative numbers, the sign propagates as the shift is made to the right, so negative numbers will remain negative (at least -1). There is no non-integer part because the ALF output is typically limited between 0 and 2 raised to the power of 1 in terms of bit depth. N=7 provides decimal precision fixed in the VVC used for ALF calculations, but other values may be used in other embodiments. Adding (1<<(N-1)) before performing the right shift >> N has the effect of rounding the fixed-point result of the scalar product.

Otherwise, output the current sample value at the current position as is (i.e., do not apply this ALF).

According to a variation of this embodiment, for an ALF filter with a centrally symmetric filter shape, in order to reduce the number of multiplication operations performed when implementing the filter and to simplify the notation, equation (1) can be reformulated as:

Where O _{_n} is the output sample at raster scan order index/position n. By raster scan order index n, we refer to the index n as the sample index increasing from left to right in a row of samples, and then from top to bottom as the rows increase. is the input sample at the same position n as the output sample (and corresponding to the position of the input sample at the center of the filter), is the i-th neighboring input sample (in the raster scan order below n) in the filter's filter shape/pattern/mask/support, and is the neighboring input sample at a mirror-space position symmetrical about the center position. Therefore, a shape symmetrical about the center means that when it is in the filter's shape/pattern/mask/support, it is also in the same filter shape/pattern/mask/support. w _{_i} is the filter coefficient associated with the neighboring input sample and w_c, w_c is the filter coefficient used for the center input sample, and c is the number of encoded filter coefficients (this is the same as the index value of the center filter coefficient, since it can be estimated from its neighboring filter coefficients and is not encoded). The value of i and its associated position correspond to the index value and the index value with superscript sign ("+" or "-") in the filter shape 306 of Figure 3-b. For example, for c, which is the index of the filter coefficient at the center of the filter shape, i = c = "6".

Figure 4-a is a flowchart illustrating the steps for filtering luminance components according to an embodiment of the present invention, and Figure 4-b provides the filter shape of the luminance filter according to an embodiment of the present invention and four possible coefficient arrangements.

Figure 4-a illustrates the main steps of the filtering process for the luminance component. Input image portion 401 contains luminance samples to be filtered. Input filtering parameters 402 include ALF parameters described with reference to the ALF syntax elements in Figure 2 (e.g., "ALF coefficient increment flag", "COeff increment prediction mode flag", filter coefficients, or any other flags or filter parameters of ALF). Before filtering, the content of the image portion is analyzed in 403. The main objective of this analysis is to allow/enable the determination of local content orientation and activity levels (see step 405). This allows/enables a local evaluation/assessment of whether the content is uniform or has any sharpness (roughly the intensity or contrast of the content), whether the content has a dominant orientation (e.g., based on edge or orientation texture), and what that dominant orientation is. For example, in VTM-3.0, the analysis includes local gradient analysis, which calculates Laplacian values for four orientations (horizontal, vertical, and two diagonals) using every two samples (i.e., quarter samples) for both horizontal and vertical directions. By segmenting samples of the input image portion into blocks 404 (e.g., 4×4 samples in VTM-3.0), and using the results of the analysis, in step 405, each block is classified into one of 25 possible classes. Each class is identified using an index (i.e., classifying the block into one of 25 categories/classes/groups) based on the Laplacian values calculated for the samples within the block. For example, in VTM-3.0, this corresponds to using 16 Laplacian values (4 orientations across 4 samples). This classification achieves partitioning based on activity, directional intensity, and separates horizontal and vertical orientations from diagonal orientations. Furthermore, in step 405, each block is associated with a transpose index. This transpose index (e.g., “transposeIdx”) can be considered supplementary/additional information to the classification to fully represent/indicate the orientation of the content. There are four possible transpose indices. When the block's class indicates that the block is horizontal or vertical, the transpose index further indicates whether the orientation is north to south, east to west, west to east, or south to north. When the block's class indicates that the block is diagonal, the transpose index further indicates whether the block's orientation is northwest to southeast, northeast to southwest, southwest to northeast, or southeast to northwest.

The class index and transpose index can be viewed as adaptive loop filter parameters for a given sample block. Step 406 uses these parameters to derive the luminance filter that will be used to filter each sample of the block. As described above with reference to Figure 2, in 402, each class is associated with an index in the coefficient table of the luminance filter. To derive the luminance filter for the 4×4 block, the transpose index allows/makes it possible to select one of the four shapes/patterns 409, 410, 411, or 412 shown in Figure 4-b. This pattern indicates how the encoded filter coefficients (e.g., based on the scan (in) order) are organized to construct the luminance filter, which is associated with the class of the block, as illustrated in the description for 204 through 206. The luminance filter has a diamond shape of size 7×7. There are 12 encoded filter coefficients (index numbers 0 through 11) for each luminance filter. The 13th coefficient (index number 12) at the center of the filter shape is derived/derived from the other filter coefficients in the same manner as the center coefficient of the chroma filter described above.

The filtering in step 407 is performed in the same way as the chroma filter to obtain the output image portion 408, that is, for each current sample block, the luminance filter derived/obtained for the current block in 406 is applied to each current sample of the current block.

When filtering brightness samples, due to orientation-based classification (class) and the associated transpose (transposeIdx), the optimal ALF filter for each class (i.e., the "orientation-based filter" selected according to local orientation (e.g., based on local gradients) tends to be a directional filter, especially one with high activity (e.g., high local gradients). For regions with the lowest activity, i.e., regions with blocks belonging to classes with relatively small local gradients, this orientation will be less pronounced, and orientation-based filters for such blocks tend not to be directional filters.

Orientation-based filters are well-suited for filtering around edges without significantly affecting edge sharpness. As described below, classes can be grouped together to reduce the signaling cost of the filter. Even with this grouping, filter optimization typically results in a directional filter. In the case of this orientation-based (ALF) filter, when the local gradient is horizontal, the filter orientation is typically vertical (i.e., the highest coefficients (absolute values) will typically be located above and below the center, while the coefficients to the left and right of the center will typically be close to zero). Similarly, when the local gradient is vertical, the filter orientation is typically horizontal, and when the local gradient is diagonal, the filter orientation is typically perpendicular to the diagonal direction of the local gradient.

When filtering chroma samples in VTM-3.0, the process is the opposite of filtering luminance samples because no classification is used (and since chroma is typically smooth), the filters used are generally more sensitive to all orientations and are not directional filters (or orientation-based filters).

In both Figures 3-b and 4-b, the filter shape is symmetrical about the center pixel. This symmetry has been chosen in the design of the ALF in the VTM software, but in variations of the embodiment, an asymmetrical shape can also be used. Then, instead of sharing 7 coefficients for 13 input samples, the same filter support/mask as in Figure 3-b can use up to 13 coefficients. And instead of sharing 13 coefficients for 25 input samples, the same filter support/mask as in Figure 4-b can use up to 25 coefficients. In other variations, the filter support/mask is even asymmetrical. In these variations, the filter shape/pattern/support/mask is transmitted along with the ALF parameters.

In other variations, the filter shape/pattern/support/mask differs from the rhombus shape. For example, in one variation it is a square, in another it is a rectangle, in yet another it is a hexagon, and in yet another it is an octagon.

In some variations, the filter shape/pattern/support/mask differs for all class orientations. In one variation, the shape is a horizontal rectangle (e.g., "-") for horizontal class/transposeIdx orientation configurations, a vertical rectangle (e.g., "-") for vertical class/transposeIdx orientation configurations, an NW-SE rectangle for Northwest-Southeast (NW-SE, e.g., "\") and Southeast-Northwest (SE-NW) class/transposeIdx orientation configurations, and a NE-SW rectangle for Northeast-Southwest (NE-SW, e.g., "/") and SW-NE class/transposeIdx orientation configurations. In other variations, the filter shape/pattern/support/mask is a horizontal-vertical intersection ("+"), a diagonal intersection ("X"), a vertical segment ("|"), a horizontal segment ("-"), a diagonal segment from top left to bottom right ("\"), a diagonal segment from top right to bottom left ("/"), or any combination of the above filter shapes/patterns/supports/masks.

Figure 6 provides a block diagram of the linear filter in the ALF. It corresponds to an example implementation of the ALF with the filter function corresponding to equation (1), where “C” 601 corresponds to I(x,y) as the sample value of the current sample, “S0” 602 to “Sn” 603 correspond to I(x+i,y+j) as the neighboring sample values (each (i,j)≠(0,0)), “wC” 604 corresponds to w(0,0) as the filter coefficient of the current sample, “w0” 605 to “Wn” 606 correspond to w(i,j) as the filter coefficient of the neighboring sample at position (x+i,y+j) (each (i,j)≠(0,0)), and “O” 607 corresponds to the output of the ALF filter, i.e., the filtered sample value of the current sample. This process takes a table of integer input sample values (i.e., "C" and "S0" through "Sn") and a table of input weights (i.e., "wC" and "w0" through "Wn") (or filter coefficients—note that this table has the same number of elements as the table of integer input sample values) as input to obtain an integer representation of a real number with fixed-point precision. Both tables have the same number of elements and are multiplied element-wise, and the results of these multiplications are summed together (i.e., obtaining a linear combination of the elements of the two tables). The result of this operation is a fixed-point precision approximation of the real number. This fixed-point value is rounded to an integer value, resulting in an output sample value of "0".

From the encoder's perspective, ALF is inspired by the Wiener filter. A Wiener filter is a linear filter (often used as a linear convolution filter in signal/image processing) that minimizes the mean square error between: 1) the estimated random processing/variable (the output of the Wiener filter), which is a linear combination of a finite number of observed processing/variables (the inputs of the Wiener filter), and 2) the desired processing/variable (the target of the Wiener filter, i.e., the original image before artifacts occur). In signal/image processing, for example, Finite Impulse Response (FIR) Wiener filters have applications in source separation or denoising. In the case of image coding, the target is the original image (before it is modified by compression/quantization), while the input is a sample from a compressed image that we wish to improve by applying a filter.

The least-squares solution to X (which is the input matrix of the observed implementations of the random processes, with each column containing an implementation for each random process) and y (which is the output row vector containing the implementations of the desired processes for the observed random processes at the same column indices) is:

Wiener Filter Coefficients Correspond to

It should be understood that "realization" is the observed value or the value of a random variable, that is, the value actually observed in practice.

In the VTM-3.0 ALF encoder, there is a rate/distortion (R/D) trade-off between optimizing the coding cost of ALF parameters (primarily determined by the cost of coding the FIR filter coefficients) using an FIR Wiener filter (or a functionally equivalent least-squares solution) and the distortion gain obtained by filtering the image using the coded ALF parameters (i.e., using an FIR filter). If the rate of coding the ALF parameters (i.e., coding cost) is not an issue, and if maximizing the peak signal-to-noise ratio (PSNR) of a given frame is the sole objective (regardless of temporal effects), then the Wiener filter will be able to achieve the optimal solution for the ALF filtering design in VTM-3.0. Therefore, the ALF encoder according to embodiments of the invention can use the same or similar filters as the linear ALF filters provided with it.

Figure 9 illustrates a flowchart of the ALF encoding process according to an embodiment of the present invention, which is implemented by modifying the ALF of VTM-3.0 described with reference to Figure 6. It should be understood that other ALFs can be modified in the same manner to implement the other embodiments according to the present invention.

ALF encoding processing begins by determining the class index and transpose index of each 4×4 block of the luminance sample 901.

Then, statistics for deriving the Wiener filter are extracted/obtained at 902. These statistics are (automatic) covariance statistics corresponding to ^XXT in Equation (3) and cross-covariance statistics corresponding to ^XyT in Equation (3). They are used to construct/obtain/estimate the (automatic) covariance matrix and cross-covariance matrix by dividing ^XXT and ^XyT by N (the number of columns in X, i.e., the number of samples to be filtered), respectively. These (automatic) covariance matrices and cross-covariance matrices are constructed/obtained/estimated at 902 for each class and for each CTU for each luminance component sample and for each CTU for each chrominance component sample. In the following description, the terms "(cross-)covariance matrix statistics" and "(cross-)covariance matrix" are used interchangeably to refer to the same thing. It should be understood that the difference between the two is that the "(cross)covariance matrix statistics" is obtained by accumulating (or summing) values, while the "(cross)covariance matrix" is also obtained by accumulating (or summing) values, but then normalized by the number of accumulations (to estimate the expected value).

For a given CTU/class/component, X is obtained as follows. Since the shape of the filter is considered symmetrical, the number of rows in X corresponds to the number of filter coefficients. One row (the last row) of X contains the realization of the center sample in the filter shape (where the center of the filter shape belongs to the given CTU/class/component), while each of the other rows with index i contains the sum of two symmetrical samples with index i in the filter shape (where the symmetrical sample is the adjacent position of the center of the filter shape belonging to the given CTU/class/component). In the case of luminance component samples, the transpose index is also used to transpose the sample positions of the filter shape according to the different shapes in Figure 4-b, such that each row i of X contains the sample statistics of the sample with index i belonging to the filter shape. For example, for the shape in Figure 4-b, X_{_i,j} contains: for i < 12, the sum of the two symmetrical adjacent positions (the j-th filter sample) with index i in the shape, and for i = 12, the j-th filter sample; where X _{_i,j} is the value at row i and column j in the matrix.

Vector y contains all the target sample values (i.e., _yj is the value of the j-th sample in the source/original image before its compression).

Instead of actually constructing/compiling the X matrix, we compute XXT by iteratively summing the results, where X i is the i-th column of X obtained for a given sample location.

Nor is the y vector actually constructed/computed. Instead, ^XyT is computed by iteratively summing the results of Xi _{y i} , where y _i is the i-th element of y and corresponds to the target sample value when the filter uses the i-th input sample of input _Xi .

In VTM-3.0, the ALF encoder attempts to reduce R/D costs, which is equivalent to a Lagrangian optimization of D+λR with a rate-distortion tradeoff. Here, D is the distortion (quadratic error), R is the rate, and λ is determined by the encoder primarily based on the VTM-3.0 encoder quantization parameters (QP), slice type (intra-frame or inter-frame), and the type of compressed component (luminance or chrominance). The ALF encoder first attempts to optimize the ALF parameters by minimizing this Lagrangian R/D cost.

If the ALF reduces the cost of the luminance component—that is, if the output distortion when the ALF is disabled is greater than the distortion when the ALF is enabled plus a multiple of λ (the rate required to signal the ALF parameters)—the encoder determines that the ALF is valid for the luminance component. Then, if valid for the luminance component, it attempts to optimize the ALF parameters for the chrominance component to see if it can improve the R/D cost of signaling these chrominance components. Based on this, the encoder can determine whether it is better to enable/enable the ALF for each of these components individually.

According to an embodiment of the invention, the ALF encoder performs the same or functionally equivalent optimization processing on its ALF parameters. According to a variation of this embodiment, at the beginning of the ALF parameter optimization processing, the ALF is set to be valid for all CTUs 903. The slice-level statistics used to construct the covariance matrix and cross-covariance matrix are obtained by aggregating statistics for each CTU where the ALF is valid/enabled. Using statistics obtained for all samples classified as belonging to a 4×4 sample block, a matrix 904 is calculated for each class of the luminance component; and a matrix 912 is calculated by aggregating (summing) statistics for the two chrominance components.

The filter optimization process for brightness begins by finding 25 sets of filters to group/merge classes together, i.e., 25 filters for the first set, 24 filters for the second set, and so on, until a final set of 1 filter (one set for each possible number of filters) 905. The encoder starts with 1 filter for each class (thus 25 filters in total), which is a Wiener filter calculated based on the covariance and cross-covariance of samples from the blocks in that class. This is the first set of 25 filters. It attempts to iteratively reduce the number of filters (one by one, until only one remains to obtain all the desired sets) by merging the filters together (i.e., merging the relevant classes of the filters together for greater precision). The encoder makes classes initially associated with two different filters share a common filter (i.e., the encoder merges the two classes to share a single associated filter). To determine the filters to be merged (i.e., the classes to be merged), covariance and cross-covariance statistics are determined for each filter. The encoder estimates/evaluates the total residual obtained after filtering all sample blocks associated with that class using the filter for each filter associated with one or more classes (indices). Then, for each pair of filters (and their associated classes), the encoder calculates the merge covariance and cross-covariance statistics to determine the Wiener filter for the merged class and the total residual obtained after filtering all sample blocks associated with the class index using the determined Wiener filter. The encoder then determines two filters for which the difference between the total residual of the merged filter (derived from the statistics associated with the class, which is associated with both filters) and the sum of the total residuals of the two filters (by adding the total residuals of one filter to the total residuals of the other filter) is minimized, and merges these two filters (making the merged filter ready for the next filter merging iteration). In short, the encoder merges (a,b) a pair of distinct filter statistics that minimizes the following: where Err(x) returns the error of the filter statistics, and a+b is the merged statistic of the two filter statistics a and b.

So far, double-precision floating-point values have been used to estimate statistics, filters, and errors. The encoder now attempts to find the optimal R/D cost tradeoff for encoding the luminance filters. Once 25 groups/classes of filters (starting with 25 filters down to 1 filter) are determined, for each group, the ALF encoder derives the integer filter coefficients for each filter (for integer value encoding and fixed-point precision calculations). Then, the optimal filter is sought in terms of R/D cost tradeoffs while using different alternative encoding schemes. The first alternative is to encode all coefficients for all filters using (exp-)Golomb coding. The second alternative uses incremental coding of the filters, where the filter coefficients of each filter are encoded differently from the filter coefficients of the previous filters (using (exp-)Golomb coding). The third alternative (R/D optimization) allows disabling some filters using a flag and encoding all filter coefficients for the filters that are not disabled (using (exp-)Golomb coding). The first two alternatives may result in a bit rate reduction, while the third alternative may result in more distortion for a smaller bit rate reduction.

The encoder employs/selects/chooses filter banks and coding trade-offs that minimize R/D costs.

Once the encoder determines/selects the luminance filter to use, for each CTU, the encoder uses CTU statistics to see if the R/D cost of the luminance sample from the filtered CTU is better than the R/D cost of the luminance sample from the same CTU without filtering. If it is not better, ALF is disabled for the luminance sample of that CTU 907. The encoder can then loop back to the luminance filter optimization step for the luminance at 908, while updating the slice statistics 904 regarding the covariance and cross-covariance statistics of the CTU (ALF enabled). For example, in VTM-3.0, the encoder loops 4 times.

Based on the R/D cost difference between applying ALF to the luminance component sample and not applying ALF, the encoder determines at 909 whether to enable/enable ALF for the luminance component.

To enable/enable ALF for the luma component, the encoder proceeds to process the chroma component at 910. To disable/disable ALF for the luma component, a signal is sent at 911 indicating that ALF is disabled/disabled, and the ALF encoding process ends.

The ALF encoding process for chroma components starts with the combined statistics of two chroma components 912, provided that ALF is enabled/enabled for all CTUs of the chip.

The encoder then determines the chroma filter 913. The encoder first determines the (floating-point) Wiener filter using CTU statistics for both chroma components. It derives integer filter coefficients. Then, for each chroma component, the encoder uses CTU statistics to see if the R/D cost of the chroma component with the filtered CTU is better than the R/D cost of the chroma component without the filtered CTU. If not, the ALF is disabled for the chroma component sample for that CTU 914. If the encoder determines that all CTUs for a given chroma component should not make the ALF effective (i.e., should be disabled), the ALF is disabled for that chroma component, and therefore no encoding is required for individual CTUs with the "enable flag" set to 0 915.

The encoder can then loop back to the chroma filter optimization step at 916, updating the slice statistics for the covariance and cross-covariance of the chroma components with CTU (ALF enabled). For example, in VTM-3.0, the encoder loops twice more.

Then, at 917, the encoder applies the ALF according to an embodiment of the invention together with the determined ALF parameters (i.e., using the chroma filter determined from step 913). Depending on the encoder configuration, the resulting image can be output and/or placed in a reference frame buffer. Finally, the encoder encodes the optimal R/D cost parameters, i.e., the ALF enable flag and (if the ALF enable flag indicates active/enabled) the determined ALF parameters 918.

It should be understood that, depending on the variant, optimization processing can be performed on other ALF parameters to optimize these parameters of the ALF.

ALF with nonlinear filtering capability

According to embodiments of the invention, the linear filtering used in steps 304 and 407 can be modified, introducing nonlinearity and improving the filtering results (achieving a better trade-off between filtering quality and coding efficiency). The goal of ALF filtering is to remove some of the "noise" (e.g., quantization noise/error) introduced by the coding tools. To remove this noise, low-pass linear filters are typically used to smooth the signal and reduce small local variations. This type of filter can introduce blurring into the filtered output, especially in areas of high contrast, such as near edges. Nonlinear filters (e.g., bilateral filters) have been developed to allow for more efficient denoising while introducing less blurring or ringing, even around edges. To this end, these nonlinear filters, like linear filters, typically rely on filtering samples based on local proximity (i.e., neighboring samples), but pay more attention to (or weight) samples with values similar to the sample to be filtered compared to samples with very different values. The weighting of neighboring values is typically done using a nonlinear function (i.e., a nonlinear mapping). Such nonlinear filters are generally more complex than linear filters and may be difficult to optimize their parameters, and/or may offer less flexibility than linear filters if this type of filter is to be used in a new ALF design.

According to embodiments of the invention, the ALF of VTM-3.0 (or any of the embodiments described above or variations thereof) is modified by introducing nonlinearities with relatively low complexity involving operations that maintain the parallelizable design of the ALF.

Looking at ALF in VTM-3.0, by simplifying the equation (1) using real number re-representation, we have (removing operations related to fixed-point representation and integer rounding):

O(x,y)＝∑ _(i,j) w(i,j).I(x+i,y+j) (4)

For this ALF, the condition is satisfied.

∑ _{(i, j)} w(i, j)＝1 (5)

This means that the sum of all filter coefficients in ALF is 1. Then it can be seen that equation (4) can be reformulated as:

O(x,y)=I(x,y)+∑ _{(i,j)≠(0,0)} w(i,j).(I(x+i,y+j)-I(x,y)) (6)

Then, the output sample O(x, y) is the result of adding the input sample I(x, y) (at the same position) to the filter coefficient vector and the local gradient vector (which is the vector of local gradients calculated as the difference between the input sample's neighboring samples and the input sample itself). In other words, the output sample (i.e., the filtered sample value) is the result of adding the input sample to the filter coefficients and the local gradient.

Instead of using the usual linear formula for the filter, according to an embodiment of the invention, the ALF filtering process is modified to introduce nonlinearity into the ALF filter. This nonlinearity is achieved using a multivariable function K(d, b) that takes the local gradient d at offset (i, j) as a first parameter, and whose value varies according to a second parameter b = k(i, j). Instead of using the local gradient at offset (i, j) in the scalar product of equation (6), K(d, b) is used in the scalar product to obtain the output sample O(x, y), which varies nonlinearly with the local gradient:

O(x,y)=I(x,y)+∑ _{(i,j)≠(0,0)} w(i,j).K(I(x+i,y+j)-I(x,y),k(i,j)) (7)

Where K(d, b) is a function of d = I(x+i, y+j) - I(x, y) as its first parameter/variable (the local gradient at offset (i, j) calculated as the difference between the neighboring sample value at position (x+i, y+j) and the current sample value at position (x, y)) and b = k(i, j) as its second parameter/variable (an additional filtering parameter). The additional filtering parameter is also determined as w(i, j). In the implementation, the values of k(i, j) and w(i, j) are determined to optimize filtering and signal-informed processing (e.g., to minimize distortion). This optimization is performed by an encoder using ALF. An example of this optimization will be provided in the description later.

Therefore, according to embodiments of the invention, input sample values are filtered by using a linear combination of an adaptive nonlinear transformation of the input sample itself and neighboring input samples. The adaptive nonlinear transformation depends on the relative position of the neighboring input samples with respect to the input sample being filtered.

Using this modified ALF, linear filtering can still be achieved for a specific K, where K satisfies the following condition: there exists b, and there exists a non-zero α, such that for all d, K(d, b) equals α multiplied by d, that is:

Make

Therefore, for certain values of b, a function K(d, b) is chosen that behaves like a linear function (i.e., a linear mapping) (i.e., the function K(d, b) satisfies condition (8)), ensuring that the modified ALF filtering process of the embodiment can still be at least as efficient as using a standard linear ALF (i.e., in the worst case, the modified ALF can achieve the same efficiency level as the ALF in VTM-3.0). For example, when using a limiting function, the parameter b can be set to the largest possible integer value (ideally, it should be set to infinity, but we can achieve the same effect using the largest possible integer value in a finite integer precision scheme) to make it behave like a linear function.

Note that in some embodiments of the invention, ALF filtering uses an alternative/filtering scheme. The multivariate function K(d, b) employing local gradients and additional parameters/variables is replaced by another multivariate function employing three parameters/variables: neighboring sample values, the sample value to be filtered, and additional parameters/variables. It is re-expressed using equation (7) while using a notation similar to that of equation (4) and satisfying condition (5):

O(x, y)=w(0,0).I(x,y)+∑ _(i,j)≠(0,0 )w(i,j).K′(I(x+i,y+j),I(x,y),k(i,j))(9)

Where K′(n, c, b) is a function that takes the neighboring sample value (n), the current sample value to be filtered (c), and the additional filtering parameter/variable (b) as its parameters/variables.

According to an embodiment of the invention, the filtering steps for both the chromaticity component (304) and the luminance component (407) are modified to use a filtering method with a nonlinear formula implemented using integer arithmetic (as described in reference equation (1)) (preferably using equation (7), but alternatively using equation (9)).

According to alternative embodiments, any choice is based on complexity, for example, modifying only one of the two filtering steps (304) or (407) to use a filter with a nonlinear formula.

According to some implementations, the parameter/variable b = k(i, j) has more than one dimension; for example, it can be K(d, b = (S _b , D _b )) = f(d, S _b , D _b ), for example, using:

S _{_b} and D _{_b} can vary with (i, j).

Using parameters/variables/functions with more than one dimension often allows for improved filtering, but signaling the filter parameters is costly. Furthermore, optimizing filter parameters in a higher-dimensional space at the encoder introduces more complexity, and computing functions at the decoder is often more complex (the more filter parameters, the more operations are required to use these functions). In this embodiment, b has a single dimension, which allows for a good trade-off for many ALF applications.

According to some embodiments, function K can also be a function comprising more than one different function for different offsets (i, j). In some embodiments, the configuration/arrangement of functions in the filter shape/pattern is predefined at both the encoder and decoder. Optionally, a configuration of function is selected from a predefined set of configurations, and the index for the selected configuration is transmitted/signaled. In other embodiments, functions are selected from a predefined set of functions for each coefficient index in the filter shape/pattern (except for the center position), and their indexes are transmitted/signaled. In a variation of such an embodiment, the individual functions are univariate functions (i.e., the K function no longer has additional parameters/variables), so instead of signaling the b parameter, the function index is signaled. This variation can be considered as using a multivariate formula, such as K(d, b) = _Kb (d). For example, b is an integer index for selecting the univariate function _Kb from the set of functions.

According to some embodiments, the function K is different for the luminance filter and the chrominance filter. According to some embodiments, the function K is different for each luminance filter. In some of these embodiments, for each filter, an index of the selected function is provided in the chip header.

According to embodiments of the invention, K is chosen to simplify computation (without introducing too much decoding complexity). For example, in such embodiments, K is simply a limiting function:

K(d, b)=max(-b, min(b, d)) (11)

Or equivalent to:

K(d,b)=min(b,max(-b,d)) (12)

Unlike some of the alternative functions described below, the limiting function does not vanish for high values of local gradients (in other words, the limiting function f(x) does not converge to zero as x approaches zero). However, it has been experimentally observed that compression using this simple function is often as efficient as, and even better than, compression using more complex functions. In fact, the limiting function does not vanish for high values of local gradients, but merely limits them, allowing the sharp edge transitions to be considered while limiting the influence of the high variance present around the region of that sharp edge transition. This makes sense since artifacts typically become stronger around sharp transitions.

The equivalent function K′ of the limiting function K using the filtering formula of equation (9) is

K′(n,c,b)=max(c-b,min(c+b,n)) (13)

Or equivalent to:

K′(n,c,b)=min(c+b,max(c-b,n))

The limiting function satisfies equation (8) as long as b is greater than or equal to the maximum possible sample value (e.g., 2 raised to the power of the image bit depth).

In the following description, "limiting range" or "limiting parameter" is used to refer to the b parameter/variable of K or K′. It is understood that this terminology can be considered a general term used to refer to parameters of nonlinear functions. Similarly, "limiting" or "limiting function" can be used to refer to the aforementioned K or K′ or their functional equivalents.

In an alternative embodiment, K can be another nonlinear function. In a variant, K is an antisymmetric function, such as a sawtooth period, corresponding to: or some triangular period corresponding to a specific case of equation (10) with respect to time; or it can also use a Gaussian kernel in a manner similar to a bilateral filter: for example, K can be any function with a threshold set above (or below) zero, as in the antisymmetric function of the variant described above.

According to an embodiment, a non-linear ALF is used instead of a linear ALF to reduce the number of sample line buffers required for ALF filtering (i.e., reduce the number of samples of the input image components that need to be processed/accessed/held in memory when filtering is performed (e.g. at the decoder)).

Depending on the variant, as a trade-off, the size of the filter shape/pattern used for the luminance filter is reduced from a 7×7 rhombus shape to a smaller filter shape/pattern. For example, a 5×5 rhombus shape is used for the luminance filter (e.g., the same shape as the chroma filter in Figure 3-b, but still using the transposed index variant). This still achieves a coding gain similar to that of a linear ALF with a 7×7 rhombus shape luminance filter (e.g., the filter shown in Figure 4-b), but with a reduction in the number of samples to be processed/accessed/stored (e.g., a reduction in the number of sample line buffers), and also a reduction in computational complexity: that is, the number of multiplications required to process the ALF filter is reduced by 6 for each filtered input sample, while achieving good coding gain.

According to one variation, when ALF filtering is applied to a sample, nonlinear ALF is used based on all neighboring samples that will be used with linear ALF. According to another variation, when ALF filtering is applied to a sample, only some of the neighboring samples that will be used with linear ALF are used with nonlinear ALF, and the remaining neighboring samples are used with linear ALF. According to yet another variation, when ALF filtering is applied to a sample, only some of the neighboring samples that will be used with linear ALF are used with nonlinear ALF, and the remaining neighboring samples are not used with linear ALF.

According to the variant, by utilizing the symmetry in these filter shapes/patterns and simplifying the linear and/or nonlinear ALF filter implementations using the same notation as in equation (2), the linear function in equation (4) can be rewritten as:

For ALF, the following conditions are met:

The linear function in equation (6) can also be rewritten as:

And the nonlinear function in equation (7) becomes:

Where k_{_i} is the filter limiting parameter associated with the filter coefficient w_{_i} .

Finally, the nonlinear function in equation (9) can be rewritten as:

According to a variation of this embodiment, the K′ function or the K function is a limiting function.

According to an embodiment, in order to reduce the computational cost involved in processing a nonlinear function compared to the computational cost of processing function K in equation (17), nonlinearity is introduced on the sum of the differences of at least two neighboring functions (i.e., a nonlinear function with the sum of two or more local gradients as variables can be used):

Equation (19) is not always equivalent to equation (17), and the filter using equation (19) may be less efficient, but it reduces computational complexity. Depending on the variant, for example when it is a K-limiting function, the number of limiting parameters/values signaled/encoded remains the same relative to the parameters/values in equation (17).

It should be understood that, according to another variation, a similar derived equation based on equation (18) with reduced complexity can be used, where the K′ function has the sum of two or more neighboring differences (local gradients) as its variable:

Figure 7 provides a block diagram of a nonlinear filter according to an embodiment of the present invention, which may be an alternative to the linear filter in the ALF in VTM-3.0 (or an additional nonlinear filter in addition to the linear filter in the ALF). It corresponds to the implementation of equation (7) with an integer fixed-point algorithm using the limiting function of equation (11). “C” 701 corresponds to I(x, y), “S0” 702 to “Sn” 703 correspond to I(x+i, y+j) (each (i, j) ≠ (0, 0)), “k0” 704 to “kn” 705 correspond to k(i, j) (each (i, j) ≠ (0, 0)), “w0” 706 to “wn” 707 correspond to w(i, j) (each (i, j) ≠ (0, 0)), and 708 corresponds to O(x, y).

Figure 5 is a flowchart outlining the modified syntax elements according to an embodiment of the present invention, providing examples of syntax elements that can be used to implement the nonlinear functions (and their parameters) described above and in their related embodiments/variations. In this embodiment, the coding filter coefficients of each filter in equation (7) (or (9)) are associated with their own limiting range, so k(i,j) can have different values that vary with offset (i,j). Most syntax elements are the same as those already used in VTM-3.0, and are illustrated with reference to Figure 2: 501, 502, 503, 504, 505, 506, 507, 508, 509, and 510 in Figure 5 have the same signaling manner and semantics as 201, 202, 210, 204, 205, 206, 207, 208, 209, and 210 in Figure 2. The new syntax elements are the limiting parameters 511 for each luminance filter, all filter coefficients (506) for each luminance filter, and all limiting parameters (511) for each luminance filter, which can be indicated, for example, by a signal in the title block. This signal is followed by the signal indicating all filter coefficients (507) and all limiting parameters (512) for the chrominance filter.

According to the embodiment, for any filter, the number of limiting parameters signaled is the same as the number of filter coefficients signaled. The limiting parameter offset (i, j) is obtained in the same manner as the filter coefficients at the same position (x+i, y+j). For chroma, they are processed in the same manner as the filter coefficient derivation process described in step 303 of FIG. 3-a (but using chroma limiting parameters); and for luminance, they are processed in the same manner as the filter coefficient derivation process described in step 406 of FIG. 4-a (but using luminance filter limiting parameters).

In an alternative embodiment, each filter has only one limiting parameter, which is used for all filter positions (i, j) ≠ (0, 0).

In an alternative embodiment, the number of limiting parameters is less than the number of filter coefficients. In a variation of this embodiment, these limiting parameters are used for a subset of filter positions with offsets (i, j), which may be predefined or deterministic. For other filter positions, normal linear filtering is performed (or in other words, K is considered an identity function at those other filter positions) or, alternatively, predefined limiting parameter values are used.

Variations of this embodiment are described with reference to Figures 16-a and 16-b. In these variations, the limiting function is applied to a subset of the filter locations; that is, the limiting pattern/shape/support/mask includes only a subset of the filter locations within the filter shape/pattern, such that only a subset of neighboring sample values/locations is limited during the filtering process (e.g., the nonlinear function is applied only to sample values from this subset). Since limiting operations can be computationally expensive, reducing the number of limiting operations involved in filtering in this way can reduce the complexity and computational cost of the filtering process.

Figure 16-a provides a 7×7 diamond filter shape with a reduced number of clipping/limiting positions, illustrating three possible arrangements/configurations of clipping/limiting positions for reducing the number of clipping operations required when filtering with a 7×7 filter shape (e.g., for luminance ALF). By using the clipping pattern shown in Figure 16-a, the number of clipping operations used for filtering is reduced because clipping operations are not applied to all filter (input) positions; that is, only a subset of filter positions (which may be predefined, signaled, or inferred) are used for clipping operations (i.e., used with a nonlinear function). In Figure 16-a, the sample positions for clipping are marked with an "X" in clipping patterns 1601, 1602, and 1603. Clipping patterns 1601 and 1602 reduce the number of clipping operations by two-thirds, while clipping pattern 1603 reduces that number by half. When using a 7×7 diamond filter shape, these limiting patterns 1601, 1602, and 1603 strike a good trade-off between output accuracy and complexity in the filtering process. Limiting patterns 1601, 1602, and 1603 work well for nonlinear ALFs based on orientation-based ALFs (i.e., filters constructed for classification based on local content orientation and activity level, such as the luma filter in VTM 3.0). However, if classification is not used, patterns 1602 or 1603 are preferred.

It should be understood that, depending on other variations, more or less limiting positions can be used. For example, a column and a row (excluding the center position) can be limited instead of limiting pattern 1601 (i.e., intersections from top to bottom and from left to right, excluding the center position), or a position on the outer edge (i.e., a larger rhombus/parallelogram limiting pattern) can be limited instead of limiting pattern 1602. In other variations of the limiting position, the limiting pattern/shape/support/mask forms a diagonal intersection "X" (excluding the center position), a vertical segment "X" (excluding the center position), a horizontal segment "-" (excluding the center position), a diagonal segment "\" from the upper left to the lower right (excluding the center position), a diagonal segment "/" from the upper right to the lower left (excluding the center position), or any combination of the above limiting patterns/shapes/supports/masks.

Figure 16-b provides a 5×5 diamond filter shape with a reduced number of clipping/limiting positions, illustrating two possible arrangements/configurations of clipping/limiting positions for reducing the number of clipping operations required when filtering using the 5×5 filter shape. As in Figure 16-a, by using the clipping patterns shown in Figure 16-b, namely clipping patterns 1604 and 1605, the number of clipping operations involved in the filtering process can be reduced when filtering using the 5×5 diamond filter shape. Clipping patterns 1604 and 1605 work well for nonlinear ALF based on orientation (i.e., for filters constructed based on local content schemes and activity levels, such as the luminance filter in VTM 3.0). However, if classification is not used, pattern 1605 is preferred.

Clamping patterns 1602 and 1605 are better for filters that do not use orientation-based classification (e.g., they are better for chroma filters than luminance filters in VTM-3.0 that depend on horizontal, vertical, or diagonal orientation) because horizontal and vertical proximity is statistically more reliable than diagonal proximity (because of their larger Euclidean distance). Therefore, it is more advantageous to use clamping only at the filtering locations of less reliable diagonal neighbor samples than at filtering locations of samples with more reliable horizontal and vertical proximity.

Using orientation-based filters (i.e., filters obtained from orientation-based classification), the clipping operation is typically applied to samples at filter locations perpendicular to the filter orientation. For example, the clipping pattern in 1601 will perform better than using a filter for classes with horizontal or vertical orientations compared to using a filter for classes with diagonal orientations. For the orientations of all four classes, clipping pattern 1602 will perform roughly the same, but for horizontal and vertical orientations, clipping pattern 1602 will be less efficient than pattern 1601. Therefore, for a luminance filter using a 7×7 filter pattern, the average performance of the clipping patterns in 1601 and 1602 is almost equal, and for a luminance filter using a 5×5 filter pattern, the average performance of the clipping patterns in 1604 and 1605 is almost equal.

According to a variant, to improve the orientation-based filter, a limiting pattern is selected from predefined patterns based on the filtering orientation. For example, limiting pattern 1601 is used for horizontal or vertical filtering, while limiting pattern 1602 is used for diagonal filtering. In yet another variant, the limiting pattern selected for vertical filtering differs from the limiting pattern selected for horizontal filtering (e.g., each is a 90° rotated version of the other), and the limiting patterns for the two diagonal filtering orientations are also different (e.g., each is a 90° rotated version of the other). In yet another variant, the limiting pattern is selected based on the transpose index (tranposeIdx) of each 4×4 block of sample values.

According to one variation, a predetermined clipping pattern exists for each filter. According to another variation, multiple (e.g., a predetermined number) clipping patterns are available for use, and for each filter index, the (clipping pattern) index of the selected clipping pattern is encoded/signaled using ALF parameters, for example, in the APS or block group header. According to yet another variation, the clipping pattern is signaled using ALF parameters. For example, the clipping pattern can be signaled by encoding/decoding a sequence of flags, with one flag for each filter (pattern) position index (e.g., equivalent to the encoded/decoded filter coefficient index), each flag indicating whether clipping is applied to the corresponding index or not. Alternatively, the clipping pattern can be signaled by encoding/decoding the clipped position index itself, or by encoding/decoding the filter (pattern) position index of the un-clipped position (preferably, signaling requires fewer bits).

Depending on the variant, when using a limiting pattern (predetermined or unpredetermined), the limiting parameters are signaled only for the filter (pattern) position where the limiting operation is applied.

In some variants, to avoid interrupting the Single Instruction, Multiple Data (SIMD) parallel implementation, clipping is applied to all filtering positions, but clipping parameters at positions where no clipping parameter is provided are considered to have linear outputs (e.g., the output of the flag function in the variant).

In an alternative embodiment, the number of limiting parameters is less than the number of filter coefficients because one or more limiting parameters are shared. For example, in the filter shapes of Figures 3-b and 4-b, some coefficient indices may share the same limiting parameter. According to a variant, the indices of the limiting parameters are indicated in a table containing multiple elements, the number being equal to half the shape size (due to symmetry) minus 1 (because there is no limiting parameter for the center coefficient position). This table allows the limiting parameter indices to be associated with individual filter coefficients. According to a variant, this table is fixed/preset by the codec. According to a variant, multiple fixed/preset tables are defined, and the indices used to identify the limiting parameters from these tables are signaled, for example, in the title block. In an alternative variant, the contents of such a table are signaled, for example, in the title block, and the contents are shared by all brightness filters.

In this embodiment, the number of possible values for the limiting parameter is limited to a small number (relative to the quality benefits, reducing encoder complexity and encoding costs). The values authorized for the limiting parameter are preferably indexed in ascending or descending order using integer value indexes. These indexes can then be mapped to the individual elements of the limiting parameter table. Instead of signaling the limiting parameter value, the index (p) of the relevant limiting parameter value in the table is signaled. In embodiments where p is the filter index in the table, an intermediate table is not needed to reduce the number of possible values for p. The number of functions available in the function table can be reduced directly, thereby reducing its size.

In one embodiment, the clipping parameter value is limited to a power of two: 2^ ^p . p is then encoded. The maximum value of p is the bit depth of the input image (higher values are not needed as they will provide the same result). In an alternative embodiment, instead of p, B _^d - p is encoded. In another embodiment, the range of p is limited to between _{p_min} and _{p_max} . For example, _{p_min} = 3 and _{p_max} = B _^d - 1. _{p_min} or _{p_max} - p can then be signaled.

In this embodiment, the authorized/permitted/available limiting parameter value of the chroma filter is different from that of the luminance filter.

According to an embodiment, a minimum index p_{_min} and a maximum index p _{_max} from a table of limiting parameters used in the chip are provided in the chip header, allowing the number of possible limiting parameter values to be limited/constrained when signaled. According to an embodiment, p_{_min} and p _{_max} are shared by the luma and chroma filters. In an alternative embodiment, p _{_min} and p _{_max} are provided only for luma, with no limitation on the chroma index. In another alternative embodiment, p _{_min} and p _{_max} are provided in the chip header for both luma and chroma.

According to an alternative embodiment, for both luminance and chrominance components, a table of authorized limiting parameter values is signaled in the chip header, or alternatively, two tables: one for luminance and one for chrominance.

In embodiments of the invention, limiting parameters 511 and 512 are not signaled in the chip header. Instead, quantization parameters (QP) are used at the encoder/decoder to determine the limiting parameters, which are used to signal the filtered samples and are based on the filter class index.

In an alternative embodiment, the limiting parameters 511 and 512 are signaled in the chip header. However, they are not used directly in the filtering process. Instead, the limiting parameters are scaled according to the quantization parameters (QP) used to signal the filtered samples.

In this embodiment, limiting parameters for the luminance filters are not provided for each individual filter. Only two limiting parameter tables are shared by all luminance filters and are signaled in the chip header. One table is for horizontal/vertical classes, and the other is for diagonal classes. Based on the class orientation, the coefficients are arranged using the method of Figure 4-b, and the limiting parameters are scaled (i.e., multiplied) accordingly by a fixed value determined from the class's activity level, deriving the limiting parameters for each class from these two tables. When classifying 4x4 blocks, the scaling takes the activity level into account. For example, for regions with high activity levels (e.g., near the edge), a higher limiting value can be used compared to regions with low activity levels (e.g., in homogeneous regions).

In an embodiment, a bit in the chip header signals whether limiting is enabled or disabled for each filter limiting index of each filter. In an embodiment where only one allowed limiting value is allowed for each limiting parameter, if the bit indicates that limiting is valid, then that limiting value is used (i.e., nothing is signaled for that limiting parameter value). Alternatively, when more than one limiting value is allowed, if a bit in a limiting index indicates that limiting is disabled for that filter limiting index, then the limiting value is not signaled, but otherwise it is signaled.

Here, the filter limiting index corresponds to the index of the limiting parameter associated with the filter coefficient index in the filter shape.

In this embodiment, the allowed clipping value depends on the slice type (e.g., it could be intra, B, or P).

In a variation of the embodiment, the allowed clipping values are: {6, 32, 181, 1024} for luminance in B or P slices; {4, 25, 161, 1024} for chrominance in B or P slices; {10, 102, 1024} for luminance in intra-slice slices; and/or {4, 24, 1024} for luminance in intra-slice slices. Therefore, any clipping parameter can take a value belonging to one of these sets (depending on the slice type and filter component). Furthermore, for each filter, for each clipping parameter, the index of that value in the set is encoded in the slice header.

In the variant, the allowed limit values are defined in the table as follows:

Where N is the number of limit values in the table (i.e., the size of the table), M is the maximum limit value (which is the latest entry in the table, e.g., M = ^2D or M = ^2D -1, where D is the sample bit depth that defines the components of the table), and "round" is the rounding operator (e.g., round to the nearest integer).

In the variant, the allowed limit values are defined in the table as follows:

Where N is the number of limit values in the table (i.e., the size of the table), M is the maximum limit value (which is the latest entry in the table, e.g., M = ^2D or M = ^2D -1, where D is the sample bit depth), A is the minimum limit value (which is the first entry in the table), and "round" is the rounding operator (e.g., round to the nearest integer).

In the embodiments, the allowed limiting values are different for each filter limiting index.

In a variation of the embodiment using a limiting function for the difference, i.e., applying the function K of equation (11) or equation (12) to the difference between neighboring sample values and the center value, a maximum allowable limiting value is defined such that the number of bits at the limiting output is reduced. This allows limiting the number of bits that the hardware implementation of filtering according to these variations needs to process. Thus, this allows, for example, a reduction in the number of logic gates/transistors in the chip. Reducing the number of bits at the input of the multiplication operator is of particular interest because multiplication requires many logic gates. For example, setting the maximum allowable limiting value to 2 raised to the power of 1 and then subtracting 1; (i.e., the maximum limiting equals 2^(bit depth - 1) - 1) allows the maximum number of bits to be reduced by 1 at the limiting output and thus at the input of the multiplication operator. And using the sample bit depth minus 2 (i.e., the maximum limiting equals 2^(bit depth - 2) - 1) allows the maximum number of bits to be reduced by 2.

In embodiments using APS, it may be impossible to distinguish between intra-frame and inter-frame slices (or intra-frame and inter-frame pictures/image portions/block groups) during clipping value derivation because APS is shared by both intra-frame and inter-frame slices/pictures/block groups/image portions. Therefore, the predefined table of allowed clipping values can no longer depend on the slice/image portion/block group type. Thus, all slice/picture/block group/image portion types share a default clipping value table. In a variant, the encoder then determines (and signals this to the decoder) whether the clipping value is limited to a subset of the clipping values from the default clipping value table (e.g., for complexity reduction considerations).

In the variant, the APS signals the available limiting values using the ALF parameter, which serves as the limiting value table. Therefore, the encoder (and decoder) can use any limiting value and signals the used limiting value using its own limiting value table.

In this variant, there is more than one default limiting value table; for example, one for inter-frame frames and another for intra-frame frames. The APS then provides information in the ALF parameter to identify the table used, such as the index of the table. In this variant, the default limiting value tables differ for luminance and chrominance.

In this embodiment, the filter's output value is also limited based on the input sample value and an additional limiting parameter (which can be added to a limiting parameter table).

According to an embodiment, in addition to the ALF validity flag, the NLALF (Nonlinear ALF) validity flag is placed at the sequence level (e.g., in the SPS), or alternatively, at the frame level (e.g., in the image header) to indicate that the nonlinear ALF is valid/used/enabled. If the flag is valid (i.e., indicating that the nonlinear ALF is valid), the ALF syntax elements include filter parameters for the nonlinear filter (e.g., the filter in Figure 5). If the flag is invalid (i.e., not indicating that the nonlinear ALF is valid), the ALF syntax elements only contain filter parameters for the linear filter (e.g., the filter in Figure 2).

According to an embodiment, if the ALF flag is valid in the chip header, then an NLALF valid flag is placed in the chip header. If the NLALF flag is valid, then the chip header contains limiting parameters for each luminance filter syntax element 511 and limiting parameters for the chrominance filter 512. If the NLALF flag is invalid, then syntax elements 511 and 512 are not present in the chip header, and the ALF uses normal linear filtering.

According to an alternative embodiment, separate NLALF validity flags are provided for luminance and chromaticity for each component type, and any one or two of these flags are used to indicate whether a nonlinear ALF is used for a sample of a particular component (e.g., they are provided in the same location as the NLALF validity flags, such as at the sequence level or in the slice header, etc.). In a variant, a nonlinear ALF is used for luminance and a linear ALF is used for chromaticity. In an alternative variant, a nonlinear ALF is used for chromaticity and a linear ALF is used for luminance.

In the variant, an NLALF luminance valid flag and an NLALF chrominance valid flag are provided in the header. The NLALF luminance valid flag indicates whether a nonlinear adaptive loop filter is used for the luminance component, and therefore indicates whether the limiting parameters of the nonlinear ALF are signaled for each luminance filter. The NLALF chrominance valid flag indicates whether a nonlinear adaptive loop filter is used for the chrominance component, and therefore indicates whether the limiting parameters of the nonlinear ALF are signaled for the chrominance filter.

In the variant, no signal is used to notify the NLALF chromaticity validity flag, but if the NLALF luminance validity flag is equal to zero, it is inferred to be equal to zero. Therefore, nonlinear ALF cannot be used only for chromaticity.

In one variant, non-linear ALF is always used for luminance (default - therefore the NLALF luminance validity flag can be omitted or assumed to be the corresponding default value), and the NLALF chrominance validity flag is used to indicate whether non-linear or linear ALF is used for chrominance. In another variant, linear ALF is always used for luminance (default - therefore the NLALF luminance validity flag can be omitted or assumed to be the corresponding default value), and the NLALF chrominance validity flag is used to indicate whether non-linear or linear ALF is used for chrominance. In another variant, the NLALF luminance validity flag is used to indicate whether non-linear or linear ALF is used for luminance, and non-linear ALF is always used for chrominance (default - therefore the NLALF luminance validity flag can be omitted or assumed to be the corresponding default value). In another variant, the NLALF luminance validity flag is used to indicate whether non-linear or linear ALF is used for luminance, and linear ALF is always used for chrominance (default - therefore the NLALF luminance validity flag can be omitted or assumed to be the corresponding default value).

In one embodiment, the NLALF valid flag (or NLALF luma valid flag or NLALF chroma valid flag) is placed in the chip header only if the NLALF flag is valid in SPS. If the NLALF valid flag (or NLALF luma valid flag or NLALF chroma valid flag) is not present, it is considered invalid by default.

In one embodiment, the NLALF flag replaces the ALF flag in the SPS and the title block. The classic (linear only) ALF can no longer be used, except in embodiments where the clipping function allows linear output for allowed clipping parameter values.

According to the embodiment, the limiting parameters are signaled in a manner similar to that used for filter coefficients. First, the (exp-)Golomb coding parameters of the limiting parameters are signaled: the minimum exponent of the (exp-)Golomb code is signaled using VLC code, and then for each (exp-)Golomb index (e.g., 3 indices for the luminance filter and 2 indices for the chrominance filter), a flag is used to signal whether the (exp-)Golomb exponent must be increased (starting from the minimum exponent) for the current and next indices.

Then, using (for unsigned integers) (exp-)Golomb codes, the limiting parameters are signaled for each (non-disabled) filter, where the (exp-)Golomb exponent with (exp-)Golomb index is associated with the coefficient index from the table storing (exp-)Golomb parameters (in a fixed table, for example, the same as for the filter coefficients).

According to an alternative embodiment, for each coding filter, for example, the (exp-)Golomb coding parameter value is signaled in the chip header, and each limiting parameter value is encoded using the provided (exp-)Golomb coding parameter value for unsigned integer (exp-)Golomb coding.

According to an embodiment, the (exp-)Golomb encoded parameter value is used as an index in a table of available (exp-)Golomb encoded parameter values to notify via signaling.

According to an embodiment, for example, a signal notification flag is used in the chip header to indicate whether incremental mode is used to signal the limiting parameters for the luminance filter. When incremental mode is active, the first filter limiting parameter is encoded as described above, but each subsequent filter limiting parameter is encoded as the difference between the subsequent filter limiting parameter and the previously encoded filter limiting parameter. In this case, the difference is encoded using signed integer (exp-) Golomb coding.

According to an embodiment, for example, a signal notification flag is used in the chip header to indicate whether the limiting parameters of the filter can use incremental mode. According to an embodiment, when incremental mode is active, a signal notification flag is used for each filter to indicate whether the limiting parameters of the filter use incremental mode. According to an embodiment, when incremental mode is active to encode the limiting parameters of a given filter, the limiting parameters are encoded one by one, with the first limiting parameter using unsigned (exp-) Golomb encoding, and subsequent limiting parameters using signed (exp-) Golomb encoding of the difference between the limiting parameter and the value of the previously encoded limiting parameter.

According to an alternative embodiment, the number of values that the limiting parameter can take is limited to two distinct values. A single bit is then used to signal which limiting parameter value must be used (the (exp-)Golomb parameter does not need to be provided).

According to an embodiment, conditional signaling of the clipping parameter is performed to minimize signaling of unnecessary data (e.g., when the filter coefficients are zero or equal to a known value of the filter on the decoder side). In a variant, such as in the chip header, the filter coefficients are signaled before the clipping parameter is signaled, and for each i-th filter coefficient that is equal to zero, its corresponding i-th clipping parameter is not signaled. This is because the i-th clipping is not useful to the filter, as the result will be multiplied by a zero coefficient, making the i-th clipping have no effect on the filtered output. In one variant, the clipping is set to a default value (e.g., based on the maximum possible value of the sample bit depth). In another variant, the filtering process does not apply clipping to input samples that are then expected to be multiplied by a zero coefficient.

According to an embodiment, a signal is used to notify a flag for each luminance filter to indicate whether a nonlinear filter is used with that luminance filter. In a variant, the limiting parameter/value of the luminance filter is only signaled when a nonlinear filter is used with that luminance filter (i.e., if the flag is valid).

According to an embodiment, a signal is used to notify a flag for the chroma filter to indicate whether a nonlinear filter is used in conjunction with the chroma filter. In a variant, the limiting parameter/value of the chroma filter is signaled only when a nonlinear filter is used for the chroma filter (i.e., if the flag is valid).

According to embodiments, one or more of the following are used to signal ALF-related information, such as ALF parameters and flags: APS syntax (Table 1); block group header syntax (Table 2); coding tree unit syntax (Table 3); and/or the non-linear ALF data syntax shown below (Table 4), which uses the same syntax naming convention as in VVC draft version 3. According to variants, all four syntaxes (e.g., Tables 1-4) are used to signal ALF-related information. According to alternative variants, a subset of the four syntaxes is used to signal ALF-related information.

Tables 1 and 2 provide advanced syntax elements to provide/notify ALF parameters with signals, such as alf_data().

Table 1 - Syntax of Adaptive Parameter Set (APS)

Table 2 - Block Group Header Syntax

The slice (header) described in other embodiments (or variations thereof) is replaced by a tile group (header) (as shown by the tile group header syntax in Table 2). Therefore, the “tile_group_alf_enabled_flag” corresponds to syntax elements 202 in Figure 2 and 502 in Figure 5, and it indicates whether ALF is valid for a tile group. Furthermore, ALF data syntax elements are not provided in the tile group header (in other embodiments with slices, they may be provided in the slice header). Instead, ALF data syntax elements are provided in a specific set of parameters (as shown in Table 1) called the Adaptive Parameter Set (APS). ALF data includes ALF syntax elements, and providing ALF data syntax elements in the APS enables the sharing of ALF parameters across more than one tile group, for example, sharing ALF parameters across multiple tile groups in the same and/or different decoded images.

The Adaptive Parameter Set (APS) syntax in Table 1 is used to define any adaptive parameter set that is signaled. In variants, the APS is included in a non-Video Coding Layer (VCL) Network Abstraction Layer (NAL) unit (e.g., called the "APS_NUT" or APSNAL unit type).

The semantics of each syntax element in APS are as follows:

Adaptive parameter set semantics

The adaptation_parameter_set_id identifies the APS that is referenced by other syntax elements. The value of adaptation_parameter_set_id should be in the range of 0 to 63 (inclusive).

The block group header syntax in Table 2 is used to define the headers for each block group. A block group header is provided for each block group. A block group consists of a set of blocks, and each block consists of a set of CTUs.

The semantics of the syntax elements in the block group header are as follows:

Block header semantics

`tile_group_aps_id` specifies the `adaptation_parameter_set_id` of the APS referenced by the tile group. The value of `tile_group_aps_id` should be in the range of 0 to 63 (inclusive). The `TemporalId` of the APS NAL cell whose `adaptation_parameter_set_id` is equal to `tile_group_aps_id` should be less than or equal to the `TemporalId` of the encoded tile group NAL cell.

Table 3 provides the syntax elements for the ALF (Enable) flags used to signal ALF validity at the CTU level for each component. These correspond to the syntax elements used in Figures 208, 209, and 210, and Figures 508, 509, and 510.

Table 3 - Syntax of Encoding Tree Unit

The coding tree unit syntax in Table 3 is used to define (encoded) coding tree units. The semantics of each syntax element of a coding tree unit are as follows:

Encoding Tree Unit Semantics

CTU is the root node that encodes a quadtree structure.

`alf_ctb_flag[cIdx][xCtb>>Log2CtbSize][yCtb>>Log2CtbSize]` equal to 1 specifies that the adaptive loop filter is applied to the coding tree block for the color component indicated by the cIdx of the coding tree unit at the luma position (xCtb, yCtb). `alf_ctb_flag[cIdx][xCtb>>Log2CtbSize][yCtb>>Log2CtbSize]` equal to 0 specifies that the adaptive loop filter is not applied to the coding tree block for the color component indicated by the cIdx of the coding tree unit at the luma position (xCtb, yCtb).

If alf_ctb_flag[cIdx][xCtb>>Log2CtbSize][yCtb>>Log2CtbSize] does not exist, it is inferred to be equal to 0.

Table 4 provides the syntax elements (i.e., the nonlinear ALF data syntax) for signaling nonlinear ALF parameters, which are built upon the ALF data syntax elements of VVC draft version 3. These syntax elements are built upon the syntax elements used in the variants described herein with reference to the nonlinear ALF syntax elements of Figure 5 (which are built upon the syntax elements of Figure 2).

Table 4 - Syntax of Nonlinear ALF Data

The nonlinear adaptive loop filter (ALF) data syntax in Table 4 is used to define the adaptive parameter set (shown in Table 1). The semantics of the nonlinear ALF data syntax elements are as follows:

Adaptive Loop Filter Data Semantics

`alf_chroma_idc` equal to 0 indicates that the adaptive loop filter is not applied to the Cb and Cr color components. `alf_chroma_idc` equal to 1 indicates that the adaptive loop filter is applied to the Cb color component. `alf_chroma_idc` equal to 2 indicates that the adaptive loop filter is applied to the Cr color component. `alf_chroma_idc` equal to 3 indicates that the adaptive loop filter is applied to both the Cb and Cr color components.

The maximum value of the truncated unary binarized tu(v), maxVal, is set to equal 3.

The variable NumAlfFilters, which specifies the number of different adaptive loop filters, is set to 25.

The increment of alf_luma_num_filters_signalled_minus1 by 1 specifies the number of adaptive loop filter classes that can be used to signal the luminance coefficients. The value of alf_luma_num_filters_signalled_minus1 should be in the range of 0 to NumAlfFilters-1 (inclusive).

The maximum value of the truncated binary binarized tb(v), maxVal, is set to be equal to NumAlfFilters-1.

`alf_luma_coeff_delta_idx[filtIdx]` specifies the index of the adaptive loop filter luminance coefficient increment of the filter class indicated by `filtIdx`, which is signaled by a signal. `filtIdx` is in the range of 0 to NumAlfFilters-1. If `alf_luma_coeff_delta_idx[filtIdx]` does not exist, it is inferred to be equal to 0.

The maximum value maxVal of the truncated binary binarized tb(v) is set to equal alf_luma_num_filters_signalled_minus1.

`alf_luma_coeff_delta_flag` equal to 1 indicates that `alf_luma_coeff_delta_prediction_flag` should not be signaled. `alf_luma_coeff_delta_flag` equal to 0 indicates that `alf_luma_coeff_delta_prediction_flag` can be signaled.

`alf_luma_coeff_delta_prediction_flag` equal to 1 specifies the luminance filter coefficient increments predicted by the signal from the incremental prediction of previous luminance coefficients. `alf_luma_coeff_delta_prediction_flag` equal to 0 specifies the luminance filter coefficient increments not predicted by the signal from the incremental prediction of previous luminance coefficients. If it does not exist, it is assumed that `alf_luma_coeff_delta_prediction_flag` equals 0.

Increasing 1 to alf_luma_min_eg_order_minus1 specifies the minimum exponent of the exp-Golomb code used to signal the luminance filter coefficients. The value of alf_luma_min_eg_order_minus1 should be in the range of 0 to 6 (inclusive).

`alf_luma_eg_order_increase_flag[i]` equal to 1 indicates that the minimum exponent of the exp-Golomb code for the luminance filter coefficients is incremented by 1. `alf_luma_eg_order_increase_flag[i]` equal to 0 indicates that the minimum exponent of the exp-Golomb code for the luminance filter coefficients is not incremented by 1.

The exponent expGoOrderY[i] used to decode the exp-Golomb code of the value of alf_luma_coeff_delta_abs[sigFiltIdx][j] is derived as follows:

expGoOrderY[i]=alf_luma_min_eg_order_minus1+1+alf_luma_eg_order_increase_flag[i]

`alf_luma_coeff_flag[sigFiltIdx]` equal to 1 specifies that the coefficients of the luminance filter indicated by `sigFiltIdx` are signaled. `alf_luma_coeff_flag[sigFiltIdx]` equal to 0 specifies that all filter coefficients of the luminance filter indicated by `sigFiltIdx` are set to 0. When it does not exist, `alf_luma_coeff_flag[sigFiltIdx]` is set to 1.

alf_luma_coeff_delta_abs[sigFiltIdx][j] specifies the absolute value of the j-th coefficient increment of the luminance filter, indicated by sigFiltIdx. If alf_luma_coeff_delta_abs[sigFiltIdx][j] does not exist, it is inferred to be equal to 0.

The exponent k of the exp-Golomb binarized uek(v) is derived as follows:

golombOrderIdxY[]＝{0,0,1,0,0,1,2,1,0,0,1,2}

k＝expGoOrderY[golombOrderIdxY[j]]

alf_luma_coeff_delta_sign[sigFiltIdx][j] specifies the sign of the j-th luminance coefficient of the filter indicated by sigFiltIdx as follows:

If alf_luma_coeff_delta_sign[sigFiltIdx][j] equals 0, then the corresponding luminance filter coefficient has a positive value.

Otherwise (alf_luma_coeff_delta_sign[sigFiltIdx][j] equals 1), the corresponding luminance filter coefficients have negative values.

When alf_luma_coeff_delta_sign[sigFiltIdx][j] does not exist, it is inferred to be equal to 0.

The variable filterCoefficients[sigFiltIdx][j] (where igFiltIdx = 0…alf_luma_num_filters_signalled_minus1, j = 0…11) is initialized as follows:

filterCoefficients[sigFiltIdx][j]=alf_luma_coeff_delta_abs[sigFiltIdx][j]*

(1-2*alf_luma_coeff_delta_sign[sigFiltIdx][j])

When alf_luma_coeff_delta_prediction_flag equals 1, filterCoefficients[sigFiltIdx][j] (where igFiltIdx = 0…alf_luma_num_filters_signalled_minus1, j = 0…11) is modified as follows:

filterCoefficients[sigFiltIdx][j]+＝filterCoefficients[sigFiltIdx-1][j]

The derivation of the luminance filter coefficients with elements AlfCoeffL[filtIdx][j] (where igFiltIdx = 0…alf_luma_num_filters_signalled_minus1, j = 0…11) is as follows

AlfCoeffL[filtIdx][j]=filterCoefficients[alf_luma_coeff_delta_idx[filtIdx]][j]

The derivation of the final filter coefficients AlfCoeffL[filtIdx][12] (filtIdx＝0...NumAlfFilters-1) is as follows:

AlfCoeffL[filtIdx][12] = 128

One requirement for bitstream consistency is that the value of AlfCoeffL[filtIdx][j] (where filtIdx = 0...NumAlfFilters-1, j = 0…11) should be in the range of -2^7 to 2^7-1 (inclusive), and the value of AlfCoeffL[filtIdx][12] should be in the range of 0 to 2^8-1 (inclusive).

The increment of alf_chroma_min_eg_order_minus1 by 1 specifies the minimum exponent of the exp-Golomb code for the chroma filter coefficients notified by the signal. The value of alf_chroma_min_eg_order_minus1 should be in the range of 0 to 6 (inclusive).

`alf_chroma_eg_order_increase_flag[i]` equal to 1 specifies that the minimum exponent of the exp-Golomb code for the chroma filter coefficients notified by the signal is incremented by 1. `alf_chroma_eg_order_increase_flag[i]` equal to 0 specifies that the minimum exponent of the exp-Golomb code for the chroma filter coefficients notified by the signal is not incremented by 1.

The exponent expGoOrderC[i] used to decode the exp-Golomb code of the value of alf_chroma_coeff_abs[j] is derived as follows:

expGoOrderC[i]=alf_chroma_min_eg_order_minus1+1+alf_chroma_eg_order_increase_flag[i]

alf_chroma_coeff_abs[j] specifies the absolute value of the j-th chroma filter coefficient. If alf_chroma_coeff_abs[j] does not exist, it is inferred to be equal to 0. Bitstream consistency requires that the value of alf_chroma_coeff_abs[j] be in the range of 0 to 2^7-1 (inclusive).

The exponent k of the exp-Golomb binarized uek(v) is derived as follows:

golombOrderIdxC[]＝{0,0,1,0,0,1}

k＝expGoOrderC[golombOrderIdxC[j]]

alf_chroma_coeff_sign[j] specifies the sign of the j-th chroma filter coefficient as follows:

If alf_chroma_coeff_sign[j] equals 0, then the corresponding chroma filter coefficient has a positive value. Otherwise (alf_chroma_coeff_sign[j] equals 1), the corresponding chroma filter coefficient has a negative value.

If alf_chroma_coeff_sign[j] does not exist, the inference is that it equals 0.

The chromaticity filter coefficients AlfCoeffC with element cC[j] (where j = 0…5) are derived as follows:

AlfCoeffC[j]=alf_chroma_coeff_abs[j]*(1-2*alf_chroma_coeff_sign[j])

The final filter coefficients for j=6 are derived as follows:

AlfCoeffC[6]＝128

The requirement for bitstream consistency is that AlfCoeffC[j] (where j = 0…5) should be in the range of -2^7-1 to 2^7-1 (inclusive), and the value of AlfCoeffC[6] should be in the range of 0 to 2^8-1 (inclusive).

A value of 0 for `alf_luma_clip` specifies that a linear adaptive loop filter is applied to the luminance component. A value of 1 for `alf_luma_clip` specifies that a nonlinear adaptive loop filter can be applied to the luminance component.

An alf_chroma_clip of 0 specifies that a linear adaptive loop filter is applied to the chroma components; an alf_chroma_clip of 1 indicates that a nonlinear adaptive loop filter is applied to the chroma components. If alf_chroma_clip does not exist, it is inferred to be 0.

A value of 1 for alf_luma_clip_default_table specifies that the default alf_luma_clipping_value table is used to convert the clipping index to the clipping value. A value of 0 for alf_luma_clip_default_table indicates that the alf_luma_clipping_value[] table exists in alf_data().

Incrementing 1 by 1 in alf_luma_num_clipping_values_minus1 indicates the size of the alf_luma_clipping_value[] table. If it does not exist, it is inferred to be equal to bitDepthY.

`alf_luma_clipping_value[clipIdx]` specifies the clipping value used when signaling the clipping index `clipIdx` in the `alf_luma_clip_idx[][]` table. If it does not exist, it is inferred that `alf_luma_clipping_value[] = {1 << bitDepthY, 1 << (bitDepthY-1), ..., 8, 4, 2}`.

For convenience, we infer that alf_luma_clipping_value[alf_luma_num_clipping_values_minus1+1] equals 1<<(bitDepthY-1).

When alf_chroma_clip_default_table equals 1, it specifies that the default alf_chroma_clipping_value[] table is used to convert the clipping index to the clipping value; when alf_chroma_clip_default_table equals 0, it indicates that the alf_chroma_clipping_value[] table exists in alf_data().

Incrementing 1 by 1 in alf_chroma_num_clipping_values_minus1 indicates the size of the alf_chroma_clipping_value[] table. If it does not exist, it is inferred to be equal to bitDepthC.

`alf_chroma_clipping_value[clipIdx]` specifies the clipping value used when signaling the clipping index `clipIdx` in the `alf_chroma_clip_idx[][]` table. If it does not exist, it is inferred that `alf_chroma_clipping_value[] = {1 << bitDepthC, 1 << (bitDepthC-1), ..., 8, 4, 2}`.

For convenience, we infer that alf_chroma_clipping_value e[alf_chroma_num_clipping_values_minus1+1] equals 1<<bitDepthC.

A value of 0 for `alf_luma_filter_clip[sigFiltIdx]` specifies that a linear adaptive loop filter is applied together with the luminance filter indicated by `sigFiltIdx`. A value of 1 for `alf_luma_filter_clip[sigFiltIdx]` specifies that a nonlinear adaptive loop filter is applied together with the luminance filter indicated by `sigFiltIdx`.

The increment of 1 in `alf_luma_clip_min_eg_order_minus1` specifies the minimum exponent of the exp-Golomb code for the luminance filter coefficients notified by the signal. The value of `alf_luma_clip_min_eg_order_minus1` should be in the range of 0 to 6 (inclusive).

`alf_luma_clip_eg_order_increase_flag[i]` equal to 1 indicates that the minimum exponent of the exp-Golomb code for the luminance filter coefficients notified by the signal is incremented by 1. `alf_luma_clip_eg_order_increase_flag[i]` equal to 0 indicates that the minimum exponent of the exp-Golomb code for the luminance filter coefficients notified by the signal is not incremented by 1.

The exponent of the exp-Golomb code used to decode the value of alf_luma_clip_idx[sigFiltIdx][j], expGoOrderYClip[i], is derived as follows:

expGoOrderYClip[i]=alf_luma_clip_min_eg_order_minus1+1+alf_luma_clip_eg_order_increase_flag[i]

alf_luma_clip_idx[sigFiltIdx][j] specifies the clipping index to be used before multiplying by the j-th coefficient of the luminance filter indicated by sigFiltIdx. When alf_luma_clip_idx[sigFiltIdx][j] does not exist, it is inferred to be equal to alf_luma_num_clipping_values_minus1+1 (infinite clipping).

The exponent k of the exp-Golomb binarized uek(v) is derived as follows:

golombOrderIdxYClip[]＝{0,0,1,0,0,1,2,1,0,0,1,2}

k＝expGoOrderYClip[golombOrderIdxYClip[j]]

The variable filterClips[sigFiltIdx][j] (where sigFiltIdx = 0...alf_luma_num_filters_signalled_minus1,j = 0…11) is initialized as follows:

filterClips[sigFiltIdx][j]=alf_luma_clipping_value[

alf_luma_clip_idx[sigFiltIdx][j]]

The luminance filter limiting value AlfClipL with element AlfClipL[filtIdx][j] (where filtIdx = 0..NumAlfFilters-1 and j = 0…11) is derived as follows:

AlfClipL[filtIdx][j]=filterClips[alf_luma_coeff_delta_idx[filtIdx]][j]

The increment of 1 in `alf_chroma_clip_min_eg_order_minus1` specifies the minimum exponent of the exp-Golomb code for the chroma filter coefficients notified by the signal. The value of `alf_chroma_clip_min_eg_order_minus1` should be in the range of 0 to 6 (inclusive).

`alf_chroma_clip_eg_order_increase_flag[i]` equal to 1 indicates that the minimum exponent of the exp-Golomb code for the chroma filter coefficients notified by the signal is incremented by 1. `alf_chroma_clip_eg_order_increase_flag[i]` equal to 0 indicates that the minimum exponent of the exp-Golomb code for the chroma filter coefficients notified by the signal is not incremented by 1.

The exponent expGoOrderC[i] used to decode the exp-Golomb code of the value of alf_chroma_coeff_abs[j] is derived as follows:

expGoOrderC[i]=alf_chroma_clip_min_eg_order_minus1+1+alf_chroma_clip_eg_order_increase_flag[i]

alf_chroma_clip_idx[j] specifies the clipping index to be used before multiplying by the j-th coefficient of the chroma filter. When alf_chroma_clip_idx[j] does not exist, it is inferred to be equal to alf_chroma_num_clipping_values_minus1+1 (infinite clipping).

The exponent k of the exp-Golomb binarized uek(v) is derived as follows:

golombOrderIdxC[]＝{0,0,1,0,0,1}

k＝expGoOrderC[golombOrderIdxC[j]]

The luminance filter clipping value AlfClipC, with element AlfClipC[j] (where j = 0…5), is derived as follows: AlfClipC[j] = alf_chroma_clipping_value[alf_chroma_clip_idx[j]].

According to the variant, alf_luma_num_clipping_values_minus1 is replaced by alf_luma_num_clipping_values_minus2, and provides a limiting value from index = 1 to alf_luma_num_clipping_values_minus2+1, and the first value of alf_luma_clipping_value[] is inferred as alf_luma_clipping_value[0] = 1 << bitDepthY. alf_chroma_num_clipping_values_minus1 is replaced by alf_chroma_num_clipping_values_minus2, and provides a limiting value from index 1 to alf_chroma_num_clipping_values_minus2+1, and the first value of alf_chroma_clipping_value[] is inferred as alf_chroma_clipping_value[0] = 1 << bitDepthC

Depending on the variant, adaptive loop filtering uses the syntax elements described above. This variant is described using the notation conventions of the VVC draft specification below.

Adaptive loop filtering

Overview

The input to this process is the array of reconstructed image samples recPictureL, recPictureCb, and recPictureCr preceding the adaptive loop filter.

The output of this process is a modified array of reconstructed image samples, alfPictureL, alfPictureCb, and alfPictureCr, after adaptive loop filtering.

The sample values in the modified reconstructed image sample arrays alfPictureL, alfPictureCb, and alfPictureCr after adaptive loop filtering are initially set to be equal to the sample values in the reconstructed image sample arrays recPictureL, recPictureCb, and recPictureCr before adaptive loop filtering, respectively.

When tile_group_alf_enabled_flag equals 1, for each coding tree unit with a luma coding tree block position (rx, ry) (where rx = 0…PicWidthInCtbs-1 and ry = 0…PicHeightInCtbs-1), the following applies:

When alf_ctb_flag[0][rx][ry] equals 1, the luminance sample coded tree block filtering process specified in the clause "Ccoded Tree Block Filtering Process for Luminance Samples" is invoked with recPictureL, alfPictureL, and the luminance coded tree block position (xCtb, yCtb) set to (rx<<CtbLog2SizeY, ry<<CtbLog2SizeY) as input, and the output is the modified filtered image alfPictureL.

When alf_ctb_flag[1][rx][ry] equals 1, the chroma sample coding block filtering process specified in the clause "chroma sample coding block filtering process" is called with recPicture set to equal recPictureCb, alfPicture set to equal alfPictureCb, and the chroma coding block position (xCtbC, yCtbC) set to equal (rx<<(CtbLog2SizeY-1), ry<<(CtbLog2SizeY-1)) as input, and the output is the modified filtered image alfPictureCb.

When alf_ctb_flag[2][rx][ry] equals 1, the chroma sample coding block filtering process specified in the clause "chroma sample coding block filtering process" is called with recPicture set to equal recPictureCr, alfPicture set to equal alfPictureCr, and the chroma coding block position (xCtbC, yCtbC) set to equal (rx<<(CtbLog2SizeY-1), ry<<(CtbLog2SizeY-1)) as input, and the output is the modified filtered image alfPictureCr.

Encoding tree block filtering of luminance samples

The inputs to this process are: the reconstructed luminance image sample array recPictureL before adaptive loop filtering, the reconstructed luminance image sample array alfPictureL after filtering, and the luminance position (xCtb, yCtb) of the top-left sample of the current luminance coding tree block for the top-left sample of the current image.

The output of this process is a modified filtered reconstructed luminance image sample array alfPictureL.

The derivation process of the filter index clause “ALF transpose of luminance samples and derivation of filter index” is invoked with the position (xCtb, yCtb) and the reconstructed luminance image sample array recPictureL as input and with the outputs filtIdx[x][y] and transposeIdx[x][y] (where x, y = 0…CtbSizeY-1) as outputs.

For the derivation of the filtered reconstructed luminance sample alfPictureL[x][y], each reconstructed luminance sample recPictureL[x][y] (where x, y = 0…CtbSizeY-1) within the current luminance coding tree block is filtered as follows:

The luminance filter coefficient array f[j] and luminance filter limiting array c[j] (where j = 0…12) corresponding to the filter specified by filtIdx[x][y] are derived as follows:

f[j]＝AlfCoeffL[filtIdx[x][y]][j]

c[j]＝AlfClipL[filtIdx[x][y]][j]

The brightness filter coefficients filterCoeff and filter clipping value filterClip are derived from transposeIdx[x][y] as follows:

If transposeIndex[x][y] == 1,

filterCoeff[]＝{f[9],f[4],f[10],f[8],f[1],f[5],f[11],f[7],f[3],f[0],f[2],f[6],f[12]}

filterClip[]＝{c[9],c[4],c[10],c[8],c[1],c[5],c[11],c[7],c[3],c[0],c[2],c[6],c[12]}

Otherwise, if transposeIndex[x][y] == 2,

filterCoeff[]={f[0],f[3],f[2],f[1],f[8],f[7],f[6],f[5],f[4],f[9],f[10],f[11],f[12]}

filterClip[]＝{c[0],c[3],c[2],c[1],c[8],c[7],c[6],c[5],c[4],c[9],c[10],c[11],c[12]}

Otherwise, if transposeIndex[x][y] == 3,

filterCoeff[]＝{f[9],f[8],f[10],f[4],f[3],f[7],f[11],f[5],f[1],f[0],f[2],f[6],f[12]}

filterClip[]＝{c[9],c[8],c[10],c[4],c[3],c[7],c[11],c[5],c[1],c[0],c[2],c[6],c[12]}

otherwise

filterCoeff[]={f[0],f[1],f[2],f[3],f[4],f[5],f[6],f[7],f[8],f[9],f[10],f[11],f[12]}

filterClip[]＝{c[0],c[1],c[2],c[3],c[4],c[5],c[6],c[7],c[8],c[9],c[10],c[11],c[12]}

The positions (hx, vy) of each corresponding luminance sample (x, y) within the luminance sample array recPicture are derived as follows:

hx＝Clip3(0,pic_width_in_luma_samples-1,xCtb+x)

vy＝Clip3(0,pic_height_in_luma_samples-1,yCtb+y)

The variable curr is derived as follows: curr = recPictureL[hx, vy]

The sum of the variables is derived as follows:

sum=filterCoeff[0]*(Clip3(-filterClip[0],filterClip[0],recPictureL[hx,vy+3]-curr)+Clip3(-filterClip[0],filterClip[0],recPictureL[hx,vy-3]-curr))

+filterCoeff[1]*(Clip3(-filterClip[1],filterClip[1],recPictureL[hx+1,vy+2]-curr)+Clip3(-filterClip[1],filterClip[1],recPictureL[hx-1,vy-2]-curr))

+filterCoeff[2]*(Clip3(-filterClip[2],filterClip[2],recPictureL[hx,vy+2]-curr)+Clip3(-filterClip[2],filterClip[2],recPictureL[hx,vy-2]-curr))

+filterCoeff[3]*(Clip3(-filterClip[3],filterClip[3],recPictureL[hx-1,vy+2]-curr)+Clip3(-filterClip[3],filterClip[3],recPictureL[hx+1,vy-2]-curr))

+filterCoeff[4]*(Clip3(-filterClip[4],filterClip[4],recPictureL[hx+2,vy+1]-curr)+Clip3(-filterClip[4],filterClip[4],recPictureL[hx-2,vy-1]-curr))

+filterCoeff[5]*(Clip3(-filterClip[5],filterClip[5],recPictureL[hx+1,vy+1]-curr)+Clip3(-filterClip[5],filterClip[5],recPictureL[hx-1,vy-1]-curr))

+ filterCoeff[6]*(Clip3(-filterClip[6], filterClip[6],recPictureL[hx,vy+1]-curr)+ Clip3(-filterClip[6], filterClip[6],recPictureL[hx,vy-1]-curr))

+filterCoeff[7]*(Clip3(-filterClip[7],filterClip[7],recPictureL[hx-1,vy+1]-curr)+Clip3(-filterClip[7],filterClip[7],recPictureL[hx+1,vy-1]-curr))

+filterCoeff[8]*(Clip3(-filterClip[8],filterClip[8],recPictureL[hx-2,vy+1]-curr)+Clip3(-filterClip[8],filterClip[8],recPictureL[hx+2,vy-1]-curr))

+ filterCoeff[9]*(Clip3(-filterClip[9], filterClip[9],recPictureL[hx+3,vy]-curr)+ Clip3(-filterClip[9], filterClip[9],recPictureL[hx-3,vy]-curr))

+ filterCoeff[10]*(Clip3(-filterClip[10], filterClip[10],recPictureL[hx+2,vy]-curr)+ Clip3(-filterClip[10], filterClip[10],recPictureL[hx-2,vy]-curr))

+ filterCoeff[11]*(Clip3(-filterClip[11], filterClip[11],recPictureL[hx+1,vy]-curr)+ Clip3(-filterClip[11], filterClip[11],recPictureL[hx-1,vy]-curr))

+filterCoeff[12]*recPictureL[hx,vy]

sum＝(sum+64)>>7

The modified filtered and reconstructed brightness image sample alfPictureL[xCtb+x][yCtb+y] is exported as follows:

alfPictureL[xCtb+x][yCtb+y]＝Clip3(0,(1<<BitDepthY)-1,sum)

Luminance sample ALF transpose and filter index derivation processing

The input to this process is: the brightness position (xCtb, yCtb) of the top-left sample of the current brightness coding tree block for the top-left sample of the current image, and the reconstructed brightness image sample array recPictureL before adaptive loop filtering.

The output of this process is a classification filter index array filtIdx[x][y] (where x, y = 0...CtbSizeY-1).

Transpose index array transposeIdx[x][y] (where x, y = 0...CtbSizeY-1).

The positions (hx, vy) of each corresponding luminance sample (x, y) within the luminance sample array recPicture are derived as follows:

hx＝Clip3(0,pic_width_in_luma_samples-1,x)

vy＝Clip3(0,pic_height_in_luma_samples-1,y)

The classification filter index array filtIdx and the transpose index array transposeIdx are derived through the following ordered steps:

1) The variables filtH[x][y], filtV[x][y], filtD0[x][y], and filtD1[x][y] (where x, y = -2…CtbSizeY+1) are derived as follows:

If both x and y are even or both x and y are odd, then the following applies:

filtH[x][y]＝Abs((recPicture[hxCtb+x,vyCtb+y]<<1)-recPicture[hxCtb+x-1,vyCtb+y]-recPicture[hxCtb+x+1,vyCtb+y])

filtV[x][y]=Abs((recPicture[hxCtb+x,vyCtb+y]<<1)-recPicture[hxCtb+x,vyCtb+y-1]-recPicture[hxCtb+x,vyCtb+y+1])

filtD0[x][y]＝Abs((recPicture[hxCtb+x,vyCtb+y]<<1)-recPictur e[hxCtb+x-1,vyCtb+y-1]-recPicture[hxCtb+x+1,vyCtb+y+1])

filtD1[x][y]＝Abs((recPicture[hxCtb+x,vyCtb+y]<<1)-recPicture[hx Ctb+x+1,vyCtb+y-1]-recPicture[hxCtb+x-1,vyCtb+y+1])

Otherwise, filtH[x][y], filtV[x][y], filtD0[x][y], and filtD1[x][y] are set to 0.

2) The variables varTempH1[x][y], varTempV1[x][y], varTempD01[x][y], varTempD11[x][y], and varTemp[x][y] (where x, y = 0...(CtbSizeY-1)>>2) are exported as follows:

sumH[x][y]＝ΣiΣj filtH[(x<<2)+i][(y<<2)+j](where i,j＝-2...5)

sumV[x][y]＝ΣiΣj filtV[(x<<2)+i][(y<<2)+j](where i,j＝-2…5)

sumD0[x][y]＝ΣiΣj filtD0[(x<<2)+i][(y<<2)+j](where i,j＝-2...5)

sumD1[x][y]＝ΣiΣj filtD1[(x<<2)+i][(y<<2)+j](where i,j＝-2...5)

sumOfHV[x][y]=sumH[x][y]+sumV[x][y]

3) The variables dir1[x][y], dir2[x][y], and dirS[x][y] (where x, y = 0...CtbSizeY-1) are derived as follows:

The variables hv1, hv0, and dirHV are exported as follows:

If sumV[x>>2][y>>2] is greater than sumH[x>>2][y>>2], then the following applies:

hv1 = sumV[x>>2][y>>2]

hv0 = sumH[x>>2][y>>2]

dirHV=1

Otherwise, the following applies:

hv1 = sumH[x>>2][y>>2]

hv0 = sumV[x>>2][y>>2]

dirHV=3

The variables d1, d0, and dirD are derived as follows:

If sumD0[x>>2][y>>2] is greater than [x>>2][y>>2], then the following applies:

d1 = sumD0[x>>2][y>>2]

d0 = sumD1[x>>2][y>>2]

dirD=0

Otherwise, the following applies:

d1 = sumD1[x>>2][y>>2]

d0 = sumD0[x>>2][y>>2]

dirD=2

The variables hvd1 and hvd0 are exported as follows:

hvd1＝(d1*hv0>hv1*d0)? d1:hv1

hvd0=(d1*hv0>hv1*d0)? d0:hv0

The variables dirS[x][y], dir1[x][y], and dir2[x][y] are exported as follows:

dir1[x][y]=(d1*hv0>hv1*d0)? dirD:dirHV

dir2[x][y]=(d1*hv0>hv1*d0)? dirHV:dirD

dirS[x][y]=(hvd1>2*hvd0)? 1:((hvd1*2>9*hv d0)?2:0)

4) The variable avgVar[x][y] (where x, y = 0…CtbSizeY-1) is derived as follows:

varTab[]＝{0,1,2,2,2,2,2,3,3,3,3,3,3,3,3,4}

avgVar[x][y]＝varTab[Clip3(0,15,(sumOfHV[x>>2][y>>2]*64)>>(3+BitDepthY))]

5) The classification filter index array filtIdx[x][y] and the transpose index array transposeIdx[x][y] (where x = y = 0 … CtbSizeY-1) are derived as follows:

transposeTable[]＝{0,1,0,2,2,3,1,3}

transposeIdx[x][y]＝transposeTable[dir1[x][y]*2+(dir2[x][y]>>1)]

filtIdx[x][y] = avgVar[x][y]

When dirS[x][y] is not equal to 0, filtIdx[x][y] is modified as follows:

filtIdx[x][y]+=(((dir1[x][y]&0x1)<<1)+dirS[x][y])*5

Filtering of chroma samples by coding tree blocks

The inputs to this process are: the reconstructed chroma image sample array recPicture before adaptive loop filtering, the reconstructed chroma image sample array alfPicture after filtering, and the chroma position (xCtbC, yCtbC) of the top-left sample of the current chroma coding tree block for the top-left sample of the current image.

The output of this process is a modified filtered and reconstructed chroma image sample array, alfPicture.

The current chroma coding tree block size ctbSizeC is exported as follows:

ctbSizeC = CtbSizeY / SubWidthC

For the derivation of the filtered reconstructed chroma sample alfPictur[x][y], each reconstructed chroma sample recPicture[x][y] (where x, y = 0…ctbSizeC-1) within the current chroma coding tree block is filtered as follows:

The positions (hx, vy) of each corresponding chroma sample (x, y) within the chroma sample array recPicture are derived as follows:

hx＝Clip3(0,pic_width_in_luma_samples/SubWidthC-1,xCtbC+x)

vy＝Clip3(0,pic_height_in_luma_samples/SubHeightC-1,yCtbC+y)

The variable curr is exported as follows:

curr = recPicture[hx, vy]

The sum of the variables is derived as follows:

sum＝AlfCoeffC[0]*(Clip3(-AlfClipC[0],AlfClipC[0],recPicture[hx,vy+2]-curr)+Clip3(-AlfClipC[0],AlfClipC[0],recPicture[hx,vy-2]-curr))

+AlfCoeffC[1]*(Clip3(-AlfClipC[1],AlfClipC[1],recPicture[hx+1,vy+1]-curr)+Clip3(-AlfClipC[1],AlfClipC[1],recPicture[hx-1,vy-1]-curr ))+AlfCoeffC[2]*(Clip3(-AlfClipC[2],AlfClipC[2],recPicture[hx,vy+1]-curr)+Clip3(-AlfClipC[2],AlfClipC[2],recPicture[hx,vy-1]-curr))

+AlfCoeffC[3]*(Clip3(-AlfClipC[3],AlfClipC[3],recPicture[hx-1,vy+1]-curr)+Clip3(-AlfClipC[3],AlfClipC[3],recPicture[hx+1,vy-1]-curr ))+AlfCoeffC[4]*(Clip3(-AlfClipC[4],AlfClipC[4],recPicture[hx+2,vy]-curr)+Clip3(-AlfClipC[4],AlfClipC[4],recPicture[hx-2,vy]-curr))

+AlfCoeffC[5]*(Clip3(-AlfClipC[5],AlfClipC[5],recPicture[hx+1,vy]-curr)+Clip3(-AlfClipC[5],AlfClipC[5],recPicture[hx-1,vy]-curr))

+AlfCoeffC[6]*recPicture[hx,vy]

sum＝(sum+64)>>7

The modified filtered and reconstructed chroma image sample alfPicture[xCtbC+x][yCtbC+y] is exported as follows:

alfPicture[xCtbC+x][yCtbC+y]＝Clip3(0,(1<<BitDepthC)-1,sum)

In the above variant, the clipping operation is applied to the difference between the neighboring samples and the filtered samples (i.e., using equation (7)). In another variant using equation (9), the semantics and filtering are modified as follows (based on the modifications to the semantics and filtering based on the above variant):

AlfCoeffL[filtIdx][12] is exported as:

AlfCoeffL[filtIdx][12]＝128-Σ _k (AlfCoeffL[filtIdx][k]<<1)(k＝0...11)AlfCoeffC[6] is derived as:

AlfCoeffC[6]＝128-Σ _k (AlfCoeffC[k]<<1)(where k＝0...5)

The bit depth restrictions on AlfCoeffL[filtIdx][12] and AlfCoeffC[6] (i.e., they should be in the range of 0 to 2^8-1 (inclusive)) are relaxed to the range of 0 to 2^9-1 or even 0 to 2^10-1 to obtain better filtering performance.

This is because, for clipping operations, the coefficients associated with neighboring sample positions often need to be higher (in absolute value) to achieve better efficiency (because once the nonlinear function is applied, the filter input samples are more reliable, i.e., more likely to provide better filtering). As a result, the sum of these coefficients (and therefore, excluding the center coefficient) often falls below -128 (in fixed-point representation), necessitating that the center coefficient be higher than 256 (in fixed-point representation). Therefore, limiting the center coefficient to the standard range of 0 to 2^8-1 means the encoder needs to find suboptimal coefficients by reducing its dynamics/amplitude, resulting in less efficient filtering. By using the standard range of 10 bits of precision for the center coefficient (i.e., 0 to 2^10-1), the encoder must use suboptimal filter coefficients much less frequently, thus increasing the overall filtering quality compared to 8 bits of precision. For ALF linear filtering, the 8-bit limitation on the center coefficient does not significantly impact filtering efficiency because neighboring filter input samples typically have smaller weights (i.e., smaller coefficients). However, for clipping, i.e., ALF nonlinear filtering, the optimal filter may assign greater weights to neighboring samples, making clipping more reliable.

The sum of variables in “Block Filtering of Luminance Samples” is derived as follows: sum＝filterCoeff[0]*(Clip3(curr-filterClip[0],curr+filterClip[0],recPictureL[hx,vy+3])+Clip3(curr-filterClip[0],curr+filterClip[0],recPictureL[hx,vy-3]))

+filterCoeff[1]*(Clip3(curr-filterClip[1],curr+filterClip[1],recPictureL[hx+1,vy+2])+Clip3(curr-filterClip[1],curr+filterClip[1],recPictureL[hx-1,vy-2]))

+filterCoeff[2]*(Clip3(curr-filterClip[2],curr+filterClip[2],recPictureL[hx,vy+2])+Clip3(curr-filterClip[2],curr+filterClip[2],recPictureL[hx,vy-2]))

+filterCoeff[3]*(Clip3(curr-filterClip[3],curr+filterClip[3],recPictureL[hx-1,vy+2])+Clip3(curr-filterClip[3],curr+filterClip[3],recPictureL[hx+1,vy-2]))

+filterCoeff[4]*(Clip3(curr-filterClip[4],curr+filterClip[4],recPictureL[hx+2,vy+1])+Clip3(curr-filterClip[4],curr+filterClip[4],recPictureL[hx-2,vy-1]))

+filterCoeff[5]*(Clip3(curr-filterClip[5],curr+filterClip[5],recPictureL[hx+1,vy+1])+Clip3(curr-filterClip[5],curr+filterClip[5],recPictureL[hx-1,vy-1]))

+filterCoeff[6]*(Clip3(curr-filterClip[6],curr+filterClip[6],recPictureL[hx,vy+1])+Clip3(curr-filterClip[6],curr+filterClip[6],recPictureL[hx,vy-1]))

+filterCoeff[7]*(Clip3(curr-filterClip[7],curr+filterClip[7],recPictureL[hx-1,vy+1])+Clip3(curr-filterClip[7],curr+filterClip[7],recPictureL[hx+1,vy-1]))

+filterCoeff[8]*(Clip3(curr-filterClip[8],curr+filterClip[8],recPictureL[hx-2,vy+1])+Clip3(curr-filterClip[8],curr+filterClip[8],recPictureL[hx+2,vy-1]))

+filterCoeff[9]*(Clip3(curr-filterClip[9],curr+filterClip[9],recPictureL[hx+3,vy])+Clip3(curr-filterClip[9],curr+filterClip[9],recPictureL[hx-3,vy]))

+filterCoeff[10]*(Clip3(curr-filterClip[10],curr+filterClip[10],recPictureL[hx+2,vy])+Clip3(curr-filterClip[10],curr+filterClip[10],recPictureL[hx-2,vy]))

+filterCoeff[11]*(Clip3(curr-filterClip[11],curr+filterClip[11],recPictureL[hx+1,vy])+Clip3(curr-filterClip[11],curr+filterClip[11],recPictureL[hx-1,vy]))

+filterCoeff[12]*recPictureL[hx,vy]

sum＝(sum+64)>>7

The sum of variables in “color sample coding tree block filtering processing” is derived as follows: sum＝AlfCoeffC[0]*(Clip3(curr-AlfClipC[0],curr+AlfClipC[0],recPicture[hx,vy+2])+Clip3(curr-AlfClipC[0],AlfClipC[0],recPicture[hx,vy-2]))

+AlfCoeffC[1]*(Clip3(curr-AlfClipC[1],curr+AlfClipC[1],recPicture[hx+1,vy+1])+Clip3(curr-AlfClipC[1],curr+AlfClipC[1],recPicture[hx-1,vy-1]))

+AlfCoeffC[2]*(Clip3(curr-AlfClipC[2],curr+AlfClipC[2],recPicture[hx,vy+1])+Clip3(curr-AlfClipC[2],curr+AlfClipC[2],recPicture[hx,vy-1]))+A lfCoeffC[3]*(Clip3(curr-AlfClipC[3],curr+AlfClipC[3],recPicture[hx-1,vy+1])+Clip3(curr-AlfClipC[3],curr+AlfClipC[3],recPicture[hx+1,vy-1]))

+AlfCoeffC[4]*(Clip3(curr-AlfClipC[4],curr+AlfClipC[4],recPicture[hx+2,vy])+Clip3(curr-AlfClipC[4],curr+AlfClipC[4],recPicture[hx-2,vy]))+AlfCoeffC[5]*( Clip3(curr-AlfClipC[5],curr+AlfClipC[5],recPicture[hx+1,vy])+Clip3(curr-AlfClipC[5],curr+AlfClipC[5],recPicture[hx-1,vy]))+AlfCoeffC[6]*recPicture[hx,vy]

sum＝(sum+64)>>7

In another variation using the 5×5 diamond filter shape for the brightness filter (to reduce the number of buffer rows required for ALF application), this variation also uses a limiting applied to the difference between neighboring sample values and the center sample value. The nonlinear ALF data syntax in Table 4 is modified as follows (changes to the previously presented ALF data syntax are underlined and bolded, and repetitions are omitted where possible):

Table 5 - Modifications to the nonlinear ALF data syntax

The number of ALF coefficients in the luminance filter becomes 13, therefore the final coefficient filter initialization process becomes:

The final filter coefficients AlfCoeffL[filtIdx][ 6 ] for filtIdx = 0..NumAlfFilters-1 are derived as follows:

AlfCoeffL[filtIdx][ 6 ] = 128

The requirement for bitstream consistency is that AlfCoeffL[filtIdx][j] (where iltIdx = 0…NumAlfFilters-1, j = 0…11) should be in the range of -2^7 to 2^7-1 (inclusive), and the value of AlfCoeffL[filtIdx][ 6 ] should be in the range of 0 to 2^8-1 (inclusive).

For "Card block filtering of luminance samples", the export processing of the filter index clause "ALF transpose of luminance samples and export processing of filter index" is modified as follows (changes to the previously presented ALF data syntax are underlined and bolded, and repeated parts are omitted as much as possible):

The export processing of the filter index clause "ALF transpose of luminance samples and export processing of filter index" is invoked with the position (xCtb, yCtb) and the reconstructed luminance image sample array recPictureL as inputs, and with filtIdx[x][y] and transposeIdx[x][y] (where x, y = 0…CtbSizeY-1) as outputs.

For the derivation of the filtered reconstructed luminance sample alfPictureL[x][y], each reconstructed luminance sample recPictureL[x][y] (where x, y = 0…CtbSizeY-1) within the current luminance coding tree block is filtered as follows:

The luminance filter coefficient array f[j] and luminance filter limiting array c[j] (where j = 0… 6 ) corresponding to the filter specified by filtIdx[x][y] are derived as follows:

f[j]＝AlfCoeffL[filtIdx[x][y]][j]

c[j]＝AlfClipL[filtIdx[x][y]][j]

Based on transposeIdx[x][y], derive the brightness filter coefficients filterCoeff and filter clipping value filterClip as follows:

If transposeIndex[x][y] = 1,

filterCoeff[]＝ {f[4],f[1],f[5],f[3],f[0],f[2],f[6]}

filterClip[]＝ {c[4],c[1],c[5],c[3],c[0],c[2],c[6]}

Otherwise, if transposeIndex[x][y] == 2,

filterCoeff[]＝ {f[0],f[3],f[2],f[1],f[4],f[5],f[6]}

filterClip[]＝ {c[0],c[3],c[2],c[1],c[4],c[5],c[6]}

Otherwise, if transposeIndex[x][y] == 3,

filterCoeff[]＝ {f[4],f[3],f[5],f[1],f[0],f[2],f[6]}

filterClip[]＝ {c[4],c[3],c[5],c[1],c[0],c[2],c[6]}

otherwise,

filterCoeff[]＝ {f[0],f[1],f[2],f[3],f[4],f[5],f[6]}

filterClip[]＝ {c[0],c[1],c[2],c[3],c[4],c[5],c[6]}

The positions (hx, vy) of each corresponding luminance sample (x, y) within the luminance sample array recPicture are derived as follows:

hx＝Clip3(0,pic_width_in_luma_samples-1,xCtb+x)

vy＝Clip3(0,pic_height_in_luma_samples-1,yCtb+y)

The variable curr is derived as follows: curr = recPictureL[hx, vy]

The sum of the variables is derived as follows:

sum=filterCoeff[0]*(Clip3(-filterClip[0],filterClip[0],recPictureL[hx,vy+2]-curr)+Clip3(-filterClip[0],filterClip[0],recPictureL[hx,vy-2]-curr))

+filterCoeff[1]*(Clip3(-filterClip[1],filterClip[1],recPictureL[hx+1,vy+1]-curr)+Clip3(-filterClip[1],filterClip[1],recPictureL[hx-1,vy-1]-curr))

+filterCoeff[2]*(Clip3(-filterClip[2],filterClip[2],recPictureL[hx,vy+1]-curr)+Clip3(-filterClip[2],filterClip[2],recPictureL[hx,vy-1]-curr))

+filterCoeff[3]*(Clip3(-filterClip[3],filterClip[3],recPictureL[hx-1,vy+1]-curr)+Clip3(-filterClip[3],filterClip[3],recPictureL[hx+1,vy-1]-curr))

+filterCoeff[4]*(Clip3(-filterClip[4],filterClip[4],recPictureL[hx+2,vy]-curr)+Clip3(-filterClip[4],filterClip[4],recPictureL[hx-2,vy]-curr))

+filterCoeff[5]*(Clip3(-filterClip[5],filterClip[5],recPictureL[hx+1,vy]-curr)+Clip3(-filterClip[5],filterClip[5],recPictureL[hx-1,vy]-curr))

+filterCoeff[6]*recPictureL[hx,vy]

sum＝(sum+64)>>7

The modified filtered and reconstructed luminance image sample alfPictureL[xCtb+x][yCtb+y] is exported as follows: alfPictureL[xCtb+x][yCtb+y]＝Clip3(0,(1<<BitDepthY)-1,sum)

In the variant that does not use the difference between neighbors and the center, AlfCoeffL[filtIdx][ 6 ] = 128 - _Σk (AlfCoeffL[filtIdx][k] << 1) (where k = 0… 5 ). A similar change is made to the limiting parameter in the filtering process.

In another variation, a 7×7 rhombus shape is used as the filter, but the limiting pattern 1601 of Figure 16-a is used as the luminance filter, and the limiting pattern 1605 of Figure 16-b is used as the chrominance filter. The nonlinear ALF data syntax in Table 4 is modified as follows (the changes made to the previously presented ALF data syntax in Table 4 are underlined and bolded, and repeated parts are omitted as much as possible):

Table 6 - Modifications to the nonlinear ALF data syntax

AlfClipPatL[] is a constant limiting pattern table for the brightness filter, and is set as follows: AlfClipPatL[] = {0,0,1,0,0,0,1,0,0,0,1,1}

AlfClipPatC[] is a constant clipping pattern table for the chromaticity filter, and it is set as follows: AlfClipPatC[] = {1,1,0,1,1,0}

Some semantic modifications are as follows:

The total variables in "Encoded Tree Block Filtering Process for Luminance Samples" are derived as follows:

sum=filterCoeff[0]*(recPictureL[hx,vy+3]-curr+recPicture L[hx,vy-3]-curr)

+filterCoeff[1]*(recPictureL[hx+1,vy+2]-curr)+recPictur eL[hx-1,vy-2]-curr)

+filterCoeff[2]*(Clip3(-filterClip[2],filterClip[2],recPictureL[hx,vy+2]-curr)+Clip3(-filterClip[2],filterClip[2],recPictureL[hx,vy-2]-curr))

+filterCoeff[3]*(recPictureL[hx-1,vy+2]-curr+recPicture L[hx+1,vy-2]-curr)

+filterCoeff[4]*(recPictureL[hx+2,vy+1]-curr+recPicture L[hx-2,vy-1]-curr)

+filterCoeff[5]*(recPictureL[hx+1,vy+1]-curr+recPicture L[hx-1,vy-1]-curr)

+filterCoeff[6]*(Clip3(-filterClip[6],filterClip[6],recPictureL[hx,vy+1]-curr)+Clip3(-filterClip[6],filterClip[6],recPictureL[hx,vy-1]-curr))

+filterCoeff[7]*(recPictureL[hx-1,vy+1]-curr+recPicture L[hx+1,vy-1]-curr)

+filterCoeff[8]*(recPictureL[hx-2,vy+1]-curr+recPicture L[hx+2,vy-1]-curr)

+filterCoeff[9]*(recPictureL[hx+3,vy]-curr)+recPictureL[hx-3,vy]-curr)

+filterCoeff[10]*(Clip3(-filterClip[10], filterClip[10],recPictureL[hx+2,vy]-curr)+Clip3(-filterClip[10],filterClip[10],recPictureL[hx-2,vy]-curr))

+filterCoeff[11]*(Clip3(-filterClip[11], filterClip[11],recPictureL[hx+1,vy]-curr)+ Clip3(-filterClip[11],filterClip[11],recPictureL[hx-1,vy]-curr))

+filterCoeff[12]*recPictureL[hx,vy]

sum＝(sum+64)>>7

The sum of variables in “color sample coding tree block filtering processing” is derived as follows: sum＝AlfCoeffC[0]*(Clip3(-AlfClipC[0],AlfClipC[0],recPicture[hx,vy+2]-curr)+Clip3(-AlfClipC[0],AlfClipC[0],recPicture[hx,vy-2]-curr))

+AlfCoeffC[1]*(Clip3(-AlfClipC[1],AlfClipC[1],recPicture[hx+1,vy+1]-curr)+Clip3(-AlfClipC[1],AlfCli pC[1],recPicture[hx-1,vy-1]-curr))+AlfCoeffC[2]*(recPicture[hx,vy+1]-curr+recPicture[hx,vy-1]-curr)

+AlfCoeffC[3]*(Clip3(-AlfClipC[3],AlfClipC[3],recPicture[hx-1,vy+1]-curr)+Clip3(-AlfClipC[3],AlfClipC[3],recPicture[hx+1,vy-1]-curr ))+AlfCoeffC[4]*(Clip3(-AlfClipC[4],AlfClipC[4],recPicture[hx+2,vy]-curr)+Clip3(-AlfClipC[4],AlfClipC[4],recPicture[hx-2,vy]-curr))

+AlfCoeffC[5]*(recPicture[hx+1,vy]-curr+recPicture[hx-1,vy]-curr)

+AlfCoeffC[6]*recPicture[hx,vy]

It should be understood that, based on the above variations and other variations using a reduced number of limiting positions (i.e., a reduced number of limiting positions, for example, for the luminance component, not applying limiting to the current (center) sample value “curr” so that “f[12]”, “c[12]”, “filterCoeff[12]” are not present, and “sum＝curr+((sum+64)>>7)” (as in Equation (17) and Figure 7) and/or not using the difference between neighboring samples and the center sample, other variations using other limiting patterns can be readily derived.

Figure 8 is a block diagram of a linear filter obtained according to an embodiment of the present invention when the nonlinear portion of the filter in Figure 7 is disabled. According to the embodiment, the linear ALF is used for filtering when the ALF is active and the nonlinear ALF is inactive. In a variation of this embodiment, the nonlinear ALF is used as defined by equation (7) and as shown in Figure 7, instead of the linear ALF as defined by equation (4) and as shown in Figure 6, and linear filtering is performed as defined by equation (6) and as shown in Figure 8. In the hardware design, it is permissible to share common circuitry between the nonlinear filter and the linear filter. When comparing Figures 7 and 8, the min/max operation is bypassed, and the process proceeds directly to the sample value. Another advantage of using this method is that when the encoder designs its filters, it does not have to ensure that the sum of all filter coefficients of each filter is 1. The result of this design is that, in addition to the current sample, a subtraction is added for each coefficient, and a multiplication of the current sample is removed.

When the maximum value of the limiting parameter is sufficiently low, the design in Figure 7 also has the advantage of saving hardware bits used to implement multiplication.

According to an embodiment, the syntax elements of the ALF that are signaled in the slice header in FIG5 are no longer signaled in the slice header, but are signaled at the frame level, such as in the image header. Then, one or more flags/syntax elements exist in the slice header to signal that parameters provided at the frame level are directly used in the slice; or that parameters provided at the frame level must be "refreshed" (i.e., updated) at the slice level. Slice-level refreshes are performed in the slice header. Refreshing filter coefficients and/or limiting parameters is accomplished by encoding the difference between the values indicated at the frame level and the values used in the slice. According to an embodiment, additional syntax elements in the slice header indicate whether the values of filter coefficients and/or limiting parameters are encoded as differences in the slice header, or indicate whether their values are encoded.

According to embodiments, filter coefficients and/or limiting parameters defined at the slice level (in the slice header) can be updated at a finer granular level (e.g., for each individual CTU). Syntax elements indicating whether an update is necessary, and update data when the syntax elements indicate that an update is necessary, are entropy-encoded in the slice data. In one embodiment, in the coded bitstream, for each CTU, the coded slice data includes an entropy-encoded flag indicating whether the luma filter/limiting parameter must be updated. If the flag indicates that an update should be performed, then for each filter coefficient table, the entropy-encoded flag indicates whether the filter must be updated. If the filter must be updated, the offsets of the coefficients (differences from reference values in the slice or from reference values in previously coded CTUs belonging to the same slice) and/or the offsets of the limiting parameters are entropy-encoded.

According to the embodiment, from the encoder's perspective, the main logical steps of the ALF encoding process of VTM-3.0 described with reference to Figure 9 are retained. The main changes are the statistical properties extracted for filter construction and the way the filter is determined. Instead of simple Wiener filter calculations, the limiting parameters are iteratively optimized to find the optimal limiting parameters and associated nonlinear Wiener filter coefficients.

According to the embodiment, we define a limit _NK of the number of possible values for the limiting parameter. _NK can be small, for example, _NK = 12 or even smaller. Then, instead of calculating the covariance and cross-covariance matrix statistics, in step 902, the encoder constructs a bi-item table E of the _NK × _NK limiting covariance matrix statistics and a single-item table y of the limiting cross-covariance matrix statistics _NK (instead of one covariance matrix statistics and one cross-covariance matrix statistics).

The limiting covariance matrix E c [a][b] is the covariance matrix estimated from (where X m  contains N implementations (observed values) of the input when using the C m -th limiting value). For example, using the shape of Figure 4-b, is the i-th column of X m  obtained for a given sample position and contains: for i <Nc (for the brightness filter of Figure 4-b, Nc = 12 is the number of filter coefficients minus 1), the sum of the results of limiting the difference between the neighboring sample (of the j-th filtered sample) and the j-th filtered sample using the m-th limiting value for two symmetrically adjacent positions with index i in the shape; and for i = 12, the j-th filtered sample; where is the value at row i and column j in matrix X m .

The amplitude-limited cross-covariance matrix _yc [c] is the cross-covariance matrix between the realization of the desired filter output (i.e. the original sample) and the realization of ^Xc .

Therefore, when we need to construct/test filters using parameters equal to the limiting parameters, we define (e.g., for a brightness filter, N _{_c} = 12 is the number of filter coefficients minus 1; it's convenient for the final value to be zero, other values can be used and have the same result (because the center coefficient 12 is not limited)). We compute the covariance matrix E and the cross-covariance matrix y, these limiting parameters such that E[i,j] = E _{_c} [c[i]][c[j]][i,j] (for each i <N_{_c} , j <N_{_c} ) and y[k] = _y _c[c[k]][k] (for each k <N_{_c} ). Then, as explained earlier for linear ALF, we use these two matrices to construct the Wiener filter.

In step 904, instead of combining the statistics of the covariance and cross-covariance matrices, the encoder combines the statistics of each limiting covariance and limiting cross-covariance matrix (which is also done by summing the statistics).

In step 905, the method for finding the optimized filter bank (i.e., the optimized combined filter) is similar to before. The main difference is that, for each test/comparison filter, instead of calculating only one Wiener filter, the limiting parameters are iteratively optimized. Each step calculates a Wiener filter for a specific limiting parameter to minimize the filter's output error (based on statistical matrix calculations). A simple optimization algorithm starts by setting all limiting parameter values ( _ci ) to zero (or alternatively, intermediate limiting values or any default values), and then iteratively performs the following steps while improving the filter (i.e., reducing output error): for each limiting parameter value, if the Wiener filter calculated using the next limiting parameter value (if available/allowed) (i.e., index < maximum index value) is better, then that limiting parameter value is adopted; or if the Wiener filter calculated using the previous limiting parameter value (if available/allowed (i.e., index > 0)) is better, then that limiting parameter value is adopted. When grouping two filters, one strategy might be to start optimization by setting each limiting parameter value to the average of the individual limiting parameter values of the two filters.

There are no actual changes to steps 906, 907, 908, 909, 910, and 911.

In step 912, the combined statistical changes are the same as described in step 904. The determination of the chroma filter in step 913 corresponds to the filter optimization described in step 905. Except for step 918, the other steps remain unchanged, in which the ALF parameters include the limiting parameters determined for each filter.

Note that in VTM-3.0 with a "normal" ALF, when the encoder derives integer filter coefficients from floating-point values, it first performs floating-point to fixed-point quantization: for each filter coefficient, it multiplies the filter coefficient by the fixed-point precision and rounds it to the nearest integer value. It then attempts to improve filter efficiency by iteratively adjusting the filter coefficients (to compensate for rounding errors). The encoder must also ensure that the sum of all filter coefficients equals 1 (multiplied by the fixed-point precision) and that the filter coefficient of the center sample does not exceed its maximum value (to ensure that its use in multiplication does not end up using too many bits).

In embodiments of the invention, the difference with neighboring samples is used as the input to the filter because the center coefficient is always 1, so it is not necessary to ensure that it does not exceed the maximum value. Furthermore, it is not always necessary to track the sum of all filter coefficients.

The advantage of these embodiments is improved coding efficiency.

The descriptions of these embodiments mention luminance and chromaticity components, but can be easily adjusted to apply to other components (such as a single luminance component or RGB components).

In the foregoing embodiments or variations, the present invention modifies the filters in 304 and 407 of Figures 3-a and 4-a to use a nonlinear function on the filter input to obtain a nonlinear filtering effect. Figure 2 is also modified to add new syntax elements for signaling additional parameters of the nonlinear function (an example of the embodiment is described with reference to Figure 5). Implicitly, the input filter parameters in 302 and 402 are modified to further include additional parameters for the nonlinear function. Finally, the filter derivation in 303 and 406 is modified to derive the nonlinear function of the filter (i.e., derive the “nonlinear filter”). Therefore, in most cases, the output image portions 305 and 408 differ from the output image portions produced in VTM-3.0 and have higher quality for equivalent/comparable input samples, and/or achieve a better trade-off between output image quality and coding efficiency. It should be understood that in another embodiment/variation, the ALF itself is modified such that it operates using any of the aforementioned nonlinear functions for filtering.

Implementation of the embodiments of the present invention

One or more of the foregoing embodiments can be implemented in the form of an encoder or decoder that performs the method steps of one or more of the foregoing embodiments. The following embodiments illustrate such an implementation.

For example, the adaptive loop filter according to any of the foregoing embodiments can be used for post-filtering 9415 performed by the encoder in FIG10 or post-filtering 9567 performed by the decoder in FIG11.

Figure 10 shows a block diagram of an encoder according to an embodiment of the present invention. The encoder is represented by connected modules, each module being adapted to implement, for example, in the form of programming instructions executed by the central processing unit (CPU) of the device, at least one corresponding step of a method for encoding images of an image sequence according to one or more embodiments of the present invention.

The encoder 9400 receives the raw sequence of digital images i0 to in 9401 as input. Each digital image is represented by a set of samples (sometimes also called pixels (hereinafter referred to as pixels)). After the encoding process is performed, the bit stream 9410 is output by the encoder 9400. The bit stream 9410 includes data of multiple coding units or image portions such as slices, each slice including a slice header for transmitting coding values of coding parameters used to encode the slice and a slice body including encoded video data.

Module 9402 divides the input digital image i0 to in 9401 into pixel blocks. These blocks correspond to image portions and can be of variable size (e.g., 4×4, 8×8, 16×16, 32×32, 64×64, 128×128 pixels, and rectangular block sizes can also be considered). An encoding mode is selected for each input block. Two types of encoding modes are provided: spatial prediction-based encoding modes (intra-frame prediction) and temporal prediction-based encoding modes (inter-frame coding, merging, skipping). Possible encoding modes are tested.

Module 9403 implements intra-frame prediction processing, wherein the block to be encoded is predicted by predictors computed from the neighboring pixels of the given block. If intra-frame coding is selected, an indication of the selected intra-frame predictor and the difference between the given block and its predictor are encoded to provide a residual.

Temporal prediction is implemented by motion estimation module 9404 and motion compensation module 9405. First, a reference image is selected from the reference image set 9416, and motion estimation module 9404 selects a portion of the reference image (also called a reference region or image portion, which is the region closest to the given block to be encoded (closest in terms of pixel value similarity)). Then, motion compensation module 9405 uses the selected region to predict the block to be encoded. The difference between the selected reference region and the given block (also called residual block/data) is calculated by motion compensation module 9405. Motion information (e.g., motion vectors) is used to indicate the selected reference region.

Therefore, in both cases (spatial and temporal predictions), when not in SKIP mode, the residuals are calculated by subtracting the predictors from the original block.

In intra-frame prediction implemented by module 9403, the prediction direction is encoded. In inter-frame prediction implemented by modules 9404, 9405, 9416, 9418, and 9417, at least one motion vector or information (data) used to identify such a motion vector is encoded for temporal prediction. If inter-frame prediction is selected, information associated with the motion vector and residual block is encoded. To further reduce the bit rate, assuming uniform motion, the motion vector is encoded by the difference relative to the motion vector predictor. Motion vector predictors from the motion information predictor candidate set are obtained from the motion vector field 9418 by the motion vector prediction and encoding module 9417.

The encoder 9400 also includes a selection module 9406, which selects an encoding mode by applying an encoding cost criterion such as a rate-distortion criterion. To further reduce redundancy, a transform module 9407 applies a transform (such as DCT) to the residual block, and the resulting transformed data is then quantized by a quantization module 9408 and entropy encoded by an entropy encoding module 9409. Finally, when not in SKIP mode and the selected encoding mode requires encoding of the residual block, the encoded residual block of the current block being encoded is inserted into the bitstream 9410.

Encoder 9400 also decodes the encoded image to generate a reference image for motion estimation of subsequent images (e.g., a reference image in reference image/picture 9416). This allows the encoder and decoder receiving the bitstream to have the same reference frame (e.g., using a reconstructed image or a portion of a reconstructed image). Inverse quantization (“dequantization”) module 9411 performs inverse quantization (“dequantization”) of the quantized data, which is then inversely transformed by inverse transform module 9412. Intra-frame prediction module 9413 uses the prediction information to determine which predictor to use for a given block, and motion compensation module 9414 actually adds the residual obtained by module 9412 to the reference region obtained from reference image set 9416. Then, post-filtering is applied by module 9415 to filter the reconstructed frame (image or image portion) of pixels to obtain another reference image of reference image set 9416.

Figure 11 shows a block diagram of a decoder 9560 that can be used to receive data from an encoder according to an embodiment of the present invention. The decoder is represented by connected modules, each module being adapted to implement corresponding steps of the method implemented by the decoder 9560, for example, in the form of programming instructions executed by the CPU of the device.

Decoder 9560 receives bitstream 9561, which includes coding units (e.g., data corresponding to an image portion, block, or coding unit). Each coding unit consists of a header containing information related to coding parameters and a body containing coded video data. As illustrated with respect to FIG10, the coded video data is entropy-coded at a predetermined number of bits for a given image portion (e.g., block or CU), and motion information (e.g., indices of motion vector predictors) is encoded. The received coded video data is entropy-decoded by module 9562. The residual data is then dequantized by module 9563, and then an inverse transform is applied by module 9564 to obtain pixel values.

The mode data indicating the encoding mode is also entropy-decoded, and based on this mode, intra-type or inter-type decoding is performed on the encoded blocks (units/sets/groups) of the image data. In the case of intra-mode, the intra-prediction module 9565 determines the intra-predictors based on the intra-prediction mode specified in the bitstream (e.g., the intra-prediction mode can be determined using data provided in the bitstream). If the mode is inter-mode, motion prediction information is extracted/obtained from the bitstream to locate (identify) the reference region used by the encoder. For example, the motion prediction information includes the reference frame index and motion vector residuals. The motion vector predictors are added to the motion vector residuals by the motion vector decoding module 9570 to obtain the motion vectors.

Motion vector decoding module 9570 applies motion vector decoding to each image segment (e.g., the current block or CU) encoded via motion prediction. Once the index of the motion vector predictor for the current block is obtained, the actual values of the motion vectors associated with the image segment (e.g., the current block or CU) can be decoded and used to apply motion compensation via module 9566. Reference image segments indicated by the decoded motion vectors are extracted/obtained from reference image set 9568, enabling module 9566 to perform motion compensation. Motion vector field data 9571 is updated with the decoded motion vectors for predicting subsequently decoded motion vectors.

Finally, the decoded block is obtained. Where appropriate, post-filtering is applied by the post-filtering module 9567. The resulting decoded video signal 9569 is then provided by the decoder 9560.

Figure 12 illustrates a data communication system that can implement one or more embodiments of the present invention. The data communication system includes a transmission device (in this case, server 9201) operable to transmit data packets of a data stream to a receiving device (in this case, client terminal 9202) via a data communication network 9200. The data communication network 9200 can be a wide area network (WAN) or a local area network (LAN). Such a network can be, for example, a wireless network (Wifi/802.11a or b or g), an Ethernet network, an Internet network, or a hybrid network consisting of several different networks. In a particular embodiment of the invention, the data communication system can be a digital television broadcasting system, wherein server 9201 sends the same data content to multiple clients. The data stream 9204 provided by server 9201 can consist of multimedia data representing video and audio data. In some embodiments of the invention, the audio and video data streams can be captured by server 9201 using a microphone and a camera, respectively. In some embodiments, the data stream can be stored on server 9201 or received by server 9201 from other data providers, or generated at server 9201. Server 9201 is equipped with an encoder for encoding video and audio streams, specifically for providing a compressed bitstream for transmission, which is a more compact representation of the data presented as input to the encoder. To achieve a better ratio of transmitted data quality to transmitted data volume, the video data can be compressed, for example, according to High Efficiency Video Coding (HEVC) format, H.264/Advanced Video Coding (AVC) format, or VVC format. Client 9202 receives the transmitted bitstream and decodes the reconstructed bitstream to reproduce the video image on a display device and reproduce the audio data using speakers.

Although this embodiment considers a streaming scenario, it will be appreciated that in some embodiments of the invention, media storage devices such as optical discs can be used for data communication between the encoder and decoder. In one or more embodiments of the invention, video images may be transmitted along with data representing compensation offsets of the reconstructed pixels to be applied to the image, in order to provide filtered pixels in the final image.

Figure 13 schematically illustrates a processing device 9300 configured to implement at least one embodiment of the present invention. The processing device 9300 may be a device such as a microcomputer, workstation, user terminal, or lightweight portable device. The device/apparatus 9300 includes a communication bus 9313 connected to:

- This refers to the CPU's central processing unit 9311, such as a microprocessor;

- Read-only memory 9307, denoted as ROM, is used to store operating device 9300 and/or computer program/instructions for implementing the present invention;

- A random access memory 9312, represented as RAM, for storing executable code of the methods according to embodiments of the present invention, and registers suitable for recording variables and parameters required for implementing the method of encoding a digital image sequence and/or the method of decoding a bit stream according to embodiments of the present invention; and

- A communication interface 9302 connected to a communication network 9303, through which digital data to be processed is transmitted or received.

Optionally, the device 9300 may also include the following components:

- A data storage component 9304, such as a hard disk, is used to store a computer program for implementing one or more embodiments of the present invention, as well as data used or generated during the implementation of one or more embodiments of the present invention;

- A disk drive 9305 for disk 9306 (e.g., storage medium), the disk drive 9305 being adapted to read data from disk 9306 or write data to disk 9306, or;

- Screen 9309, which is used to display data and/or serve as a graphical interface for user interaction by means of keyboard 9310, touch screen or any other indication/input device.

Device 9300 can be connected to various peripheral devices such as digital camera 9320 or microphone 9308, each of which is connected to an input/output card (not shown) to provide multimedia data to device 9300.

The communication bus provides communication and interoperability between various elements included in or connected to the device 9300. The representation of the bus is not limiting, and in particular, the central processing unit 9311 is operable to communicate instructions directly or by means of other elements of the device 9300 to any element of the device 9300.

Disk 9306 may be replaced by any information medium such as a rewritable or non-rewritable compact disc (CD-ROM), ZIP disc, or memory card, and generally by an information storage component that can be read by a microcomputer or processor. Disk 9306 may be integrated into or not integrated into a device, may be portable, and is adapted to store one or more programs that execute to enable the implementation of a method for encoding digital image sequences and/or a method for decoding bit streams according to the present invention.

Executable code can be stored in read-only memory 9306, hard disk 9304, or removable digital media (such as, for example, disk 9306 as described above). According to a variant, the executable code of the program can be received via interface 9302 through communication network 9303 to be stored in one of the storage components of device 9300 (e.g., hard disk 9304) before execution.

The central processing unit 9311 is adapted to control and direct the execution of instructions or software code portions of one or more programs according to the invention, and instructions stored in one of the aforementioned storage components. Upon power-up, one or more programs stored in non-volatile memory (e.g., on hard disk 9304, disk 9306, or in read-only memory 9307) are transferred to random access memory 9312 (which then contains the executable code of one or more programs) and registers for storing variables and parameters necessary for implementing the invention.

In this embodiment, the device is a programmable device that implements the invention using software. However, alternatively, the invention can be implemented in hardware (e.g., in the form of an application-specific integrated circuit or ASIC).

Implementation of the embodiments of the present invention

Implementation of the embodiments of the present invention

It should also be understood that, according to other embodiments of the invention, a decoder according to the above embodiments is provided in a user terminal such as a computer, mobile phone (cellular phone), tablet, or any other type of device capable of providing/displaying content to a user (e.g., a display device). According to yet another embodiment, an encoder according to the above embodiments is provided in an image capture device, which further includes a camera, video camera, or webcam (e.g., a closed-circuit television or video surveillance camera) for capturing and providing content for encoding by the encoder. Two such embodiments are shown below with reference to Figures 14 and 15.

Figure 14 is a diagram illustrating a webcam system 9450 including a webcam 9452 and a client device 9454.

The webcam 9452 includes an image-capturing unit 9456, an encoding unit 9458, a communication unit 9460, and a control unit 9462. The webcam 9452 and a client device 9454 are interconnected via a network 9200 to enable communication between them. The image-capturing unit 9456 includes a lens and an image sensor (e.g., a charge-coupled device (CCD) or complementary metal-oxide-semiconductor (CMOS)), captures an image of a subject, and generates image data based on the image. The image can be a still image or a video image. The image-capturing unit may also include zoom and/or pan components, respectively adapted for zooming or panning (optically or digitally). The encoding unit 9458 encodes the image data using the encoding methods described in the first to fifth embodiments. The encoding unit 9458 uses at least one of the encoding methods described in the first to fifth embodiments. For other examples, the encoding unit 9458 may use a combination of the encoding methods described in the first to fifth embodiments.

The communication unit 9460 of the network camera 9452 transmits the encoded image data, which is encoded by the encoding unit 9458, to the client device 9454.

Furthermore, the communication unit 9460 can also receive commands from the client device 9454. These commands include commands for setting parameters for encoding by the encoding unit 9458. The control unit 9462 controls other units in the network camera 9452 based on the commands received by the communication unit 9460. The client device 9454 includes a communication unit 2114, a decoding unit 2116, and a control unit 9468. The communication unit 2118 of the client device 9454 can transmit commands to the network camera 9452. Additionally, the communication unit 2118 of the client device 9454 receives encoded image data from the network camera 9452. The decoding unit 9464 decodes the encoded image data using one or more of the decoding methods described in the foregoing embodiments. For other instances, the decoding unit 9464 may use a combination of the decoding methods described in the foregoing embodiments. The control unit 9468 of the client device 9454 controls other units in the client device 9454 based on user operations or commands received by the communication unit 2114. The control unit 9468 of the client device 9454 can also control the display device 9470 to display the image decoded by the decoding unit 9466.

The control unit 9468 of the client device 9454 also controls the display device 9470 to display the values of the parameters for specifying the network camera 9452 (e.g., the parameters for encoding the encoding unit 9458).

The control unit 9468 of the client device 9454 can also control other units in the client device 9454 based on user input via the GUI displayed on the display device 9470. The control unit 9468 of the client device 9454 can also control the communication unit 9464 of the client device 9454 based on user input via the GUI displayed on the display device 9470, to transmit commands specifying values for parameters of the network camera 9452 to the network camera 9452.

Figure 15 is a diagram illustrating a smartphone 9500. The smartphone 9500 includes a communication unit 9502, a decoding/encoding unit 9504, a control unit 9506, and a display unit 9508. The communication unit 9502 receives encoded image data via a network. The decoding/encoding unit 9504 decodes the encoded image data received by the communication unit 9502. The decoding unit 9504 decodes the encoded image data using one or more of the decoding methods described in the foregoing embodiments. The decoding/encoding unit 9504 may also use at least one of the encoding or decoding methods described in the foregoing embodiments. In other instances, the decoding/encoding unit 9504 may use a combination of the decoding or encoding methods described in the foregoing embodiments. The control unit 9506 controls other units in the smartphone 9500 according to user operations or commands received by the communication unit 9502. For example, the control unit 9506 controls the display unit 9508 to display the image decoded by the decoding/encoding unit 9504.

The smartphone may also include an image recording device 9510 (e.g., a digital camera and associated circuitry) for recording images or videos. Such recorded images or videos can be encoded by a decoding/encoding unit 9504 under the instruction of a control unit 9506. The smartphone may also include a sensor 9512 adapted to sense the orientation of the mobile device. Such a sensor may include an accelerometer, gyroscope, compass, global positioning (GPS) unit, or similar position sensor. This sensor 9512 can determine whether the smartphone has changed orientation, and such information can be used when encoding video streams.

Although the invention has been described with reference to embodiments, it should be understood that the invention is not limited to the disclosed embodiments. Those skilled in the art will understand that various changes and modifications can be made without departing from the scope of the invention as defined by the appended claims. All features disclosed in this specification (including any appended claims, abstract, and drawings), and/or all steps of any disclosed method or process, can be combined in any combination except for at least some mutually exclusive combinations of such features and/or steps. Unless otherwise expressly stated, the various features disclosed in this specification (including any appended claims, abstract, and drawings) may be replaced by alternative features for the same, equivalent, or similar purposes. Therefore, unless otherwise expressly stated, the various features disclosed are merely examples of a general series of equivalent or similar features.

It should also be understood that any result of the above comparison, determination, evaluation, selection, execution, conduct, or consideration (e.g., a selection made during encoding or filtering processing) may be indicated in or determined/inferred from data in the bitstream (e.g., flags or information indicating the result), such that the indicated or determined/inferred result can be used in processing, rather than actually being performed, for example, during decoding processing.

In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite articles "a" or "an" do not exclude multiple elements. The mere fact that different features are recited in mutually different dependent claims does not indicate that a combination of these features cannot be used advantageously. Reference numerals appearing in the claims are for illustrative purposes only and should not be used to limit the scope of the claims.

In the foregoing embodiments, the described functions can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions can be stored as one or more instructions or code on or transmitted through a computer-readable medium, and can be executed by a hardware-based processing unit.

Computer-readable media may include computer-readable storage media, which correspond to tangible media such as data storage media or communication media that include, for example, any medium facilitating the transfer of a computer program from one place to another according to a communication protocol. In this way, computer-readable media may generally correspond to (1) a non-transitory tangible computer-readable storage medium or (2) a communication medium such as a signal or carrier wave. Data storage media may be any available medium accessible by one or more computers or one or more processors to retrieve instructions, code, and/or data structures for implementing the techniques described herein. Computer program products may include computer-readable media.

By way of example and not limitation, such computer-readable storage media may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer. Furthermore, any connection may be appropriately referred to as a computer-readable medium. For example, coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technology (such as infrared, radio, and microwave) are included in the definition of medium if instructions are sent from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technology (such as infrared, radio, and microwave). However, it should be understood that computer-readable storage media and data storage media do not include connections, carrier waves, signals, or other transient media, but rather refer to non-transient tangible storage media. The terms disk and disc used herein include compact discs (CDs), laser discs, optical discs, digital versatile discs (DVDs), floppy disks, and Blu-ray discs, where disks typically copy data magnetically, while discs optically reproduce data using lasers. The above combinations should also be included within the scope of computer-readable media.

Instructions can be executed by one or more processors, such as one or more digital signal processors (DSPs), general-purpose microprocessors, application-specific integrated circuits (ASICs), field-programmable gate/logic arrays (FPGAs), or other equivalent integrated or discrete logic circuits. Therefore, the term "processor" as used herein can refer to any of the foregoing structures or any other structure suitable for implementing the techniques described herein. Additionally, in some aspects, the functionality described herein can be provided within dedicated hardware and/or software modules configured for encoding and decoding, or incorporated into combined codecs. Furthermore, the technique can be fully implemented within one or more circuit or logic elements.

While the invention has been described with reference to embodiments, it should be understood that the invention is not limited to the disclosed embodiments. For the avoidance of doubt, the following statements form part of the description. The claims follow the statements and are marked as claims.

Statement 1. A method for controlling an adaptive loop filter for one or more image portions of an image, the method comprising controlling filtering on a first sample based on a plurality of neighboring sample values of a first sample value of the image portion, wherein the control comprises using a nonlinear function having one or more variables based on one or more neighboring sample values and one or more filter variables signaled in an adaptive parameter set.

Statement 2. According to the method described in Statement 1, the nonlinear function includes one or more limiting functions, and the one or more filter variables include one or more limiting parameters determined using information provided in the adaptive parameter set.

Statement 3. The method according to Statement 2, wherein two or more limiting functions share the same limiting parameter determined by information signaled in the adaptive parameter set.

Statement 4. The method described in Statement 2 or Statement 3, wherein an index used to identify the limiting parameter values from a plurality of limiting parameter values is used to determine the limiting parameter to be used with the limiting function, and one of the plurality of limiting parameter values corresponds to the maximum sample value based on bit depth. Appropriately, an index is provided in the Adaptation Parameter Set.

Statement 5. The method according to any one of Statements 2 to 4, wherein an index for identifying the limiting parameter values from a plurality of limiting parameter values is used to determine the limiting parameter used with the limiting function, and the plurality of limiting parameter values include the same number of values for both when the first sample value is a luminance sample value and when the first sample value is a chrominance sample value. Suitably, the plurality of limiting parameter values include four (allowed/available/possible) values. Suitably, the plurality of limiting parameter values include the same number of (allowed/available/possible) values for two or more of one or more image portions of the image. Suitably, the plurality of limiting parameter values include four (allowed/available/possible) values for two or more of one or more image portions of the image. Suitably, the plurality of limiting parameter values include the same number of (allowed/available/possible) values for all image portions of the image or all image portions of an image sequence. Suitably, the plurality of limiting parameter values include four (allowed/available/possible) values for all image portions of the image or all image portions of an image sequence.

Statement 6. According to any of the methods described in the foregoing statements, the number of one or more filter variables notified by signaling is less than the number of one or more variables based on one or more sample values among the neighboring sample values. Properly, the number of one or more filter variables notified by signaling is half the number of one or more variables based on one or more sample values among the neighboring sample values.

Statement 7. According to any of the methods described in the foregoing statements, one or more variables based on one or more neighboring sample values include the difference between the first sample value and each of the one or more neighboring sample values.

Statement 8. According to any of the methods described in the foregoing statements, the output of the nonlinear function is used as an input parameter of the adaptive loop filter. Appropriately, the outputs from two or more nonlinear functions (or limiting functions) are used as input parameters of the adaptive loop filter.

Statement 9. The method according to Statement 8, wherein: the adaptive loop filter uses one or more filter coefficients; and the filter coefficients and their associated filter variables can be determined using indices of neighboring sample values.

Statement 10. According to the method described in Statement 9, the index of a neighboring sample value is associated with its position in the scan order, and the association can be derived using a transposed index. Appropriately, the derived association corresponds to one of the four possible arrangements in Figure 4-b: {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11} for 409; {9, 4, 10, 8, 1, 5, 11, 7, 3, 0, 2, 6} for 410; {0, 3, 2, 1, 8, 7, 6, 5, 4, 9, 10, 11} for 411; or {9, 8, 10, 4, 3, 7, 11, 5, 1, 0, 2, 6} for 412.

Statement 11. According to any of the methods described in the foregoing statements, the nonlinear function comprises one or more limiting functions, and each of the one or more limiting functions returns one of the following: max(-b,min(b,d)), min(b,max(-b,d)), max(c-b,min(c+b,n)), min(c+b,max(c-b,n)), max(-b,min(b,d1+d2)), min(b,max(-b,d1+d2)), max(2*c-b,min(2*c+b,n1+n2)) or min(2*c+b,max(2*c-b,n1+n2)), where c is the first sample value, n or n1 or n2 are the neighboring sample values, d = n-c, d1 = n1-c, d2 = n2-c, and b is the limiting parameter.

Statement 12. A method for processing one or more portions of an image, the image portions having associated chroma samples and luminance samples, wherein the method includes determining, based on information obtained from a bitstream or a first sample value and one or more neighboring sample values, at least one of the following: whether to use a filter controlled by the method according to any of the foregoing descriptions; enabling or disabling the use of the filter; or filter parameters, filter coefficients, or filter variables for use with the filter when filtering the first sample value.

Statement 13. The method according to Statement 12, wherein the information obtained from the bitstream includes a flag provided for one of the luminance or chrominance components, the flag indicating at least one of the following for that component: whether a filter controlled by any one of Statements 1 to 10 is used; or whether the use of said filter is enabled or disabled.

Statement 14. The method according to Statement 12 or 13, wherein the information obtained from the bitstream includes a flag provided for one or more image portions, the flag indicating at least one of the following for one or more image portions: whether a filter controlled by any one of Statements 1 to 10 is used; or whether the use of said filter is enabled or disabled.

Statement 15. A method for encoding one or more images, the method comprising controlling a filter for one or more portions of the image according to any one of Statements 1 to 10, or processing the image according to any one of Statements 12 to 14.

Statement 16. The method according to Statement 15 further includes: receiving an image; encoding the received image and generating a bitstream; and processing the encoded image, wherein the processing includes control according to any one of Statements 1 to 11, or processing according to any one of Statements 12 to 14.

Statement 17. When subordinate to any of Statements 12 to 14, the method according to Statement 15 further includes providing the information in the bit stream.

Statement 18. When subordinate to Statement 11, the method described according to Statement 17 includes: the variables of the nonlinear function further comprising one or more filter coefficients, which depend on the indices of one or more neighboring sample values; and providing said information including providing indices in the bit stream for determining one or more limiting parameters and for deriving the association between the indices of neighboring sample values and their positions in scan order, one or more transposed indices. Appropriately, the derived association corresponds to one of the four possible arrangements in Figure 4-b: {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11} for 409; {9, 4, 10, 8, 1, 5, 11, 7, 3, 0, 2, 6} for 410; {0, 3, 2, 1, 8, 7, 6, 5, 4, 9, 10, 11} for 411; or {9, 8, 10, 4, 3, 7, 11, 5, 1, 0, 2, 6} for 412.

Statement 19. The method according to any one of Statements 16 to 18 further includes: selecting a nonlinear function or one or more limiting functions from a plurality of available functions; using the selected function when processing the coded image; and providing information in the bit stream for identifying the selected function.

Statement 20. A method for decoding one or more images, the method comprising controlling a filter for one or more portions of the image according to any one of Statements 1 to 11, or processing the image according to any one of Statements 12 to 14.

Statement 21. The method according to Statement 20 further includes: receiving a bit stream; decoding information from the received bit stream to obtain an image; and processing the obtained image, wherein the processing includes control according to any one of Statements 1 to 11, or processing according to any one of Statements 12 to 14.

Statement 22. When subordinate to any of statements 12 to 14, the method according to statement 21 further includes obtaining the information from the bit stream.

Statement 23. According to Statement 21, when subordinate to Statement 11...

The method further includes one or more filter coefficients, which depend on the indices of one or more neighboring sample values, as well as one or more transposed indices obtained from the bitstream for determining one or more limiting parameters and for deriving the association between the indices of neighboring sample values and their positions in the scan order. Appropriately, the derived association corresponds to one of the four possible arrangements in Figure 4-b: {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11} for 409; {9, 4, 10, 8, 1, 5, 11, 7, 3, 0, 2, 6} for 410; {0, 3, 2, 1, 8, 7, 6, 5, 4, 9, 10, 11} for 411; or {9, 8, 10, 4, 3, 7, 11, 5, 1, 0, 2, 6} for 412.

Statement 24. The method according to any one of Statements 21 to 23 further includes: obtaining information from the bit stream for identifying a nonlinear function or one or more limiting functions from a plurality of available functions; and using the identified function when processing the obtained image.

Statement 25. An apparatus comprising one or more of the following: a controller configured to perform the method according to any one of statements 1 to 11 or 12 to 14; an encoder configured to perform the method according to any one of statements 15 to 19; or a decoder configured to perform the method according to any one of statements 20 to 24.

Statement 26. A program that, when run on a computer or processor, causes the computer or processor to perform the method according to any one of statements 1 to 11, 12 to 14, 15 to 19, or 20 to 24.

Statement 27. A computer-readable storage medium storing a computer program as described in Statement 26.

Statement 28. A signal carrying an information dataset for encoding an image using the method according to any one of Statements 15 to 19 and represented by a bit stream, the image comprising a set of reconstructable samples, each reconstructable sample having a sample value, wherein the information dataset includes control data for controlling filtering of a first reconstructable sample based on the sample values of neighboring samples of the first reconstructable sample.

Claims (8)

1. A method for encoding one or more portions of an image using adaptive loop filtering, the method comprising: Image data is obtained from a bitstream; The current sample of an image portion is filtered using a nonlinear equation, which includes a limiting function, each having a limiting parameter for limiting the difference between the current sample's neighboring sample values and the current sample's value, wherein the current sample's value is a chroma sample value. Wherein, if the difference is greater than the value indicated by the limiting parameter, the difference is limited to the value indicated by the limiting parameter, and the limiting parameter of each limiting function depends on the position of the neighboring sample value. 2. The method according to claim 1, wherein, The limiting function returns one of max(-b,min(b,d)), min(b,max(-b,d)), max(c-b,min(c+b,n)), and min(c+b,max(c-b,n)), and... c is the first sample value, n is the neighboring sample value, d = n - c, and b is the limiting parameter. 3. A method for decoding one or more portions of an image using adaptive loop filtering, the method comprising: Obtain image data from a bitstream; The current sample of an image portion is filtered using a nonlinear equation, which includes a limiting function, each having a limiting parameter for limiting the difference between the current sample's neighboring sample values and the current sample's value, wherein the current sample's value is a chroma sample value. Wherein, if the difference is greater than the value indicated by the limiting parameter, the difference is limited to the value indicated by the limiting parameter, and the limiting parameter of each limiting function depends on the position of the neighboring sample value. 4. The method according to claim 3, wherein, The limiting function returns one of max(-b,min(b,d)), min(b,max(-b,d)), max(c-b,min(c+b,n)), and min(c+b,max(c-b,n)), and... c is the first sample value, n is the neighboring sample value, d = n - c, and b is the limiting parameter. 5. An apparatus for encoding an image, the apparatus being configured to perform the method according to claim 1 or 2. 6. An apparatus for decoding an image, the apparatus being configured to perform the method according to claim 3 or 4. 7. A computer-readable storage medium storing a computer program that, when run on a computer or processor, causes the computer or processor to perform the method according to claim 1 or 2. 8. A computer-readable storage medium storing a computer program that, when run on a computer or processor, causes the computer or processor to perform the method according to claim 3 or 4.