CN111614960B - Quick coding and decoding method for frame level division according to image content characteristics - Google Patents

Abstract

The invention relates to the technical field of data compression, in particular to a fast encoding and decoding method for frame-level division according to image content characteristics. In the encoding method, by classifying the categories according to the different characteristics of the images at the frame level, the qualified category images can be encoded directly by skipping the string prediction mode, and the overall encoding time can be reduced on the premise of ensuring the same encoding efficiency; , when the image of the category is not encoded in the string prediction mode, the encoding end does not need to write the coding unit level identifier of whether the current coding unit adopts the string prediction mode into the code stream. Correspondingly, there is no need to parse the coding unit during decoding. Level identification, encoding and decoding time can be shortened, and encoding and decoding efficiency can be improved.

Description

A fast coding and decoding method for frame-level division based on image content features

Technical field:

The invention relates to the technical field of data compression, in particular to a fast encoding and decoding method for frame-level division according to image content characteristics.

Background technique:

During the "anti-epidemic" period of the whole world, the remote online interactive activities represented by large-scale telecommuting, online teaching, and remote conferences have played an irreplaceable role and have had a profound impact on the form of social activities. Screen content encoding is the key technology used for screen sharing and display in these remote online interactive activities. These screen contents are ubiquitous and will be one of the main sources of incremental multimedia streaming. As a decisive technology to solve the bottleneck problem of data transmission between "cloud-end/edge", "end/edge-end/edge" in various applications, and "human-computer" interaction lag, the screen content coding technology will be deeply Change the way of "human-human" interaction, "human-machine" interaction, and even "machine-machine" interaction. Therefore, the research on screen content coding technology has become an urgent and inevitable demand. At the same time, with the continuous penetration of mobile devices and smart terminals/edge devices into every corner of social life, the interaction between various devices is becoming more and more important. In addition to the traditional two important factors of image distortion and encoding efficiency, "encoding and decoding time" and "battery endurance", which are directly related to the computational complexity of image encoding and decoding, have increasingly become very important factors. At present, the latest international video coding VVC (Versatile Video Coding) standard has about 25% higher coding efficiency (AI configuration) than HEVC (High Efficiency Video Coding) standard, but the coding complexity is increased by about 26 times. Such a high level of complexity is unbearable for screen content encoding applications on mobile, smart terminals/edge devices. Therefore, it is urgent to explore a screen content encoding technology with "low complexity and high encoding efficiency".

The screen content is regarded as an image that can contain part of the natural image content area and is more general than natural images. Screen content exhibits different characteristics from natural images, mainly including: repeating patterns of various sizes and shapes, relatively small number of colors in coding units in discontinuous tone areas, and several other characteristics (such as sharp, drastic content, no noise, etc.). In the existing hybrid coding framework, the coding blocks of image content with different characteristics are coded using an increasing number of different prediction modes, that is, each coding unit in the image is pre-coded using various prediction modes in turn, and the calculation The rate-distortion value obtained by each mode is obtained, and the mode with the smallest rate-distortion value is selected as the optimal mode of the coding unit for coding.

In the latest VVC standard and my country's AVS3 (the third generation of Audio Video Coding Standard) standard and the previous version or the extended version standard, various prediction modes mainly include intra prediction mode, inter prediction mode, Intra Block Copy (Intra Block). Copy, IBC), Palette mode, string matching (also known as string prediction) mode, etc. For example, in addition to the traditional intra-frame prediction mode, other intra-frame prediction modes appear in the VVC standard. The IBC algorithm uses a limited number of fixed-size (such as 4x8, 8x4, 8x8, 16x4, etc.) block matching within a search range formed by a part of the current frame reconstruction area (current CTU and left CTU). Palette mode uses a Palette to save the representative color of the current CU or the most recent CU. The string prediction mode takes pixels as a unit, searches for an optimal reference string according to a certain matching order and rule, and finally divides the CU into one or more reference strings and/or one or more unmatched pixels. Different prediction modes have irreplaceable roles for different image contents. Generally speaking, intra-frame prediction mode and inter-frame prediction mode are very effective for continuous-tone areas generated by cameras, such as natural images; while IBC algorithm, Palette mode, serial prediction mode are very effective. Predictive mode works well for screen images that contain computer-generated non-contone areas, including text, graphics, icons, charts, and more. In fact, the IBC algorithm and Palette mode are special cases of the string prediction mode.

An image is also called a frame of image. A frame of image is divided into maximum coding units, and the maximum coding unit is divided into one or more coding units, which are determined by the coding tree. Different prediction modes have different effects on image content of different features. The string prediction mode is very effective for screen images with discontinuous areas, but has little effect on content images with few discontinuous areas. In the existing coding algorithms with the burst prediction mode, on the one hand, the precoding method of the burst prediction mode is tried for all image contents, which not only does not help to improve the coding efficiency, but increases the coding time; In the aspect, regardless of whether there is a coding unit in the image that selects the string prediction mode, it is necessary to write the coding unit level identifier of whether the current coding unit adopts the string prediction mode into the code stream. Correspondingly, the decoder also needs to parse the string prediction mode code. Unit level identifier. Therefore, the existing encoding and decoding methods have the problems of long encoding and decoding time and efficiency to be improved. In view of this, this case came into being.

Invention content:

The invention discloses an encoding method and a decoding method respectively. At the image (frame level) level, preset division rules are used to classify image content with different characteristics, and the image content is adaptively carried out according to the divided categories. The encoding and decoding of the non-string prediction mode can improve the encoding and decoding efficiency.

In order to achieve the above-mentioned purpose of the invention, the coding method disclosed by the present invention adopts the following technical solutions:

A fast coding method for frame-level division according to image content features, including:

Step 1: at the image level, use preset division rules to classify the image content features, and divide the image content into category 1 and category 2;

Step 2: According to the image content classification result, set the string prediction mode frame-level identification information of the current image, and write the string prediction mode frame-level identification information into the code stream; if the image is classified into category one, then the string of the image is The prediction mode frame-level flag is set to "true", and if the image is divided into category two, the string prediction mode frame-level flag of the image is set to "false";

Step 3: Determine the setting result of the frame-level flag of the string prediction mode of the current image:

If the frame-level flag of the string prediction mode is set to "false", all coding units in the picture are directly pre-coded in the non-string prediction mode instead of the string prediction mode for precoding, and then the current coding unit is marked with the non-string prediction mode for precoding. And the encoding information of the non-string prediction mode is written into the code stream;

If the frame-level flag of the string prediction mode is set to "true", the following operations are repeatedly performed on each coding unit in the entire image until all coding units are encoded: use the string prediction mode and the non-string prediction mode to perform precoding; Then judge whether the mode with the least rate distortion of the current coding unit is the string prediction mode: if the mode with the least rate distortion of the current coding unit is the string prediction mode, then set the current coding unit string prediction mode flag to "true" and write the code stream, Write the encoding information of the prediction mode of the current coding unit string into the code stream; if the mode with the least rate distortion of the current coding unit is the non-string prediction mode, set the prediction mode flag of the current coding unit string to "false" and write it into the code stream , and write the non-string prediction mode identifier of the current coding unit and the encoding information of the non-string prediction mode into the code stream.

Further, the non-string prediction mode is used in any combination of one or more of the following modes: intra-frame prediction mode; inter-frame prediction mode; IBC mode; Palette mode.

Further, the preset division rule in the step 1 adopts one of the following rules or a combination thereof: (1) The total number of pixels with the same hash value i, Num _hashvalue=i, satisfies the preset condition A and is divided into category one, otherwise It is divided into category two; (2) the total number of pixels of adjacent hash values between Num _hashvalue=i and Num _{hashvalue=i+1} meet the preset condition B, and it is divided into category one, otherwise, it is divided into category two.

Further, the preset condition A is used in any combination of one or more of the following ways: greater than a fixed threshold; equal to a fixed threshold; greater than a non-fixed threshold; equal to a non-fixed threshold; and the preset condition B is used in one or more of the following ways Use in any combination: (1) the absolute value of the difference between the smaller value and the larger value in the total number of pixels of adjacent hash values is greater than a certain fixed threshold; (2) the value in the total number of pixels of adjacent hash values The absolute value of the difference between the smaller and the larger is greater than a certain non-fixed threshold.

Further, when the hash value is calculated, one or more of the following methods are used in any combination: (1) different hash value calculation methods are used for pixels of different sampling formats; (2) pixels of different sampling formats are used The same hash value calculation method; (3) different hash value calculation methods are used for pixels with different bit numbers; (4) the same hash value calculation method is used for pixels with different bit numbers.

Further, the calculation method of the hash value of the pixel adopts one or more of the following methods in any combination: (1) take the hash value formed by concatenating several bits of the same number of each component of the pixel; (2) take the pixel (3) Take the hash value formed by concatenating several bits in the same position of each component of the pixel in series; (4) Take the difference of each component of the pixel A hash value formed by concatenating several bits of the position.

The present invention also discloses a decoding method after data encoding is completed by the above encoding method, and the content includes:

Step 1: Parse the compressed data stream to obtain the frame-level identification information of the string prediction mode of the current image;

Step 2: Determine the setting result of the frame-level flag of the string prediction mode of the current image:

If the frame-level flag of the serial prediction mode is set to "false", parse the compressed data stream, obtain the identification information of the non-sequential prediction mode of the current decoding unit, and use the non-sequential prediction mode to decode the current decoding unit;

If the frame-level flag of the string prediction mode is set to "true", parse the compressed data stream, obtain the coding unit-level flag information of the current decoding unit string prediction mode, and determine whether it is "true" or "false": if the coding unit-level flag information If it is "true", the current decoding unit is decoded using the string prediction mode; if the coding unit level identification information is "false", the compressed data stream is parsed, the current decoding unit non-string prediction mode identifier is obtained, and the current decoding unit is decoded. Units are decoded using a non-string prediction mode.

The encoding and decoding method disclosed in the present invention, by classifying the categories according to different characteristics of the image at the frame level, can directly skip the string prediction mode for encoding the qualified category images. The overall encoding time; in addition, when the image of the category is not encoded in the string prediction mode, the encoder does not need to write the coding unit level identifier of whether the current coding unit adopts the string prediction mode into the code stream. There is no need to parse the coding unit level identifier, the encoding and decoding time can be shortened, and the encoding and decoding efficiency can be improved.

The present invention will be further described below through the accompanying drawings and specific embodiments.

Description of drawings:

Fig. 1 is the flow chart of the coding method adopted in the embodiment of the present invention;

FIG. 2 is a flowchart of a decoding method adopted in an embodiment of the present invention.

Detailed ways:

This embodiment first discloses a fast coding method for frame-level division according to image content characteristics, as shown in FIG. 1 , the content includes:

Step 1: at the image level, use preset division rules to classify the image content features, and divide the image content into category 1 and category 2;

Step 2: According to the image content classification result, set the string prediction mode frame-level identification information of the current image, and write the string prediction mode frame-level identification information into the code stream, and the string prediction mode frame-level identification is represented by "pic_use_sp_flag"; if the image If the image is classified into category one, set the "pic_use_sp_flag" of the image to "true", and if the image is classified into category two, set the "pic_use_sp_flag" of the image to "false";

Step 3: Determine the setting result of the string prediction mode frame level flag "pic_use_sp_flag" of the current image:

If "pic_use_sp_flag" is set to "false", all coding units in the picture are directly pre-coded using the non-string prediction mode instead of the string prediction mode. The encoding information of the prediction mode is written into the code stream;

If "pic_use_sp_flag" is set to "true", repeat the following operations for each coding unit in the entire image until all coding units are coded: use string prediction mode and non-string prediction mode for precoding; then determine the current Whether the mode with the least rate distortion of the coding unit is the string prediction mode: If the mode with the least rate distortion of the current coding unit is the string prediction mode, set the string prediction mode flag cu_sp_flag of the current coding unit to "true" and write it into the code stream. The coding information of the coding unit string prediction mode is also written into the code stream; if the mode with the least rate distortion of the current coding unit is the non-string prediction mode, then the current coding unit string prediction mode flag cu_sp_flag is set to "false" and written into the code stream, Also write the non-string prediction mode identifier of the current coding unit and the encoding information of the non-string prediction mode into the code stream.

The non-string prediction mode can be used in any combination of one or more of the following modes: intra-frame prediction mode; inter-frame prediction mode; IBC mode; Palette mode.

The preset division rule in the above step 1 can be one of the following rules or a combination of them: Rule 1: The total number of pixels with the same hash value i Num _hashvalue=i satisfies the preset condition A to be divided into category 1, otherwise divided into Category 2: Here, the preset condition A is used in one or more of the following ways: greater than the fixed threshold; equal to the fixed threshold; greater than the non-fixed threshold; equal to the non-fixed threshold. Rule 2: The total number of pixels of adjacent hash values between Num _hashvalue=i and Num _{hashvalue=i+1} satisfy the preset condition B, and it is divided into category 1, otherwise it is divided into category 2; here, the preset condition B adopts one of the following methods One or more arbitrary combinations are used: (1) the absolute value of the difference between the smaller value and the larger one in the total number of pixels of adjacent hash values is greater than a certain fixed threshold; (2) the difference between adjacent hash values The absolute value of the difference between the smaller value and the larger value in the total number of pixels is greater than a certain non-fixed threshold.

The hash value calculation method in the preset division rule can be used in one of the following ways or in any combination: (1) different hash value calculation methods are used for pixels of different sampling formats; (2) different sampling formats are used for calculating the hash value. The same hash value calculation method is used for the pixels of different bit numbers; (3) different hash value calculation methods are used for pixels with different bit numbers; (4) the same hash value calculation method is used for pixels with different bit numbers.

The calculation method of the hash value of the pixel can be used in one of the following ways or in any combination: (1) Take the hash value formed by concatenating several bits of the same number of each component of the pixel; (2) Take each pixel of the pixel. Hash value formed by concatenating several bits of different numbers of components; (3) Take the hash value formed by concatenating several bits in the same position of each component of the pixel; (4) Take several bits in different positions of each component of the pixel Concatenated hash values.

For example, when calculating the hash value of a three-component pixel, the total number of bits is set to 12 bits as an example: when choosing the method of taking the same number of bits in the same position of each component and concatenating them to form a hash value, you can Take the middle 4 bits of each component, take a total of 12 bits; or take the lower 4 bits of each component, take a total of 12 bits; choose the same position of each component with different numbers of bits in series to form a hash value. , you can take the middle 6 bits of pixel component 1, the middle 3 bits of component 2, and the middle 3 bits of component 3.

Regarding the preset division rules in the encoding method, a specific embodiment is given below. This embodiment takes a three-component pixel YUV image as an example, and the description is as follows:

1) For each pixel in the image, calculate a 12-bit hash value (range 0 to 4095) in the following way, and take the hash value formed by concatenating the middle 4 bits of the Y, U, and V components respectively. The formula expresses As follows: hashValue=((Y&0x3c<<6)|((U&0x3c)<<2)|((V&0x3c)>>2); ("&" in the formula means binary bitwise AND operation; "0x" means hexadecimal Notation; "<<" and ">>" represent binary left shift and right shift operations, respectively, and "|" represent binary bitwise OR operation).

2) Count the total number of pixels with the same hash value i (0≤i≤4095), denoted as Num _hashvalue=0 , Num _hashvalue=1 ,...Num _{hashvalue=4095} ;

3) Count the total number of pixels with the same hash value that meet the following conditions, and the total number is recorded as RColorNO (initialized to 0):

Traverse Num _hashvalue=0 , Num _hashvalue=1 ,...Num _{hashvalue=4095} ;

If Num _hashvalue=i is greater than or equal to a certain threshold A (such as threshold A is a fixed value of 1000, or a non-fixed value, such as (image height * image width) >> 11), then RColorNO increases by 1;

4) Count the total number of pixels of adjacent hash values between Num _hashvalue=i and Num _{hashvalue=i+1} that satisfy the following conditions, and the total number is recorded as DColorNO (initialized to 0):

Traverse Num _hashvalue=0 , Num _hashvalue=1 ,...Num _{hashvalue=4094} :

Let long_chain=Num _{hashvalue=the larger value between i} and Num _{hashvalue=i+1} ;

Let short_chain=Num _{hashvalue=the smaller value between i} and Num _{hashvalue=i+1} ;

If (((short_chain+1)<<threshold B) is less than long_chain), add 1 to DColorNO; (here, the threshold B is 7)

5) According to the values of RColorNO and DColorNO, the image is divided into two categories:

If (DColorNO/(RColorNO+1)>threshold C), the image is classified as class one, otherwise the image is classified as class two (here, the threshold C is 0.2 for example).

In order to fully illustrate the specific application of the encoding method of the present invention, the following complete examples are provided:

Encoding method embodiment: still take the three-component image as an example to illustrate, the encoding method is as follows:

Step 1: At the image (frame level) level, use the following division rules to divide image content with different characteristics, and divide the image content into two preset categories:

Sub-step 1: For each pixel in the image, calculate a 12-bit hash value (ranging from 0 to 4095) in the following manner, and take the hash value formed by concatenating the middle 4 bits of the Y, U, and V components respectively;

hashValue=((Y&0x3c<<6)|((U&0x3c)<<2)|((V&0x3c)>>2)

Sub-step 2: Count the total number of pixels with the same hash value i (0≤i≤4095), denoted as Num _hashvalue=0 , Num _hashvalue =1,...Num _{hashvalue=4095} ;

Sub-step 3: Count the total number of pixels with the same hash value that meet the following conditions, denoted as RColorNO (initialized to 0):

Traverse Num _hashvalue=0 , Num _hashvalue=1 ,...Num _{hashvalue=4095} :

If Num _hashvalue=i is greater than or equal to the threshold A, then add 1 to RColorNO;

Sub-step 4: Count the total number of pixels of adjacent hash values between Num _hashvalue=i and Num _{hashvalue=i+1} that satisfy the following conditions, denoted as DColorNO (initialized to 0):

Traverse Num _hashvalue=0 , Num _hashvalue=1 ,...Num _{hashvalue=4094} :

Let long_chain=Num _{hashvalue=the larger value between i} and Num _{hashvalue=i+1} ;

Let short_chain=Num _{hashvalue=the smaller value between i} and Num _{hashvalue=i+1} ;

If (((short_chain+1)<<threshold B) is less than long_chain), add 1 to DColorNO;

Sub-step 5: According to the values of RColorNO and DColorNO, the image is divided into two categories:

If (DColorNO/(RColorNO+1)>threshold C), the image is classified as class one; otherwise, the image is classified as class two.

Step 2: According to the image content division result, set the flag bit pic_use_sp_flag to indicate the frame-level flag information of the current image string prediction mode, and write the code stream:

If the image falls into category one, then

pic_use_sp_flag is set to true.

otherwise

pic_use_sp_flag is set to false.

Step 3: According to the frame-level identification information of the string prediction mode, when the frame-level identification of the string prediction mode is false, skip the precoding of the string prediction mode, and do not need to write the encoding unit-level identification of the string prediction mode into the code stream, but directly All coding units in the picture are pre-coded using the non-string prediction mode, and the non-string prediction mode identifier of the current coding unit and the encoding information of the non-string prediction mode are written into the code stream.

For each coding unit in the entire image, repeat the following operations until all coding units are encoded:

If pic_use_sp_flag is true, then

Precoding using a string prediction mode;

Precoding using non-string prediction mode;

Then judge whether the mode with the smallest rate distortion of the current coding unit is the string prediction mode:

If the mode with the least rate distortion of the current coding unit is the string prediction mode, write the current coding unit string prediction mode flag cu_sp_flag as true into the code stream, and write the encoding information of the string prediction mode into the code stream;

If the mode with the least rate distortion of the current coding unit is the non-string prediction mode, write the current coding unit string prediction mode flag cu_sp_flag as false into the code stream, and write the current coding unit non-string prediction mode flag and the encoding information of the non-string prediction mode into the code stream code stream.

In step 3 in the specific embodiment given above, when it is determined that pic_use_sp_flag is true, the precoding sequence may also be: first perform non-string prediction mode precoding, and then perform string prediction mode precoding. Alternatively, the order of precoding may also be as follows: firstly, the partial non-string prediction mode is used for precoding, then the string prediction mode is used for precoding, and then the partial non-string prediction mode is used for precoding. Different precoding order settings have a certain impact on coding efficiency. For example, when one of the non-string prediction modes and the string prediction mode have the same rate-distortion, the current coding unit will select the non-string prediction mode; When the non-string prediction mode precoding has an early termination algorithm, and the conditions for the early termination algorithm are satisfied, subsequent steps may not be performed, which is helpful for improving coding efficiency. However, no matter how the above-mentioned precoding order is set, it is applicable to the coding method provided by the present invention.

Corresponding to the above encoding method, the content of the decoding method includes:

Step 1: parse the compressed data stream to obtain the information of the string prediction mode frame level flag pic_use_sp_flag of the current image;

Step 2: Determine the setting result of the string prediction mode frame-level flag pic_use_sp_flag of the current image:

If pic_use_sp_flag is false, parse the compressed data stream, obtain the non-string prediction mode identification information of the current decoding unit, and use the non-string prediction mode to decode the current decoding unit;

If pic_use_sp_flag is true, parse the compressed data stream, obtain the coding unit-level flag cu_sp_flag information of the current decoding unit string prediction mode, and determine whether it is "true" or "false": if cu_sp_flag is true, use string prediction for the current decoding unit mode for decoding; if cu_sp_flag is false, parse the compressed data stream, obtain the non-string prediction mode flag of the current decoding unit, and use the non-string prediction mode to decode the current decoding unit. Among them, the cu_sp_flag information exists in the data stream in a separate manner (ie, a separate identification bit); it can also be combined with other information, and one of the identification bits is used to represent the cu_sp_flag.

The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Other modifications or equivalent replacements made by those of ordinary skill in the art to the technical solutions of the present invention should be covered as long as they do not depart from the spirit and scope of the technical solutions of the present invention. within the scope of the claims of the present invention.

Claims (7)

1. carry out the fast coding method of frame-level division by image content feature, it is characterized in that: content comprises: Step 1: at the image level, use preset division rules to classify the image content features, and divide the image content into category 1 and category 2; Step 2: According to the image content classification result, set the string prediction mode frame-level identification information of the current image, and write the string prediction mode frame-level identification information into the code stream; if the image is classified into category one, then the string of the image is The prediction mode frame-level flag is set to "true", and if the image is divided into category two, the string prediction mode frame-level flag of the image is set to "false"; Step 3: Determine the setting result of the frame-level flag of the string prediction mode of the current image: If the frame-level flag of the string prediction mode is set to "false", all coding units in the picture are directly pre-coded in the non-string prediction mode instead of the string prediction mode for precoding, and then the current coding unit is marked with the non-string prediction mode for precoding. And the encoding information of the non-string prediction mode is written into the code stream; If the frame-level flag of the string prediction mode is set to "true", the following operations are repeatedly performed on each coding unit in the entire image until all coding units are encoded: use the string prediction mode and the non-string prediction mode to perform precoding; Then judge whether the mode with the least rate distortion of the current coding unit is the string prediction mode: if the mode with the least rate distortion of the current coding unit is the string prediction mode, then set the current coding unit string prediction mode flag to "true" and write the code stream, Write the encoding information of the prediction mode of the current coding unit string into the code stream; if the mode with the least rate distortion of the current coding unit is the non-string prediction mode, set the prediction mode flag of the current coding unit string to "false" and write it into the code stream , and write the non-string prediction mode identifier of the current coding unit and the encoding information of the non-string prediction mode into the code stream. 2. The fast coding method for frame-level division according to image content features according to claim 1, wherein the non-string prediction mode adopts one or more of the following modes in any combination: intra-frame prediction mode; frame Inter prediction mode; IBC mode; Palette mode. 3. The fast coding method for frame-level division by image content feature according to claim 1, wherein the preset division rule in the step 1 adopts one of the following rules or a combination thereof: (1) The total number of pixels with the same hash value i Num _hashvalue=i satisfies the preset condition A is divided into category 1, otherwise it is divided into category 2; (2) The total number of pixels with adjacent hash values Num _hashvalue=i and Num _hashvalue=i+ Between ₁ and 1, if the preset condition is met, B is divided into category 1, otherwise, it is divided into category 2. 4. The fast coding method for frame-level division according to image content characteristics according to claim 3, wherein the preset condition A is used in any combination of one or more of the following methods: greater than a fixed threshold; equal to a fixed Threshold; greater than the non-fixed threshold; equal to the non-fixed threshold; the preset condition B adopts the following method: the absolute value of the difference between the smaller and the larger of the total number of pixels of adjacent hash values is greater than a certain fixed threshold. 5. The fast coding method for frame-level division according to image content features according to claim 3, wherein: when the hash value is calculated, one or more of the following methods are used in any combination: (1) to Pixels with different sampling formats use different hash value calculation methods; (2) use the same hash value calculation method for pixels with different sampling formats; (3) use different hash value calculation methods for pixels with different bit numbers; (4) The same hash value calculation method is used for pixels with different bit numbers. 6. The fast coding method for frame-level division according to image content features according to claim 5, wherein: the hash value calculation method of the pixel adopts one of the following methods or any combination of them: (1) Take the hash value formed by concatenating several bits of the same number of each component of the pixel; (2) Take the hash value formed by concatenating several bits of different numbers of each component of the pixel; (3) Take each component of the pixel (4) Take the hash value formed by concatenating several bits in different positions of each component of the pixel. 7. according to the decoding method after the arbitrary described encoding method of claim 1 to 6 completes data encoding, it is characterized in that: its content comprises: Step 1: Parse the compressed data stream to obtain the frame-level identification information of the string prediction mode of the current image; Step 2: Determine the setting result of the frame-level flag of the string prediction mode of the current image: If the frame-level flag of the serial prediction mode is set to "false", parse the compressed data stream, obtain the identification information of the non-sequential prediction mode of the current decoding unit, and use the non-sequential prediction mode to decode the current decoding unit; If the frame-level flag of the string prediction mode is set to "true", parse the compressed data stream, obtain the coding unit-level flag information of the current decoding unit string prediction mode, and determine whether it is "true" or "false": if the coding unit-level flag information If it is "true", the current decoding unit is decoded using the string prediction mode; if the coding unit level identification information is "false", the compressed data stream is parsed, the current decoding unit non-string prediction mode identifier is obtained, and the current decoding unit is decoded. Units are decoded using a non-string prediction mode.

Images