WO2001045385A1

WO2001045385A1 - Image processing device and method and recording medium

Info

Publication number: WO2001045385A1
Application number: PCT/JP1999/007129
Authority: WO
Inventors: Yasuhiro Tanaka; Tadashige Kojima
Original assignee: Intermedia Inc.
Priority date: 1999-12-17
Filing date: 1999-12-17
Publication date: 2001-06-21
Also published as: JP3530844B2

Abstract

An image processing device comprising a quantizing range calculating unit (33) for determining a range of difference values with occurrence frequencies larger than a specified one, a quantizing unit (35) for quantizing difference values within the range, an entropy encoding unit (37) for reversibly encoding at least quantized difference values and a compressed data creating unit (3c) for creating compressed data including encoded difference value data corresponding to encoded difference values, wherein difference values falling within the quantizing range only are quantized and at least quantized difference values are entropy-encoded and reversibly compressed. Since the above limited quantizing and encoding do not subject difference values outside the quantizing range to quantizing errors, compressed data are available that can restore still images far less in visual degradation than conventional ones while maintaining a comparatively high compression ratio.

Description

TECHNICAL FIELD Image processing method and apparatus and recording medium

The present invention relates to an image processing method and apparatus for processing a still image such as an art image or a medical image having a natural gradation and a recording medium, and particularly to a method for compressing still image data obtained by digitization. zTechnology for restoration. Background art

Various methods have been established for this type of conventional image processing method, and each method is often a hybrid processing using a plurality of methods. This process can be roughly divided into two processes. The process of converting the still image data into a data stream that is easy to compress in consideration of the characteristics of the still image (called a modeling process), and the redundancy of the data sequence. It consists of the process of suppressing and converting to efficient code (called the encoding process).

As a standard image processing method for compression, JPEG (Joint Photographic and Expert Experts Group) can be mentioned. In this method, first, the still image data is divided into blocks composed of a specified number of pixels (generally, 8 × 8 = 64 pixels), and DCT is applied to all pixels for each block. Performs orthogonal transformation by (Discrete Cosine Transform).

Then, after quantizing the DCT coefficients and omitting the information, for each block, the DC component and the AC component are subjected to the event peak coding in order from the lowest frequency. ing. Applying such processing As a result, a high compression ratio can be obtained. However, in the case of such a conventional method, there are the following problems.

In other words, JPEG compresses each block, so when restoring still image data, a phenomenon called block distortion occurs in which distortion occurs at the block boundaries of the still image.For example, continuous contours become zigzag. I will.

In addition, since the high-frequency component of the AC component is roughly quantized, mosquito noise that causes bleeding or graininess in the outline occurs, and overall color shift due to quantization occurs. , visual degradation is very large trough cormorant drawback, there is a force ^s.

Furthermore, when implementing the conventional method, there is a problem that a considerably complicated configuration is required to execute the DCT processing, and the processing load is heavy.

Because of the above-mentioned problems, it is often inappropriate especially for art images and medical images that require high-resolution still images as a top priority after restoration.

Furthermore, since high-resolution still images are increasingly used in recent years, the amount of data tends to be enormous.Therefore, a compression / decompression method capable of high-speed and high-resolution restoration is desired. It is rare.

The present invention has been made in view of such circumstances, and it is possible to obtain an image with little visual deterioration after restoration while maintaining a relatively high compression ratio by limiting the object of quantization. It is an object of the present invention to provide an image processing method and an apparatus thereof and a recording medium capable of performing the method. Disclosure of the invention An image processing method according to the present invention is an image processing method for compressing still image data, which includes dividing a still image data into blocks each having an arbitrary number of pixels, A reference pixel selection process of selecting one pixel as a reference pixel, pixels of each peripheral pixel other than the reference pixel in each block, and reference pixels in a block corresponding to each peripheral pixel A difference value calculation process for calculating the difference value of the values, an occurrence ratio calculation process for calculating the appearance ratio of each difference value, and a difference value range larger than a specific occurrence ratio is obtained from the relationship between the difference value and the appearance ratio. A quantization range calculating step of obtaining a range as a quantization range, a quantization step of quantizing a difference value within the quantization range, a pixel value of a reference pixel, a quantized difference value, and the dividing step The pixel value of the fractional pixel that is out of the block at An entropy encoding process of performing reversible encoding on at least the quantized difference values of the difference values outside the quantization range so as to suppress data redundancy; and The reference pixel data according to the pixel value of the fractional pixel data, the fractional pixel data according to the pixel value of the fractional pixel, the out-of-range pixel data corresponding to the difference value outside the quantization range, and the encoded difference value And generating a compressed data including the encoded difference value data, the decompression data including the quantization table, and the image status of the still image data. is there.

In a general still image having a natural gradation, there is a high probability that the pixel value of an arbitrary pixel in the image and the pixel value of an adjacent pixel take an extremely close value. The higher the value, the more noticeable.

Therefore, if the distribution of values consisting of the difference between the reference pixel in the block and the pixel value of the surrounding pixels is determined for such a still image, for example, the appearance of the difference value as shown in the graph of FIG. The rate will be concentrated around “0”.

Since there is a statistical bias in the appearance rate of the difference value, the original image It is possible to express a still image with a small amount of information, and by calculating such a difference value, it is modeled into a form suitable for processing in the encoding process .

By the way, as apparent from FIG. 4, the appearance rate of the difference value sharply decreases as the value increases. Such a large difference value often corresponds to a contour portion in a still image, but since it has a large effect on the image quality, it should not be quantized.

In other words, a range of difference values larger than a specific appearance rate is obtained from the difference value distribution, and only the difference values within this range are quantized. Therefore, the difference value within this range includes a quantization error. However, this difference value corresponds to a low frequency component and has little effect on image quality.

Next, of the pixel value of the reference pixel, the quantized difference value, the pixel value of the fractional pixel located outside the block, and the difference value outside the quantization range, at least the quantized difference value is selected. The difference value is subjected to end-to-end coding and lossless compression is performed.

In this way, quantization is performed only on the difference values of pixel values within a specific quantization range corresponding to low-frequency components, and lossless coding is performed on other values at least as quantized difference values. Therefore, errors in high-frequency components can be suppressed in the compressed data obtained in this manner.

As described above, according to the method invention according to the present invention, only the difference value within the quantization range among the difference values is quantized, and the pixel value of the reference pixel, the quantized difference value, and the block At least the quantized difference value among the pixel values of the fractional pixels located outside and the difference value outside the quantization range is subjected to the lossy Be-encoding and subjected to lossless compression.

In this way, quantization is limited so that quantization errors are not generated in addition to the difference value within the quantization range. Degradation of Restored Still Image at Restoration Can be obtained.

In addition, since the reference pixels are held over the entire still image, it is possible to reduce the occurrence of overall color shift. Also, since the high-frequency components are held in a completely reversible state, the sharpness of the still image is not easily reduced. Furthermore, since the DCT processing is unnecessary, the processing is relatively simple, so that the processing load can be reduced and the compression processing can be performed at high speed.

Further, in the image processing method according to the present invention, in the above image processing method, in the entropy encoding step, a pixel value of a reference pixel, a pixel value of a fractional pixel, and a difference value outside a quantization range are provided. Among them, those that have a compression effect are also subjected to reversible encoding so as to suppress data redundancy, and in the storing step, data newly encoded in the It is preferable to store the data in place of the original data.

In this way, the data amount can be further reduced by applying reversible entropy coding not only to the quantized difference values but also to those that can provide a compression effect. Therefore, the compression ratio of the compressed data can be further increased.

Further, an image processing method according to the present invention is an image processing method for restoring still image data compressed by the above image processing method, wherein the reference pixel data corresponding to the pixel value of the reference pixel and a fractional pixel Restoration including fractional pixel data according to the pixel value, out-of-range pixel data according to the difference value outside the quantization range, encoded difference value data according to the encoded difference value, and quantization table Reversible data restoring process of restoring reversibly encoded data among compressed data including data for use and image status of still image data before compression, and A temporary difference value is obtained based on the quantized difference value restored on the basis of the quantization table. A temporary pixel value calculating process of adding the pixel value of each peripheral pixel as a temporary pixel value by adding to the pixel value of the reference pixel in the block corresponding to each temporary difference value; A prediction process in which the pixel value of a surrounding pixel is determined as a prediction pixel value so as to suppress a quantization error based on the pixel value of a reference pixel located near each of the surrounding pixels. The static pixel value based on a reference pixel value, a predicted pixel value of a surrounding pixel corresponding to each reference pixel value, a pixel value of each fractional pixel, a difference value of each pixel outside the quantization range, and an image status. And a generation process for generating still image data as restored still image data.

First, all the losslessly encoded data is restored, and a provisional difference value is obtained for the quantized difference value of the data based on the quantization table. Based on the provisional difference value and the pixel value of the corresponding reference pixel, the pixel value of the surrounding pixel is determined as the provisional pixel value. Although this temporary pixel value includes a quantization error, a predicted pixel value is determined so as to suppress the quantization error based on the pixel value of the reference image located near the temporary pixel value.

The restored still image data is obtained based on the predicted pixel value, the reference pixel value, the pixel value of the fractional pixel, the difference value of each pixel outside the quantization range, and the like thus obtained.

Although the restored still image data is obtained as described above, since the pixel values restored based on completely reversible reference pixel data are almost uniformly dispersed throughout the still image, the quantization error is reduced. Can be prevented.

In addition, since high-frequency components can be completely restored without errors, no errors occur in pixel values corresponding to the contours of the image.

Furthermore, since the pixel value of the surrounding pixel including the quantization error is predicted using the reference pixel of the adjacent block, the quantization included in the restored surrounding pixel is predicted. Errors can be suppressed.

According to the method invention as described above, since the pixel values of the completely restored reference pixels are almost uniformly dispersed throughout the still image, the quantization error is prevented from being dispersed throughout the restored still image. And overall color shift can be suppressed.

In addition, since high-frequency components can be completely restored without errors, there is no error in the pixel values corresponding to the contours of the image, and the sharpness of the still image before compression can be faithfully restored compared to the conventional example. it can.

Furthermore, by predicting the pixel value of the surrounding pixels including the quantization error with the reference pixel of the adjacent block, the quantization error of the surrounding pixels can be suppressed. Can be greatly suppressed as compared with the conventional example. In addition, since the inverse DCT processing is not required, the processing is simpler than the conventional example, so that the processing load can be reduced and the restoration processing can be performed at high speed.

Further, in the image processing method according to the present invention, in the image processing method described above, in the predicting step, a reference pixel positioned across each peripheral pixel is used, and a line connecting these pixels is determined according to a quantization width. It is preferable that the provisional pixel value is shifted to be the predicted pixel value of each surrounding pixel.

According to this method invention, since the pixel values of the surrounding pixels are obtained using the reference pixels that do not include the error, the quantization error included in the predicted pixel value of the surrounding pixels can be suppressed. Therefore, the block distortion generated at the block boundary can be suppressed, and the phenomenon that a continuous contour becomes zigzag can be suppressed.

Further, an image processing apparatus according to the present invention is an image processing apparatus for compressing still image data, comprising: a block dividing unit that divides the still image data into blocks having an arbitrary number of pixels; One pixel of the package is the reference pixel A reference pixel selection unit to be selected as a reference pixel, a difference for calculating a pixel value difference between each peripheral pixel other than the reference pixel in each block, and a reference pixel in a block corresponding to each peripheral pixel. A value calculation unit, an appearance rate calculation unit for calculating an appearance rate of each difference value, and a range of difference values larger than a specific appearance rate from a relationship between the difference value and the appearance rate, and this range is set as a quantization range. A quantization range calculation unit to be obtained; a quantization unit that quantizes a difference value within the quantization range; a pixel value of a reference pixel; a quantized difference value; and a block in the block division unit. Lossless encoding is performed so as to suppress data redundancy with respect to the pixel value of the fractional pixel that has become a fraction and the difference value outside the quantization range, which is at least the quantized difference value. The entropy encoding unit to be applied and the reference pixel data corresponding to the pixel value of the reference pixel. Fractional pixel data according to the pixel value of the fractional pixel, out-of-range pixel data according to the difference value outside the quantization range, and encoded difference value data according to the encoded difference value. A compressed data generation unit that generates compressed data including restoration data including a quantization table and image status of the still image data.

In the device configured in this way, the still image data from the original image data storage unit is divided into blocks by the block division unit, and one pixel in each block is used as a reference in the reference pixel selection unit. Select to pixel. Then, the difference calculation unit calculates a difference between the pixel values of the reference pixel and the surrounding pixels to give a statistical bias to the difference value, and the appearance rate calculation unit obtains the appearance rate. From the distribution of the difference values, a range of the difference value larger than a specific appearance rate is obtained by the quantization range calculation unit, and only the difference value within this range is quantized by the quantization unit.

Next, in the entropy coding unit, at least the pixel value of the reference pixel, the quantized difference value, the pixel value of the fractional pixel located outside the block, and the difference value outside the quantization range are calculated. In both cases, the quantized difference value is entropy-coded to perform lossless compression. The compressed data generated in this way is stored in the storage unit, but only the difference between pixel values within a specific quantization range is subjected to quantization, and other values are at least quantized. Since lossless encoding is performed on the difference value, it is possible to suppress errors in high frequency components in the compressed data.

Therefore, according to the apparatus of the present invention, the above-described method invention for compression can be suitably implemented.

Further, an image processing device according to the present invention is an image processing device for restoring still image data compressed by the image processing device, wherein the reference pixel data according to the pixel value of the reference pixel and a block Fractional pixel data according to the pixel value of the pixel that has become a fraction outside the range, pixel data outside the range according to the difference value outside the quantization range, and encoded difference value according to the encoded difference value. Reversible data restoration unit for restoring losslessly encoded data among compressed data including data, restoration data, and the image status of the still image data before compression; A quantized data restoring unit that determines a temporary difference value based on the quantized difference value restored based on the difference value data and the quantization table; Each temporary difference A temporary pixel value calculation unit that adds the pixel value of each peripheral pixel as a temporary pixel value by adding to the pixel value of the reference pixel in the block corresponding to the value, and each of the temporary pixel value calculated by the temporary pixel value calculation unit. A prediction unit that obtains a pixel value of a surrounding pixel as a predicted pixel value so as to suppress a quantization error based on a temporary pixel value of the surrounding pixel and a pixel value of a reference pixel located near each surrounding pixel. , The reference pixel value in each block, the predicted pixel value of the surrounding pixels corresponding to each reference pixel value, the pixel value of each fractional pixel, the difference value of each pixel outside the quantization range, A generating unit that generates the still image data as restored still image data based on the image status.

In the device configured as described above, the compressed data is stored in the compressed data storage unit. Among the compressed data, all of the losslessly encoded data is restored by the lossless data restoration unit, and the quantized difference value is used as a temporary difference value by the quantization data restoration unit. Once obtained, the pixel values of the surrounding pixels are determined as temporary pixel values by the temporary pixel value calculation unit based on the temporary difference value and the pixel value of the corresponding reference pixel. Although this temporary pixel value includes a quantization error, a prediction pixel value is obtained by the prediction unit so as to suppress the quantization error based on the pixel value of the reference image positioned so as to sandwich the temporary pixel value. .

The restored still image data is obtained in the restored data storage unit based on the predicted pixel value thus obtained, the reference pixel value, the pixel value of the fractional pixel, the difference value of each pixel outside the quantization range, and the like.

Although the restored still image data is obtained in this way, since the pixel values restored based on completely reversible reference pixel data are dispersed throughout the still image, quantization errors are dispersed throughout the restored still image. Can be prevented. Also, since high-frequency components can be completely restored, no error occurs in pixel values corresponding to the outline of the image.

Furthermore, since the pixel value of the surrounding pixel including the quantization error is predicted using the reference pixel of the adjacent block, the quantization error included in the restored surrounding pixel can be suppressed.

Therefore, according to the apparatus of the present invention, the above-described method invention for restoration can be suitably implemented. BRIEF DESCRIPTION OF THE FIGURES

FIG. 1 is a block diagram showing a schematic configuration of the image processing apparatus, and FIG. 2 is a block diagram functionally showing the image processing apparatus related to compression.

Fig. 3 is a schematic diagram showing the relationship between the reference pixel and surrounding pixels in the block. And

FIG. 4 is a distribution diagram showing an example of a difference value appearance frequency.

FIG. 5 is a diagram schematically showing a state of a block after quantization, and FIG. 6 is a flowchart showing a compression process.

FIG. 7 is a flowchart showing the compression process.

FIG. 8 is a flowchart showing an entropy encoding process, and FIG. 9 is a block diagram functionally showing an image processing apparatus relating to restoration.

FIG. 10 is a diagram provided for explanation of the prediction process.

FIG. 11 is a diagram provided for explanation of the prediction process.

FIG. 12 is a flowchart showing the restoration process.

FIG. 13 is a flowchart showing the restoration processing. BEST MODE FOR CARRYING OUT THE INVENTION

The following are forms for solving the conventional problems.

FIG. 1 is a block diagram showing a hardware configuration of the image processing apparatus according to the present invention.

The apparatus of this embodiment is configured by a computer system, and a CPU 1 executes a process of compressing still image data and a process of restoring compressed still image data according to a processing program stored in an internal memory 3. Run. CPU 1 and internal memory 3 are connected via bus line 5.

The internal memory 3 has a pre-processing storage section 3b for storing data before compression processing and decompression processing and data being processed, in addition to a program storage section 3a for storing processing programs, and a compression processing section. It has a post-processing storage unit 3c for storing data after and after restoration processing. The CPU 1 also includes, via an input / output interface (IZO) 9, a recording medium driver 15 into which the recording medium 13 is loaded, an external storage device 17 that stores various data and files, and a still image It is also connected to a display device 19 such as a monitor display used to display data as images, an input device 21 such as a keyboard or mouse, and an image input device 23 such as a scanner. I have.

The processing program read from the recording medium 13 loaded in the recording medium driver 15 is transferred and stored in the program storage section 3a, and the CPU 1 executes the processing described below. It has become so. Examples of the recording medium 13 include a floppy disk, a magneto-optical disk, a CD-ROM, and a memory card.

In the following description, each process according to the processing program is also functionally shown in a block diagram for easy understanding.

The compression processing by the image processing apparatus configured as described above will be described with reference to the functional block diagram of FIG.

In this device, first, an image file containing still image data to be processed is input from the image input device 23 or the external storage device 27 and stored in the pre-processing storage portion 3b.

The still image data referred to here is generally in a reversible image format such as BMP (bitmap) or TIFF (tiff), but it is a specified process that converts the data into a format that can be processed as described later. The format of the still image data input from the image input device 23 does not matter as long as the procedure is performed.

The pre-processing storage unit 3b stores the image status of the image file including information such as image size, color format (RGB, CMYK, black and white), and resolution. A work area necessary for processing is secured based on the task, and image information based on still image data is developed in this area.

In the block division unit 25, for example, as shown in the schematic diagram of FIG. 3, an arbitrary pixel group (horizontal jX vertical k pixels) based on the image information of the pre-processing storage unit 3b is processed. The work of dividing into minute blocks BL (m × n pixels, for example, 3 × 3 pixels) is performed.

Depending on the size of the still image data, a fractional pixel that is not taken as the block B L may occur, but the pixel value of such a fractional pixel PA is stored in the work area as it is.

For the image information divided for each block BL, the reference pixel selection unit 27 selects one reference pixel (for example, the central pixel P 5) for each block BL. I do. From the viewpoint of processing speed, it is preferable that the reference pixel is common to each block B L, but a different reference pixel may be selected for each block B.

After the reference pixel is selected for each block BL, the difference value calculation unit 29 calculates the difference between the pixel value of the reference pixel in each block BL and the pixel values of surrounding pixels located around it. Ask. As a result, statistical bias can be imparted to a general still image, and the efficiency of subsequent quantization and coding can be increased.

After calculating the difference between the pixel values of the reference pixel and the surrounding pixels in all the blocks BL, the appearance rate calculator 31 calculates the appearance rate of the difference value.

The quantization range calculation unit 33 determines a range to be quantized using the difference value appearance rate. That is, instead of quantizing all the difference values as in the conventional example, the quantization is selectively performed only in a certain range where the appearance rate is high.

By the way, in a general still image with natural gradation, the pixel value of a certain pixel, There is a high probability that the pixel value of a pixel adjacent thereto takes a value close to the pixel value. Therefore, when the distribution of the difference values is obtained, for example, as shown in the graph of FIG. 4, the appearance rate of the difference values is concentrated in a range R 2 near “0”.

Therefore, except for the ranges R1 and R3, which correspond to the contours of the image and have a low appearance rate corresponding to the high-frequency component, a considerable compression rate can be obtained simply by quantizing only the constant range R2 having a high appearance rate. Can be expected.

The quantization unit 35 actually performs quantization on each difference value for the quantization range determined in this way.

In the entropy encoding unit 37, the pixel value of the fractional pixel PA, the pixel value of the reference pixel P5, and the quantized value corresponding to the low frequency component If the compression effect is obtained, the difference between the pixel value of the fractional pixel, the pixel value of the reference pixel, and the difference value out of the quantization range is obtained. The information is compressed, and the quantized difference value is always encoded to compress the amount of information.

By the above processing, for example, as shown in FIG. 5, in the blocks BL (1) to BL (3), all the surrounding pixels except for the central reference pixel P5 indicated by hatching are quantized. On the other hand, in the block BL (j + 1), the surrounding pixels at the lower right are non-quantized and invertible. As is clear from the above-mentioned appearance rate, most of the pixel group is quantized and made reversible.

Then, the compressed still image data is stored in the post-processing storage unit 3c as compressed data together with the decompression data including the quantization table required for decompression. In addition, output to external storage device 17 if necessary

The post-processing storage unit 3c corresponds to the compressed data generation unit in the present invention.

Next, the compression process is performed with reference to the flowcharts in FIGS. 6 to 8. Will be described step by step.

[Step S 1]

The file containing the still image data input from the image input device 23 or the external storage device 17 is opened, and necessary data and the like are stored in the pre-processing storage section 3b.

[Step S 2]

Based on the image status included in this file, a queryer required for processing is secured in the pre-processing storage unit 3b.

[Step S 3]

Based on the above file, the image information of the still image data is developed in the work area of the pre-processing storage unit 3b.

[Steps S4 to S6]

The image information developed in the work area is divided into small blocks BL, for example, 3 × 3 pixels as shown in FIG. In this case, the image information may not be divisible by 3, but for those fractional pixels PA, their pixel values ν _{ι> Λ} are stored in the work area as they are.

Step S4 described above corresponds to the dividing process in the present invention.

[Step S 7]

The image information divided for each block B L is selected for each block B L, for example, the central pixel Ρ5 is set as a reference pixel. It is preferable that the reference pixel # 5 be at the same position in each block in consideration of later processing. The pixel value of the reference pixel # 5 thus selected is stored in the work area.

Step S7 corresponds to a reference pixel selection step in the present invention. [Step S 8]

The pixel value V y of the reference pixel P 5 in each block B L and surrounding pixels P 1 to P 4,? 6 to? The difference values dif (1) to dif (8) from the pixel values 1 to 4 and V6 to 9 of 9 are obtained. The difference values dif (1) to dif (8) thus obtained are overwritten with the image information of the work area to save resources. If there are sufficient resources, the difference values dif (l) to dif (8) may be stored in another area.

This step S8 corresponds to a difference value calculating step in the present invention.

[Step S 9]

The processes in steps S7 and S8 described above are performed for all the blocks BL, and when the process is completed, the process proceeds to the next step S10.

[Step S10, S11]

The appearance rate is calculated for the obtained difference value dif.

For example, if the still image has an 8-bit gradation, the difference between the pixel values can range from 255 to 255 (± 2 to the eighth power). As shown in Fig. 4, a histogram is created by plotting the difference value diί between 255 and 255 on the horizontal axis and the number of pixels corresponding to the vertical axis. More specifically, the number of occurrences of each difference value may be divided by the total number of pixels, but the number of occurrences may simply be used as the appearance rate.

In this case, this calculation includes an error for the number of fractional pixels Ρ 、, but since it accounts for a very small percentage of the total number of pixels, ignoring it has no effect.

Note that these steps S10 and S11 correspond to the appearance rate calculation process of the present invention. [Step S 1 2]

Here, the range to be quantized is determined from the appearance rate obtained as described above.

For example, as shown in FIG. 4, a range R2 in which the total appearance frequency accounts for 95% or more of the total number of pixels is obtained, and this range R2 is set as a quantization range. In this way, quantization is performed only for the range R2 having a high appearance rate.

Next, an arbitrary quantization width is set for a difference value existing in the quantization range R2, and a quantization code is assigned to generate a quantization table. The quantization width may be set at equal intervals, but for a difference value with a low frequency of appearance, the quantization width may be set slightly larger to generate a quantization table that performs coarse quantization. Les ,.

This step corresponds to the quantization range calculation process in the present invention.

[Steps S13 to S16]

Using the quantization table generated in the above steps, all the difference values dif are quantized to generate quantized difference values Qdif.

At this time, only the difference value dif within the quantization range R2 is subject to quantization. As for the difference value dif outside the quantization range R2 (that is, the difference value dif within the range R1, R3), the difference value dif without being quantized is stored in the pre-processing storage unit 3b. You. At this time, the image information in the work area is overwritten to save resources, but if there is enough resources, it may be stored in another area.

These steps S13 to S16 correspond to the quantization process in the present invention.

"Step S 17" The data generated through the above-described steps includes a pixel value V _{K HP} of the reference pixel P 5, a quantized difference value Q dif, a pixel value ν _{ι) Λ of} the fractional pixel PA, and a value outside the quantization range. The entropy coding for suppressing the redundancy of the information amount is performed on these difference values dif.

Specifically, as described below, general bottom-up Huffman coding is used.

The following steps described with reference to FIG. 8 correspond to the entropy coding process in the present invention.

[[Step S20]]

The pixel value V _{li EF} of the reference pixel P 5, the quantized difference value Q dif, the pixel value V of the fractional pixel PA, and the difference value dif outside the quantization range are treated as symbols in Huffman coding. Sorting is performed according to the appearance rate corresponding to the symbol.

[[Step S2 1]]

Assign a code (eg, “1” to one and “0” to the other) for the least significant symbol and the second least significant symbol.

[[Step S22, S23]]

The appearance rate of the least significant symbol and the second least significant symbol are merged and newly defined as a merged symbol. Then, the above steps are repeatedly performed until the number of symbols becomes “1”.

[[Step S2 4]]

After the loop consisting of the above steps is completed, the Huffman table is created by reading back the accumulated assignment codes.

[[Step S25]]

The above processing is performed for each symbol, and is repeated until a Huffman table is created for each symbol. [[Step S26]]

After a Huffman table is created for each symbol, it is determined for each symbol whether or not the compression effect of entropy coding can be obtained based on its own Huffman table, and processing is performed based on this determination. Branch. This determination can be made easily based on the frequency of appearance and the code length of the Huffman table.

[[Step S28, S29]]

Among the symbols of the pixel value V _{li l; l of} the reference pixel P 5, the pixel value V _M of the fractional pixel PA, and the difference value dif out of the quantization range, the symbol determined to have the compression effect is represented by the corresponding Huffman table. If it is determined that there is no compression effect by performing encoding with reference to, the actual value as it is is used.

In general, the higher the resolution of an image, the greater the entropy (amount of information), and the effect is not so much obtained in the case of Huffman coding as a typical example. Is very small as a percentage of the total number of pixels, so even if the actual value is used when it is determined that there is no compression effect, there is no significant effect on the overall compression ratio .

In this way, not only the quantized difference value Q di 〖but also a compression effect can be obtained by applying reversible end-port Bee coding to further reduce the data amount. Can be. Therefore, the compression ratio of the compressed data can be further increased.

[[Step S29]]

Here, only the quantized difference value Qdif is encoded with reference to the Huffman table, and stored in the post-processing storage unit 3c. By this step S29, at least the quantized difference value Q d i 〖is subjected to reversible coding with suppressed data redundancy.

Now, return to the flowchart of FIG. [Step S 18]

Reference pixel data corresponding to the pixel value of the reference pixel P5 stored in the work area of the pre-processing storage unit 3b, fractional pixel data corresponding to the pixel value ν _{ι> Λ} of the fractional pixel PA, and post-processing storage The encoded difference value data corresponding to the encoded difference value Q dif stored in the unit 3c, the restoration data including the quantization table and the Huffman table used for quantization, and a still image The compressed data including the image status of the data is stored in the post-processing storage unit 3c.

This step corresponds to the generation process in the present invention.

[Step S 19]

The compressed data generated as described above is output to the external storage unit 17 and stored as a file.

Thus, only the difference value within the quantization range among the difference values is quantized, and the pixel value of the reference pixel, the quantized difference value, and the pixel value of the fractional pixel located outside the block are calculated. In addition, among the difference values outside the quantization range, at least quantized difference values are subjected to loss-to-loss compression by performing edge-to-peak coding. In this way, the quantization and encoding targets are limited so that quantization errors are not generated in addition to the difference values within the quantization range. It is possible to obtain compressed data that can suppress the visual degradation of the restored still image at the time of restoration as compared with the example.

In addition, since the reference pixels are held over the entire still image, overall color shift is unlikely to occur, and since the high-frequency components are held in a completely reversible state, the sharpness of the still image is reduced. Nikure, Furthermore, since the DCT processing is unnecessary, the processing is relatively simple, so that the processing load can be reduced and the compression processing can be performed at high speed.

In the entropy coding in step S17 described above, Huffman coding is used as an example, but the present invention is not limited to this. For example, the following method may be used.

(1) Block coding In Huffman coding, coding was performed for each symbol alone, but depending on the number of symbols and the conditions of their occurrence probabilities, the information sources to be processed are combined into an arbitrary block. The average code length can be shortened by performing coding.

In particular, if the entropy is considered to be small even if a certain amount of continuous blocks are generated (such as high-resolution images) (the probability of occurrence of data is biased to a particular symbol to a certain extent), it is appropriate. In some cases, the compression rate can be further improved by considering the blocks with a large number of blocks as one symbol and performing the Huffman coding described above.

(2) Arithmetic coding In this method, each symbol is represented as a real number segment between real numbers 0 and 1. Similar to Huffman coding, when performing arithmetic coding, it is necessary to know the symbol occurrence probability in advance.

(3) Universal encoding This is the encoding used in most file archivers and compression utilities these days.

In this coding, unlike the algorithms described above, it is not necessary to know the symbol occurrence probability in advance. Although there is a drawback that small files may not be effective enough, when dealing with large files as in the present invention, the average code length becomes longer as the data length becomes longer. It is considered to be an effective encoding means because it converges on the speech.

For this encoding, a given symbol sequence stores several symbols before and after the symbol currently being encoded in a buffer, and uses this buffer as a dictionary to perform encoding using a slide dictionary. Law or symbol column There is a dynamic dictionary method that dynamically creates a dictionary when encoding, and uses the index numbers of the dictionary to represent previously appearing symbols (columns).

It is also possible to change the encoding method depending on the type of data such as the pixel value of the reference pixel and the quantized difference value, or to perform the entropy encoding by combining a plurality of the above encoding methods. You can do it.

In the above description, in addition to the quantized difference value, the pixel value of the reference pixel, the pixel value of the fractional pixel, and whether or not the difference value outside the quantization range has a compression effect are determined. If it is determined that there is a compression effect, entropy coding is applied to that as well. However, it is also possible to omit the decision flow and apply the event-port coding only to the quantized difference value.

Next, with reference to a functional block diagram of FIG. 9, a description will be given of a process when the compressed data generated by the above-described compression process> is restored by the image processing apparatus configured as described above.

The external storage device 17 stores a file including the compressed data generated by the above-described compression processing, and the compressed data of this file is provided to the pre-processing storage unit 3b.

Real data decoding unit 4 corresponding to the reversible data restoration unit in the present invention 4

1 is for the data included in the compressed data that has been subjected to lossless encoding, and refers to the encoding table (Huffman table) used for encoding. Restore.

The data includes the reference pixel data corresponding to the pixel value V, _.cl . Of the reference pixel P5 and the fraction pixel data corresponding to the pixel value ν ,, _Λ of the fractional pixel PA out of the block. Data, out-of-range pixel data corresponding to the difference value dif outside the quantization range, encoded difference value data corresponding to the encoded difference value Q dif, and the quantization table used during the compression processing. For example, there are data for restoration including an image and an encoding table, and an image status including information such as an image size, a color format, and a resolution.

The quantized data decoding unit 43 restores only the quantized irreversible data. Therefore, only the quantized difference value data decoded from the encoded difference value Qdif is targeted.

At this time, the quantization table is referred to, but the temporary difference value is temporarily determined by temporarily using the center value of the quantization width as a temporary restoration value.

Note that the quantized data decoding unit 43 corresponds to the quantized data restoration unit in the present invention.

The data adding unit 45 adds the temporary difference value calculated as described above and the pixel value V ^. Of the reference pixel P5 in the corresponding block to obtain the pixel value of each surrounding pixel. Is calculated as a temporary pixel value.

The data adding section 45 corresponds to the provisional pixel value calculating section in the present invention.

The restored data prediction unit 47 performs prediction to achieve visual smoothness and corrects the data. Specifically, the temporary pixel values of the surrounding pixels between the reference pixel P5 are corrected using the pixel value V ^ of the reference pixel P5 and a linear function.

FIG. 10 is a plot of pixel values in a horizontal direction at a position including the reference pixel P5 of the blocks BL (1) to BL (3) shown in FIG. 5 for a general image. FIG. 9 is a diagram showing an example of a case where the user has turned on the camera.

In such a general image, the reference pixel P5 in each of the blocks BL (1) to BL (3) and the quantized peripheral pixel P located between the reference pixels P5 are used. For the pixel values of P4 and P6, the pixel values of the pixels quantized by the general method before quantization are the pixel values of each reference pixel P5 and the pixel values of surrounding pixels P4 and P6. It is reasonable to assume that it is likely to exist in a smooth curved line connecting.

In other words, the primary function connecting the reference pixel P5 of the block BL (1) and the reference pixel P5 of the block BL (2) adjacent to the block BL (1) is their If the distance change in the horizontal direction is dX, the change in pixel value is dy, and the pixel value of the reference pixel P5 of the block BL (1) is ya, any point X between them The predicted value y is expressed by the following equation (1) (shown by a dotted line in FIG. 11).

y = (d y / d x) x + ya, 1)

This prediction value y is the temporary pixel value of the peripheral solid pixels obtained by the data adding section 4 5 described above, is shifted within the range of the quantization width Q _w in the direction of the prediction value y, and the shift The value is adopted as the predicted pixel value.

Specifically, the temporary pixel value of the peripheral pixel P5 located at the block BL (1) is V, and the temporary pixel value of the peripheral pixel P4 located at the block BL (1) is V ′. When, respectively shift in the range of quantization width Q _w to the predicted value y side represented by the above formula (1), each predicted pixel value V _{l) fi,} V ,,, (figure Black circle).

As described above, since the pixel values of the surrounding pixels P 6 and P 4 are obtained by using the reference pixels P 5 that do not include the quantization error, the quantization values included in the predicted pixel values V _B and V of the surrounding pixels P 6 and P 4 are obtained. The conversion error can be suppressed. Therefore, it is possible to suppress the block distortion that occurs at the block boundary, which has been a problem in the conventional example, and to suppress the phenomenon that a continuous contour line becomes zigzag. 47 corresponds to the prediction unit in the present invention. The pixel value of the reference pixel in each block restored in this way, the predicted pixel value of the surrounding pixel value, the pixel value of the fractional pixel, the difference value outside the quantization range, and the image status are stored after processing. Stored in part 3c. Then, it is output to the external storage device 17 or the display device 19 if necessary.

The post-processing storage unit 3c corresponds to the generation unit in the present invention.

Next, the restoration processing will be described with reference to the flowcharts of FIGS.

[Step S 30]

The file including the compressed data input from the external storage device 17 is opened and stored in the pre-processing storage unit 3b.

[Step S31]

Based on the image status included in this file, there are four types of data: reference pixel data according to the pixel value of the reference pixel, encoded difference value data according to the encoded difference value, and out of quantization range. Acquires out-of-range pixel data according to the difference value and fractional pixel data according to the fractional pixel value, and acquires the encoding table (Huffman table in this example) and the quantization table required for restoration. I do.

Further, a query area of a size necessary for the restoration processing according to the image status is secured in the pre-processing storage unit 3b. Then, if there is data that has not been encoded or quantized, such as the pixel value of the reference pixel, it is output as it is into the query.

[Steps S32 to S35]

Four types of data: Of the reference pixel data, the encoded difference value data, the out-of-range difference value data, and the fractional pixel data, data that has been reversibly encoded is restored. If lossless encoding has been performed, restoration is performed with reference to the Huffman table, and the restored data is placed in the work area of the pre-processing storage unit 3b. These processes are repeatedly executed until all the data subjected to the lossless encoding is restored. From the rules at the time of the compression process, at least the encoded difference value data is restored to obtain a quantized difference value.

The above steps correspond to the reversible data restoration process in the present invention.

[Steps S36 to S39]

Next, restoration processing is performed on the quantized difference value, which is the difference value within the quantization range determined to be irreversible by quantization.

Specifically, using the quantization table used at the time of the compression processing, the quantized difference value is restored as a temporary difference value. Then, each difference value is added to the pixel value of the reference pixel in the corresponding block, and each pixel value is determined as a temporary pixel value for all surrounding pixels of all blocks. The temporary pixel value obtained in this way is mapped to the work area.

Note that the above steps S36 to S39 correspond to the provisional pixel value calculation process in the present invention.

[Step S40]

Here, the temporary pixel value (including the quantization error) of each surrounding pixel obtained as described above is predicted to suppress the quantization error based on this and the temporary pixel value of each surrounding pixel. Perform processing.

Reference is now made to FIG.

[[Step S43]]

The block adjacent to the block containing the surrounding pixels to be predicted is extracted, and the reference pixel is extracted. [[Step S44]]

From these positional relationships, the X-coordinates of surrounding pixels that are prediction points are extracted.

[[Step S 4 5]]

As described when calculating the above equation (1), the slope dy / dx of the line containing the surrounding pixel to be predicted and the straight line passing between the reference pixels of the adjacent block is calculated as follows. Ask.

[[Step S46]]

Since the above equation (1) is obtained in the above step S45, the predicted value y of the prediction point is calculated based on the equation (1).

[[Step S47]]

The temporary pixel value of each surrounding pixel is shifted to the predicted value y obtained in the above step within the range of the quantization width, and the pixel value is adopted as the predicted pixel value. As a result, the quantization error can be suppressed, and the block distortion generated at the block boundary can be suppressed.

[[Step S48, S49]]

After processing all surrounding pixels in a block, the above steps are repeatedly performed until all blocks are processed.

Return to FIG.

[Step S 4 1]

Data group that has been expanded as is or restored in the work area: reference pixel values in each block, predicted pixel values of surrounding pixels corresponding to each reference pixel value, and pixel values of fractional pixels Then, an image status is added to the difference value of each pixel outside the quantization range, and restored still image data is generated so that the image processing device can take the image data. The restored still image data is stored in the post-processing storage unit 3c.

[Step S42] The restored still image data is output to the external storage device 17 and saved as a file.

In this way, the still image data compressed by the above <compression process> is restored as the restored still image data, but the pixel values of the completely restored reference pixels are almost the same over the entire still image. Since they are uniformly distributed, it is possible to prevent the quantization error from being dispersed throughout the restored still image, and to suppress the overall color shift.

Furthermore, by predicting the pixel value of the surrounding pixels including the quantization error with the reference pixel of an adjacent block, the quantization error of the surrounding pixels can be suppressed. It can be greatly reduced compared to the example.

Further, since the inverse DCT processing is not required, the processing is simpler than in the conventional example, so that the processing load can be reduced and the restoration processing can be performed at high speed.

In the above description, the prediction processing of step S40 is collectively performed after obtaining all temporary pixel values in steps S36 to S39, but in order to increase the processing efficiency, step S3 is performed. After step 7, the prediction processing of step S40 may be performed sequentially. However, in this case, the loop of step S49 in FIG. 13 is unnecessary because there is a no-rape in step S39 in FIG. This eliminates the need to read the mapped data prior to the prediction processing and perform the mapping again after the prediction processing, thereby improving the efficiency.

In the above description, when calculating the predicted pixel value, the predicted pixel value is expressed by equation (1). Although a linear function is used, the following various functions may be used.

(1) Agrange polynomial

The m-first degree polynomial connecting the m points is given by Lagrange's formula

That is, in the interval [0, m] where the start point is 0 and the end point is m, P „(X)) = f j j = 0, 1,…,, m

The polynomial that satisfies is

P (x) = ∑w. (X) f.

(X-χ ₀ノ (-Xj_, (χ-χ _{ί +} !)) (Χ-χ J

_{_{(x; -.. 0;}} ■ j- j ^ (χ; -. x j + 1) .., it is given by (Xj-.

X j indicates the position of the reference pixel, and ί ”. Indicates the pixel value of the reference pixel. At this time, the pixel value of an arbitrary surrounding pixel X is compared with the solution obtained by the above polynomial, and a value closer to the quantization width is corrected as a restoration value.

(2) Newton difference quotient

To find the polynomial, a simpler Newton difference quotient formula below may be used.

f) = f (x _Q ) + U-,)) f (x «, x,) + (x-x.) -x,) f (X (,» X ,, x ₂ ) ⁺

+ (xx ₀ ) (x-,) (-x _m- ,) f (X ().

+ (xx ₀ ) (xx.) (xx _m ) f (x, x _(l , x ,, x „,)

Note that f (x,., X _{i + 1} ) in the equation is

f (x _{i + 1} ) -f ( _Xi )

f (x ;, Xw i + r) X l. + 'l— x l.

It is defined as the difference quotient represented by As is clear from this equation, the difference Quotient is the slope of a line connecting the data points _(Xi, y ,.) and a _{_{(x i + 1 y i +}} l).

Note that the higher-order difference quotient f (Xi, x _{i + 1} x _{i + r + 1} ) is

f (x _{i + 1} , x _{i + 2}・ ••, x _{i + r + 1} ) -f (x _i , x _{i + i} >'· _{ί ί + Γ} )

Are defined sequentially.

When using the above polynomial, it is necessary to pay attention to the selected number of reference pixels used in the equation. In other words, if the number of selections is too large, the amount of computation for solving higher-order polynomials will increase at an accelerating rate.

In addition, the function called “analysis function” is guaranteed to be the only smoothest function in the interval [x ₍₎ xj (that is, the m-first derivative is a continuous function). May vibrate at both ends of the section (runge phenomenon), which may cause a problem at the end of the section. For these reasons, it is not preferable to increase the number of data points.

(3) Correction by n-th spline function

In the above-mentioned method of sliding a polynomial (this is called an interval polynomial), it is clear that smoothness is not guaranteed in a portion between sliding adjacent sections (hereinafter referred to as a “node”). (Ie, the first derivative in the clause is discontinuous).

For this problem, a function that guarantees the contact condition in the nodes between the sections of each polynomial is the spline function. There are various ways to find this spline function, but the following method is often used simply and conveniently.

The second derivative f, where f is the cubic spline function in the small section [X j, is

X. X χ-χ.

f. "(x) = a. —— ^ + σ '

j + ij + 1 σ x '' h "= x _{j +1} -xj

二 (積分) = Φ y _{j + 1} + Φ Yj + h (( ³ -φ) σ _{j + 1} + (φ ^ +) σ.) Χ- χ.

φ = φ ^ Ι-φ

h.

Is obtained.

This shows the conditions of the spline function (interpolation conditions, 0, second-order contact conditions).

Is the first derivative of f

f. (x) = A. + h. ((3 ³ -l) σ _{j + 1-} (3 (^ ² -l) σ)

yj +厂^y j

If continuity is guaranteed in Δ, that is, f 'j- ^ Xj): f' j (Xj) is guaranteed.

Thus, in the end, it is sufficient to solve the following n−1 system of simultaneous equations with σ ”, j = l, ……, N−1 as unknowns.

2 (h ₀ + h〗) σ _t + h! Σ ₂ = Δ, -Δ ₀

hj_, σ _j _, + 2 (h _j _ ₁ + h _j ) σ j + hj σ _{j + 1} = Δ Δ j_, (j = l, '', Ν-1) h _N — ₂ σ _Ν _ ₂ + 2 (h _N — ₂ + —,) σ _Ν —, = Δ Δ _Ν — ₂

Industrial applicability

As described above, the image processing method, the apparatus, and the recording medium according to the present invention can be used for the apparatus for compressing / decompressing still image data and the software in general. Digital image processing devices that perform processing such as compression and decompression, shooting devices such as digital cameras and digital video cameras for obtaining still image data, and computers Image processing board, computer image processing software, and floppy disk, CD-ROM, MO, etc. that store these image processing software. Suitable for medium.

Claims

The scope of the claims

1. An image processing method for compressing still image data,

A dividing process of dividing the still image data into blocks having an arbitrary number of pixels;

A reference pixel selection process in which one pixel in each block is selected as a reference pixel;

A difference value calculation process for calculating a difference value between each peripheral pixel other than the reference pixel in each block and a reference pixel in the block corresponding to each peripheral pixel, and an appearance for calculating an appearance rate of each difference value Rate calculation process,

A quantization range calculating step of obtaining a range of the difference value larger than a specific appearance rate from the relationship between the difference value and the appearance rate, and obtaining this range as a quantization range;

A quantization process for quantizing the difference value within the quantization range,

At least quantization among the pixel value of the reference pixel, the quantized difference value, the pixel value of a fractional pixel that has fallen out of the block in the division process, and the difference value outside the quantization range An end-to-peak coding process for applying reversible coding to the already-processed difference value so as to suppress data redundancy;

Reference pixel data corresponding to the pixel value of the reference pixel, fractional pixel data corresponding to the pixel value of the fractional pixel, out-of-range pixel data corresponding to the difference value outside the quantization range, and encoded difference value Generating compressed data including encoded difference value data according to the following, decompression data including a quantization table, and an image status of the still image data;

An image processing method comprising:

2. In the image processing method according to claim 1,

In the entropy encoding process, the pixel value of the reference pixel and the fractional pixel Lossless encoding is performed to suppress data redundancy even for pixel values and difference values outside the quantization range that have a compression effect.

The image processing method, wherein in the storing step, data newly coded in the entropy coding step is stored in place of original data.

3. An image processing method for restoring still image data compressed by the image processing method according to claim 1,

Reference pixel data corresponding to the pixel value of the reference pixel, fractional pixel data corresponding to the pixel value of the fractional pixel, out-of-range pixel data corresponding to the difference value outside the quantization range, and encoded difference value Of compressed data including the encoded differential value data according to the above, decompression data including a quantization table, and the image status of the still image data before compression, Reversible data restoration process to restore,

A provisional difference value is obtained based on the quantized difference value restored based on the encoded difference value data and a quantization table, and each provisional difference value is stored in a block corresponding to each provisional difference value. A temporary pixel value calculating step of adding the pixel value of each surrounding pixel as a temporary pixel value by adding to the pixel value of the reference pixel of

Prediction that obtains the pixel value of a surrounding pixel as a predicted pixel value so as to suppress a quantization error based on the temporary pixel value of each surrounding pixel and the pixel value of a reference pixel located near each surrounding pixel. Process

The reference pixel value in each block, the predicted pixel value of surrounding pixels corresponding to each reference pixel value, the pixel value of each fractional pixel, the difference value of each pixel outside the quantization range, and the image status Generating the still image data as restored still image data based on

An image processing method comprising:

4. The image processing method according to claim 3, In the above-mentioned prediction process, a temporary pixel value is shifted according to the quantization width on a line connecting the reference pixels located between the respective neighboring pixels to obtain a predicted pixel value of each neighboring pixel. An image processing method comprising:

5. An image processing apparatus for compressing still image data,

A block dividing unit for dividing the still image data into blocks each having an arbitrary number of pixels;

A reference pixel selection unit for selecting one pixel in each block as a reference pixel;

A difference value calculation unit that calculates a difference value between each peripheral pixel other than the reference pixel in each block and a reference pixel in the block corresponding to each peripheral pixel, and an appearance rate of each difference value An appearance rate calculation unit for calculating

A quantization range calculating unit that obtains a range of the difference value larger than a specific appearance rate from a relationship between the difference value and the appearance rate, and obtains this range as a quantization range;

A quantization unit that quantizes a difference value within a quantization range,

At least one of the pixel value of the reference pixel, the quantized difference value, the pixel value of the fractional pixel that has become a fraction outside the block in the block division unit, and the difference value outside the quantization range An end-to-end coding unit that performs reversible coding so as to suppress data redundancy with respect to the quantized difference value, reference pixel data corresponding to the pixel value of the reference pixel, and fractional pixels Fractional pixel data according to the pixel value of, the pixel data outside the range according to the difference value outside the quantization range, the encoded difference value data according to the encoded difference value, and the quantization table A compressed data generation unit that generates compressed data including decompression data including the following, and image status of the still image data:

An image processing apparatus comprising:

6. The image processing apparatus according to claim 5,

The entropy encoding unit includes a pixel value of a reference pixel and a pixel of a fractional pixel. Lossless coding is performed to suppress data redundancy for values and difference values outside the quantization range that have a compression effect.

The image processing apparatus according to claim 1, wherein the storage unit stores the data newly coded by the event-to-peak coding unit instead of the original data.

7. An image processing device for restoring still image data compressed by the image processing device according to claim 5,

Reference pixel data corresponding to the pixel value of the reference pixel, fractional pixel data corresponding to the pixel value of the pixel that has become a fraction outside the block, and pixels outside the range corresponding to the difference value outside the quantization range Of the compressed data including the data, the encoded difference value data corresponding to the encoded difference value, the decompression data, and the image status of the still image data before compression, the lossless encoding is performed. A reversible data restoration unit for restoring the applied data;

A quantized data restoring unit for obtaining a temporary difference value based on the quantized difference value restored based on the encoded difference value data and a quantization table; A temporary pixel value calculation unit that adds the difference value to a pixel value of a reference pixel in a block corresponding to each temporary difference value and obtains a pixel value of each surrounding pixel as a temporary pixel value;

Based on the provisional pixel value of each surrounding pixel obtained by the provisional pixel value calculation unit and the pixel value of a reference pixel located in the vicinity of each surrounding pixel, the surrounding pixels are suppressed so as to suppress the quantization error. A prediction unit for determining a pixel value of a pixel as a predicted pixel value;

The reference pixel value in each block, the predicted pixel value of surrounding pixels corresponding to each reference pixel value, the pixel value of each fractional pixel, the difference value of each pixel outside the quantization range, and the image status A generating unit that generates the still image data as restored still image data based on

An image processing apparatus comprising:

8. The image processing apparatus according to claim 7,

The prediction unit shifts the provisional pixel value according to the quantization width on the line connecting the reference pixels using the reference pixels positioned across each of the surrounding pixels to obtain the predicted pixel value of each of the surrounding pixels. An image processing apparatus characterized by the above-mentioned.

9. A recording medium storing an image processing program for compressing still image data,

A reference pixel selection process for selecting one pixel in each block as a reference pixel;

A difference value calculation process of calculating a difference value between each peripheral pixel other than the reference pixel in each block and a reference pixel in the block corresponding to each peripheral pixel;

An appearance rate calculation process for finding an appearance rate of each difference value;

From the relationship between the difference value and the appearance rate, a range of difference values larger than a specific appearance rate is obtained, and this range is used as a quantization range. Quantization processing,

At least one of the pixel value of the reference pixel, the quantized difference value, the pixel value of a fractional pixel that has become a fraction outside the block in the division processing, and the difference value outside the quantization range An edge-to-peak coding process for performing reversible coding on the already-processed difference value so as to suppress data redundancy;

Reference pixel data corresponding to the pixel value of the reference pixel, fractional pixel data corresponding to the pixel value of the fractional pixel, out-of-range pixel data corresponding to the difference value outside the quantization range, and encoded difference value A generation process of generating compressed data including encoded difference value data corresponding to the above, decompression data including a quantization table, and an image status of the still image data; Recording medium for storing a program for controlling a computer so as to execute the program.

10. A recording medium storing an image processing program for restoring still image data compressed by the program stored in the recording medium according to claim 9.

Reference pixel data corresponding to the pixel value of the reference pixel, fractional pixel data corresponding to the pixel value of the fractional pixel, out-of-range pixel data corresponding to the difference value outside the quantization range, and encoded difference value Of compressed data including the encoded differential value data according to the above, decompression data including a quantization table, and the image status of the still image data before compression, Reversible data restoration processing to restore,

A provisional difference value is obtained based on the quantized difference value restored based on the encoded difference value data and the quantization table, and each provisional difference value is defined as a reference in a block corresponding to each provisional difference value. A temporary pixel value calculation process in which the pixel value of each peripheral pixel is obtained as a temporary pixel value by adding to the pixel value of the pixel;

Prediction in which the pixel value of a surrounding pixel is determined as a predicted pixel value so as to suppress a quantization error based on the temporary pixel value of each surrounding pixel and the pixel value of a reference pixel located near each surrounding pixel. Processing,

Based on the reference pixel value in each block, the predicted pixel value of surrounding pixels corresponding to each reference pixel value, the pixel value of each fractional pixel, the difference value of each pixel outside the quantization range, and the image status Generating the still image data as restored still image data by using

Recording medium for storing a program for controlling a computer so as to execute the program.