JP2000101847A - Image signal coding method, image signal decoding method and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To shorten a maximum code length by finding out a region shared by codes that are not almost actually in use in the case of coding an actual image and using this region for an escape region so as to reduce number of codes thereby decreasing number of bits to mutually identify the codes. SOLUTION: A zero coefficient of a quantization coefficient AC component received by a zero discrimination section 11 is fed to a run counter 12 and a non-zero coefficient is fed to a size discrimination section 13, and when the run counter 12 counts a consecutive number of zero coefficients and the size discrimination section 13 detects the non-zero coefficient, run information of the run counter 12 is fed to a run comparison section 14, which outputs a logical value 1 when the run information is equal to a prescribed threshold r-limit or over. When the non-zero coefficient amplitude is classified and grouped by its quantity into group numbers and a group number that is an output of the section 13 is equal to a prescribed threshold s-limit or over, a size comparison section 15 provides an output of logical value 1. The outputs of the comparison sections 14, 15 are fed to an AND element 16, where they are ANDed and an output of the AND element 16 is fed to a code assignment section 17. The section 17 outputs a code word denoting an escape region when the output of the AND element 16 has a logical value 1, and a code word corresponding to the input run information and the group number and an added bit when the AND output has a logical value 0.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for encoding and decoding an image signal, and more particularly to a two-dimensional variable-length encoding method and apparatus, and a two-dimensional variable-length decoding method.
Related to the device. More specifically, the present invention relates to a method and an apparatus for performing encoding and decoding of an image signal that enable the code length of image data to be reduced.

[0002]

2. Description of the Related Art With the spread of computers, digital image signals have been widely used. Since such a digital image signal has a large amount of data, a compression method for reducing the data amount by suppressing redundancy for communication and storage is generally applied.

[0003] There are various algorithms for the compression method. Among them, discrete cosine transform (Discrete Cosine Tra), which is one of orthogonal transform algorithms, is used.
The nsform (hereinafter referred to as DCT) method has been used in a wide range of fields because it has been adopted as an international standard algorithm for coding still images and moving images.

In such an orthogonal transformation algorithm, first, an image is divided into pixel blocks which are rectangular regions of N pixels × N pixels (N is a positive integer). An orthogonal transform is performed for each of the divided pixel blocks, and N × N transform coefficients are obtained.
These transform coefficients are divided by a quantization step width predetermined for each coefficient position and rounded to obtain an N × N quantized coefficient in which the number of bits of the transform coefficient is reduced. Further, as for the quantization coefficient, the entire data amount is reduced by variable length coding in which a short codeword is assigned to a code with a high appearance probability and a long codeword is assigned to a code with a low appearance probability.

[0005] The variable length coding of the algorithm adopted in the standard method uses two-dimensional coding as described below from the viewpoint of coding efficiency.

FIG. 11 shows N = 8, that is, an 8 × 8 transform coefficient matrix and a zigzag scan order. The element at the upper left is called a DC component, and the other elements are called AC components. The AC component is two-dimensionally encoded by the following procedure.

For the AC component, the quantization coefficient matrix is scanned in a zigzag order as shown by an arrow in FIG. 11, and a run having a continuous length of a coefficient having a value of zero is obtained. Further, it is determined whether the amplitude value of the non-zero coefficient that occurs after the run belongs to any of a plurality of groups classified according to the magnitude of the value, and the size that is the group number is determined. A corresponding codeword is read from a two-dimensional Huffman table set in advance for these combinations of run and size, and a value indicating the amplitude value in the group is output as a value of a non-zero quantization coefficient together with an additional bit. I do.

By the way, JPEG, which is a coding standard for still images, and H.264, which is a coding standard for moving images, are used. 261, M
In PEG, the two-dimensional Huffman code table has the following differences.

FIG. 12 shows a two-dimensional Huffman table (code table) recommended for an AC component of a luminance signal in the JPEG algorithm. FIG.
The array of numerical values in the middle represents the length (number of bits) of the Huffman code word assigned to the combination of run and size. In the JPEG two-dimensional Huffman code table, codes corresponding to all combinations of runs and sizes are prepared. In the case of the baseline system which is the basic algorithm, up to 15 runs and up to 10 sizes are used. However, as shown in FIG. 12, in the JPEG Huffman table, in the column of size 0, the range of runs from 1 to 14 is an unused area. The code at the position of size 0, run 15 is the ZRL code used most recently when a run of length 16 has occurred, and the code at the position of size 0, run 0 is that all the remaining coefficients in the block are This is an EOB code indicating that it is zero. When these special codes are included, 162 code words are allocated.

The array of numerical values shown in FIG. 12 is the length of the Huffman code word assigned to the combination of run and size, that is, the number of bits, and is short in a portion where the frequency of each run and size combination is high. A bit number is assigned, and a long bit number is assigned to a portion where the occurrence frequency is low. In the example shown in FIG. 2, a 16-bit code is assigned as the maximum code length. It is presumed that these combinations of runs and sizes of the areas to which the 16-bit codes are assigned have a low occurrence frequency.

FIG. 13 shows H.264, a video coding standard for videoconferencing. 261 is a two-dimensional Huffman table. H. At 261, the orthogonally transformed transform coefficient is replaced by a quantization index that is an integer in the range of −127 to 127 by a quantization process. These are consecutive lengths (runs) of zero as in the case of JPEG.
And the following combination of the magnitude (level) of the non-zero value. In addition, H. In 261, codewords are allocated to combinations of runs 0 to 63 and absolute values of levels 1 to 127. In FIG. 13, when a numerical value is entered at a position corresponding to a combination of a run and a level, it indicates that a codeword having the bit number of the numerical value is assigned. Other combinations of run and level are coded using escape codes because they are infrequently occurring events. In this case, each of the escape code and the run has a fixed length of 6 bits, and the level has a fixed length of 8 bits, for a total length of 20 bits. The number of codewords is 66 in total.

[0012]

In the field of copiers, the shift to digitalization is rapidly progressing for the purpose of realizing high image quality by advanced image quality control and multifunction such as sharing with printers and faxes. Images handled by such digital copiers and printers have been increasing in capacity as demands for higher image quality and higher speed have increased.

For example, at present, about 16
In the future, the resolution of dots (400 [dots / 25.4 mm]) will increase to about 24 dots (600
[Dot / 25.4 mm]) is expected. The number of pixels at this time is based on the ISO, A4 original (21
0 mm x 297 mm), 210 x 24 x 297 x
24 = 5040 × 7128 = 35,925,129 pixels. The data amount per pixel is monochrome 2
In the case of an image with 56 gradations, one pixel is one byte, and YM
In the case of a full-color image composed of four color components of CK, each pixel has 4 bytes.

Further, it is required that these images be subjected to processing such as printing at a speed exceeding, for example, 40 sheets per minute. With the increasing demand for such high-speed processing of large-capacity image data, an encoding technique has been required for the purpose of reducing the image data storage capacity and the bandwidth at the time of data transfer.

In particular, in a copying machine or a printer having an image forming unit such as xerography, it is necessary to supply an image signal at a high speed corresponding to a high-speed image output. For this reason, the speed at which code data is decoded is an important factor influencing productivity, and a high-speed decoding capability has been required in order to increase productivity.

Images (still images) handled by such digital copiers and printers are compared with moving images from the viewpoint of large-volume image data continuous in the time axis direction.
There are the following features and restrictions. (1) Since the decoded image is printed and watched for a long time, there is a high demand for the image quality of the result obtained as an output, and the compression ratio, which is a factor of deteriorating the image quality, cannot be so high. (2) In a multi-page document read from a scanner or the like, there is no correlation between consecutive pages. For this reason, it is not possible to use inter-frame prediction in which only a difference between consecutive frames, such as a moving image, is encoded to suppress redundancy.

Images (still images) handled by the printer have the above-mentioned restrictions, and the number of input pixels is large, the image quality is required to be high, and the encoding efficiency cannot be so high. For this reason, the performance required for the decoding process becomes severe with the increase in the capacity of the image data.

In a decoding process with many restrictions, JP
When the EG algorithm is applied, it is conceivable that the decoding device is configured by an LSI. In the case of the JPEG algorithm, the LSI has a configuration including a variable length decoding unit, an inverse quantization unit, and an inverse orthogonal transform unit. Among them, processes such as inverse orthogonal transform and inverse quantization can be relatively easily speeded up by using an LSI, but the following problems are involved in implementing a variable-length decoding unit into an LSI.

That is, the variable-length decoding unit determines the input variable-length codeword and associates it with a specific combination of run and level. For this association process, the Huffman table of the JPEG algorithm is
It is possible to implement (a) a configuration that performs matching with all codes registered in the table, (b) a configuration in which logic circuits are combined in multiple stages, and the like. Was not satisfied.

Therefore, the configuration of the present invention solves the above problem by reducing the number of code words set in the two-dimensional Huffman table and shortening the maximum code length.

For example, the H.264 described in the section of the prior art has been described. 26
It is conceivable to employ a two-dimensional Huffman coding table into which escape codes are introduced, as in the moving image coding standard such as 1. That is, as shown in FIG. 13, a code word having a bit number corresponding to a numerical value is assigned to a position corresponding to a combination of a run and a size with a high frequency of occurrence. The present invention attempts to apply a technique similar to the moving picture coding method to be coded by using a still picture coding system.

However, H. In the algorithm of the moving image coding standard such as H.261, the input is originally a difference image between frames, and further different from the still image coding algorithm such as a difference in quantization processing, a difference in a combination of a run and a level, and the like. There are many points. For this reason, H. It is difficult to apply the Huffman code table of H.261 to a still image coding system as it is. H. The escape area of the encoding standard of H.261 (moving image) has a complicated shape as shown in FIG. 13, and there is a problem that the determination of escape / non-escape becomes complicated.

On the other hand, in the JPEG algorithm, a recommended table for two-dimensional Huffman coding is prepared. This table is a general-purpose table that shows an average good coding performance for various images, although the coding efficiency is inferior to the Huffman table optimized for each image. It has been known.

In view of the above, the configuration of the present invention is
Based on the PEG algorithm, (1) it can be easily determined whether or not an escape code is used; (2)
To take advantage of the general-purpose performance of the recommended Huffman table of JPEG, (3) to reduce the number of coded symbols by introducing escape codes, and (4) to reduce the maximum code length. The purpose is to solve.

[0025]

In order to solve the above-mentioned problems, an image signal encoding method according to the present invention provides a method of quantizing a transform coefficient obtained by performing an orthogonal transform on an image signal. An image signal code to be encoded by combining a run having a continuous length of zero, a size indicating the size of a non-zero quantization coefficient, and additional information for representing a value of the non-zero quantization coefficient. In the quantization method, it is determined whether or not a combination of a run and a size of a quantization coefficient is included in an escape area defined by a combination of a run and a size that is equal to or larger than a predetermined run threshold and equal to or larger than a predetermined size threshold. If it is determined that the combination of the run and the size of the quantization coefficient is included in the escape area, an escape code indicating that the area is the escape area, a value of the run, and a non-zero When the combination of the run and the size of the quantization coefficient based on the image signal is determined not to be included in the escape area, the coding is performed in combination with the combination of the run and the size. The encoding is performed by combining a predetermined code word and additional information indicating a value of a non-zero quantization coefficient.

Further, in the image signal encoding method according to the present invention, the encoding is performed by using a code table in which code words corresponding to a plurality of combinations of runs and sizes are encoded. , Are shown as rectangular areas in the code table.

Further, the image signal encoding method according to the present invention comprises:
The code word corresponding to the combination of run and size included in the escape area is deleted from the code table, and the code table that indicates the escape area is added to the code table is corrected.

Further, in the image signal encoding method of the present invention, the codewords set in the code table are set with identification bits for identifying codewords having the same code length, and the escape area is set. The number of identification bits is reduced and set according to the number of code words deleted from the code table.

Further, in the image signal encoding method according to the present invention, the escape area is set as a combination area of a run and a size of a quantization coefficient having a low frequency of occurrence.

Further, in the image signal encoding method of the present invention, when it is determined that the combination of the run and the size of the quantized coefficient is included in the escape area, an escape code indicating that the combination is an escape area is provided. , And the value of the run and the value of the non-zero quantized coefficient are each assigned to a fixed code length region, and are combined and encoded.

Further, in the image signal encoding method of the present invention, the length of the code word assigned to the escape code is determined to be included in the escape region from the maximum code length set for the quantization coefficient outside the escape region. And a code length longer than, or at least equal to, the code length obtained by subtracting each fixed code length indicating the value of the run and the value of the non-zero quantized coefficient.

Further, the image signal encoding method of the present invention is characterized in that an escape area having a run threshold value of 4 and a size threshold value of 3 is set, and the maximum code length is reduced by 2 bits.

Further, the image signal encoding apparatus of the present invention
The orthogonal transformation is performed on the image signal, and the quantization coefficient obtained by quantizing the obtained transformation coefficient is determined to be zero by the zero determination means for determining that the quantization coefficient is zero and the zero determination means. Zero counting means for counting runs indicating the continuous length of the quantized coefficient, non-zero coefficient determining means for determining a size indicating the size of the quantized coefficient determined to be non-zero by the zero determining means, In an image signal encoding apparatus including a code allocating means for encoding a code word corresponding to a combination of sizes and additional information representing a value of a non-zero quantization coefficient, a run for comparing a run with a preset run threshold value A comparison unit, a size comparison unit that compares the size with a preset size threshold, and a run determined by the run comparison unit to be equal to or larger than the run threshold, and the size is equal to or larger than the size threshold by the size comparison unit. An escape information determining unit that determines that the determined case is a combination of a run and a size included in the escape area, and a code word allocating unit that switches a code allocating operation corresponding to an output of the escape information determining unit. In the escape information determination means, when it is determined that the combination of run and size is included in the escape area, by combining the escape code indicating the escape area, the value of the run, and the value of the non-zero quantization coefficient, If the escape information determination means determines that the combination of run and size is not included in the escape area, the predetermined codeword corresponding to the combination of run and size and a non-zero quantization count And code word allocating means for encoding the additional information representing the value of.

Further, in the image signal encoding apparatus according to the present invention, the escape information judging means is constituted by a logical operation element for outputting a logical product of comparison results of the run comparing means and the size comparing means.

Further, in the image signal coding apparatus according to the present invention, the code word allocating means has a code table for obtaining a predetermined code word corresponding to a combination of a run and a size. As described above, the area equal to or larger than the size threshold is set as an escape area, the corresponding code word is not provided in the escape area, and the code table is a table having an escape code indicating that the area is an escape area. And

Further, the image signal encoding apparatus of the present invention
An escape area with a run threshold of 4 and a size threshold of 3 is set, and the maximum code length is reduced by 2 bits.

Further, according to the image signal decoding method of the present invention, code data is input, bit data of a predetermined length is read from the input code data, and a variable length code portion located at the head of the read bit data is read. After decoding, information indicating whether the area is an escape area, the length of an additional bit,
Run, and code conversion to output the entire length of the variable length code and the information associated with it, information indicating whether or not it is an escape area, additional bit length, run, and variable length code and Based on the overall length of the accompanying information,
It is characterized by executing quantization coefficient reconstruction for reading information from bit data and restoring quantization coefficients.

Further, the image signal decoding apparatus of the present invention has a storage means for inputting code data and temporarily storing input code data, a bit cutout means for reading bit data of a predetermined length from the storage means, Decodes the variable-length code portion located at the beginning of the code, and outputs information indicating whether or not the area is an exceptional area, an additional bit length, a run, and the entire length of the variable-length code and the accompanying information. The information is read from the bit data based on the code conversion means and information indicating whether or not the area is an exception area, the additional bit length, the run, and the entire length of the variable length code and the accompanying information. Reconstructing means for restoring the quantization coefficient.

[0039]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention will be described below. (1) Concept of the configuration of the present invention (2) Outline of coding apparatus and decoding apparatus of the present invention (3) Details of coding method of the present invention (3. 1) Setting of escape area, setting parameters and number of reduced codes (3.2) Method of shortening maximum code length (4) Details will be sequentially described in accordance with the operation items of the encoding device and the decoding device of the present invention.

(1) Concept of Configuration of the Present Invention First, the concept of the configuration of the present invention will be described. The present invention sets an escape area in a two-dimensional Huffman coding table used in a still image coding algorithm, facilitates determination of an escape area / non-escaping, and makes use of general-purpose performance of a recommended Huffman table of JPEG. , The number of coded symbols and the maximum code length can be reduced.

FIG. 1 shows an example in which an escape area is set in a two-dimensional Huffman table. The Huffman table shown in FIG. 1 has the same configuration as the table shown in FIG. 12, but differs in that an escape area is set. The hatched area in FIG. 1 is an escape area. The hatched area in FIG. 1 is an area whose size is larger than s_limit (size threshold) and whose run is larger than r_limit (run threshold). In the encoding of the present invention, this area is treated as an escape area. The details of the s_limit and r_limit settings will be described later.

(2) Overview of the Encoding and Decoding Devices of the Present Invention FIG. 2 shows an embodiment of a two-dimensional Huffman encoding device of the present invention. The two-dimensional Huffman coding apparatus of FIG. 2 includes a zero coefficient discriminator 11 for discriminating whether the value of an AC component of an input quantized coefficient is zero or non-zero, and counts the number of consecutive zero coefficients. The run counter 12 determines which of the groups classified according to the magnitude of the non-zero coefficient belongs to a group number, which is set, and which of the numerical values in the group the non-zero coefficient corresponds to. A size determination unit 13 that outputs an additional bit to indicate the run information, a run comparison unit 14 that compares input run information with a predetermined threshold r_limit, a size comparison unit 15 that compares an input group number with a predetermined threshold s_limit, and a run comparison unit An AND element 16 for outputting the logical product of the output of the size comparison unit 15 and the output of the size comparison unit 15. The input run information, group number, additional bit, and non-zero coefficient are converted according to escape information. Constituted by the code assignment unit 17 for Huffman coding.

The operation of the two-dimensional Huffman coding apparatus of the present invention shown in FIG. 2 will be described in detail after the description of the coding method using the escape area of the present invention.

FIG. 3 shows an embodiment of a two-dimensional Huffman decoding apparatus according to the present invention. The two-dimensional Huffman decoding device shown in FIG. 3 includes a buffer 21 for temporarily storing input code data, a bit cutout unit 22 for reading code data from the buffer 21 with a variable bit length, and a bit string cut out by the bit cutout unit. As an input, a pattern that matches a predetermined plurality of bit patterns is detected, and additional bit lengths, run information, the sum of the matched bit string lengths and the additional bit lengths allocated to the plurality of bit patterns, and escape information are detected. And a reconstruction unit 24 that reads out additional bits and run information from the corresponding bit field from the bit string extracted by the bit extraction unit 22, and restores the quantized coefficient value.

The operation of the two-dimensional Huffman decoding apparatus of the present invention shown in FIG. 3 will also be described in detail after the description of the encoding method using the escape area of the present invention.

(3) Details of Encoding Method of the Present Invention (3.1) Setting of Escape Area, Setting Parameters and Number of Reduced Codes FIG. 4 shows the compression when a plurality of images are encoded using the JPEG algorithm. It shows the difference in the occurrence distribution of the combination of run and size due to the difference in the rate. In FIG. 4, the horizontal axis indicates size (the size increases from left to right), and the vertical axis indicates a run (the run length increases downward).
Is set, and the hatched area is
The portion where the combination of run and size has occurred even once is shown. The leftmost diagram in FIG. 4 shows a code generation area when the compression ratio (scaling factor) is small, and the middle is a middle compression ratio and the right end is a large compression ratio.

The encoding uses a recommended quantization table for a luminance component defined in the JPEG algorithm. The compression ratio was changed by changing the scaling factor. The scaling factor is a positive real value (not including 0) multiplied by the entire quantization step width to uniformly increase or decrease the quantization step width for dividing each transform coefficient in the quantization process. When the scaling factor is larger than 1, the compression ratio increases (the amount of code decreases),
When the scaling factor is smaller than 1, the compression ratio becomes smaller.

FIG. 4 shows that the distribution in which the combination of run and size occurs changes by changing the compression ratio. When the compression ratio is low, the size appears to be relatively large, but long runs hardly occur. On the other hand, when the compression ratio increases, a large size does not often occur, and conversely, a long run tends to occur.

However, it can be seen that in any of the cases where the compression ratio is small, medium, and large, almost no occurrence occurs in the region of the combination where the run and the size are both large.

On the other hand, in the two-dimensional Huffman code table of FIG. 12, when either the run or the size is large, a 16-bit code having the maximum code length is uniformly allocated. This is due to the fact that JPEG limits the maximum code length to 16 bits, and the frequency of each of the combinations of runs and sizes included in these areas to which 16-bit codes are assigned is not equal. Unthinkable.

Therefore, based on the result of the code generation region shown in FIG. 4, it is determined that the region having both the large run and the large size is the region having a low occurrence frequency among the regions to which the 16-bit code is allocated shown in FIG. . Therefore, in the present invention, these areas are set as escape areas.

FIG. 1 shows an escape area set in a JPEG two-dimensional Huffman table. FIG.
In the two-dimensional Huffman table, the run is r_limit by using two parameters r_limit (run threshold) and s_limit (size threshold).
As described above, the rectangular area whose size is equal to or larger than s_limit is set as the escape area. By setting such an escape area, the escape area and the other area can be distinguished by simply comparing the run and the size with each threshold value. In addition, the escape area set in this way covers a combination having a large run and a large size, which hardly occurs in an actual image.

When the escape area is set in the two-dimensional Huffman table by the above procedure, the number of codes can be reduced. The Huffman table shown in FIG. 1 is similar to the Huffman table shown in FIG.
Is a table having areas of 0 to 11 and run (r) of 0 to 15. As described above, the code at the position of the size 0 and the run 15 is the ZRL code used at the time when the run of the length 16 occurs, and the code at the position of the size 0 and the run 0 is EOB code indicating that all remaining coefficients are zero. When these special codes are included, 162 code words are normally allocated.
In the configuration of the present invention, as shown as a hatched area in FIG. 1, a region exceeding r_limit (run threshold) and s_limit (size threshold) is set as an escape region, and the code of this escape region is reduced.

R_limit (run threshold) and s_lim
By setting a region exceeding it (size threshold) as an escape region, the sign of the hatched region shown in FIG. 1 is reduced. As a result, the number of reduced codes is
The following equation: Number of reduced codes = (11−s_limit) × (16
−r_limit).

The total number of codes left in the two-dimensional Huffman table after code reduction by setting the escape area is: total code number = 162 + 1− (11−s_limit) ×
(16-r_limit).

The reason why 1 is added to the original total code number 162 is that an escape code indicating an escape area is added.

Whether or not an area is included in the escape area can be easily identified by r_limit (run threshold) and s_limit (size threshold). When it is determined whether or not the quantization coefficient is included in the escape area, the quantization coefficient is subjected to different encoding inside and outside the escape area.

FIG. 5 shows the difference in the structure of the code word assigned to the combination of run and size inside and outside the escape area.

As shown in the upper part of FIG. 5, with respect to the DC component, Huffman code 2 to 9 bits and additional bits 0 to 11 are allocated, and a maximum of 20 bits are output.

For a combination of a run and a size determined that the run is not larger than r_limit (run threshold) and the size is not within the range larger than s_limit (size threshold) and is determined to be outside the escape area, FIG. As shown in the middle part of FIG.
A 6-bit Huffman code and 1 to 10 additional bits are output, resulting in a maximum bit output of 26 bits. The components outside the DC component and the escape area are the same as the conventional bit output.

A combination of a run and a size whose run is larger than r_limit (run threshold) and whose size is larger than s_limit (size threshold) and which is determined to be within the escape area is shown in the lower part of FIG. After encoding an escape code of length n bits,
4 bits for runs up to ~ 15, -1023 ~ 10
23 quantization coefficients are represented by 11 bits, and a total of (n + 15)
Output a bit. The code length n assigned to the escape code will be described later.

Hereinafter, a description will be given of a method for determining the set position of the escape code in the two-dimensional Huffman table and the codeword length n. In the JPEG Huffman table, as shown in FIG.
The area up to is an unused area (strictly speaking, it is used in the JPEG extension mode, but is not within the scope of the present invention).

In the embodiment of the present invention described below, an escape code is set at the position of size 0 and run 1 in the Huffman table. However, the present invention is not limited to this, and escape codes may be set in other places in the above-described range.

The number of bits of a code word to be assigned to an escape code should be determined based on the sum of the appearance frequencies of combinations of runs and sizes included in the escape area. However, in practice, it is impossible to estimate the original frequency of occurrence from the number of allocated bits, so the code length of the escape code is determined based on the following idea.

First, it is conceivable to determine the length of the escape code from the condition that the code length does not change before and after the introduction of the escape area. As described above, the maximum code length outside the escape area is a total of 26 bits including 16 bits for the Huffman code and 10 bits for the additional bits.
On the other hand, since the maximum code length in the escape area is assigned 4 bits for the run to be added and 11 bits for the size, the code length n of the escape code is set to 11 bits. The maximum code length to be encoded can be made equal. As a result, it is possible to eliminate the increase and decrease of the code amount before and after the introduction of the escape area.

However, by setting the code length n of the escape code to 11 bits, one of the combinations of the run and the size to which the 11-bit code is assigned in the original JPEG code table becomes 11 bits. The code word is replaced by a code word having more than bits, which causes a problem of a decrease in efficiency.

As described above, when a code having an extremely short number of bits is used as the escape code, it is considered that the code allocation to a combination of a run and a size that frequently occurs is affected, and the coding efficiency is reduced. For this reason, the escape code is assigned as long as possible, and the original JPEG
It is conceivable that the encoding efficiency can be maintained.

However, if a too long code word is assigned to the escape code, on the other hand, a total of 15 bits, ie, 4 bits for the run and 11 bits for the size, are added, resulting in a longer overall code length. There is a problem in terms of conversion efficiency.

Therefore, as described above, since the escape area is set as an area having a particularly low occurrence frequency among the combinations of the run and the size, the escape area is one bit smaller than 11 bits.
A description will be given assuming that a code length of 12 bits, which is a bit longer, is given.

Of course, if a representative set of images to be encoded can be selected in advance, a preliminary encoding experiment is performed on these, and the bit length of the escape code is statistically determined from this. Is also good. In the present invention, the code length of the escape code is not limited to only 12 bits.

A procedure for setting an escape code having the set position and code word length determined as described above in an existing JPEG code tree will be described. This is to ensure that the escape code in the code data is uniquely identified from other Huffman codes.

In JPEG, a Huffman code tree is generated from two arrays, bits and values.
The following contents are set in bits and values, respectively. bits: an array indicating how many bits of codes are present. The maximum code length is up to 16 bits. values: coding elements arranged in order of occurrence frequency. In the case of the AC coefficient, the run and the size are represented by 4 bits each, and each position in the two-dimensional Huffman table is associated with an 8-bit numerical value (0 to 255) with the run being higher.

FIG. 6 is a diagram for explaining a method of assigning code lengths in JPEG. The first element of bits indicates the number of code words having a code length of 1 bit. In FIG. 6, b
Since the head element of “its” is 0, a 1-bit code word is not used. The second element of bits indicates the number of 2-bit codes, and there are two in FIG. As a result,
The first two elements of values will be assigned a 2-bit codeword. In FIG. 6, 0x01, 0x
02 corresponds to this. Similarly, the codeword length of the value element is determined by sequentially assigning the code length set in bits to the value element. It should be noted that the method of assigning a bit pattern serving as a code word is the same as the method described in the JPEG specification, and a description thereof will be omitted.

FIG. 7 is a diagram for explaining a procedure for setting an escape code in a JPEG values array. The escape code is set as a code word having a length of 12 bits as described above. Size 0 in the 2D Huffman table,
The position of the run 1 is 0x1 according to the address calculation described above.
It becomes 0 (10 in hexadecimal), and this value is set in values.

The upper part of FIG. 7 is an excerpt of a part of JPEG values, and shows a code to which a code near 12 bits is assigned. There are four codes set as 12-bit codes in the JPEG Huffman code table. These are 0x24,
0x33, 0x62, and 0x72, and their positions in the two-dimensional Huffman table are represented by (run, size) as (2,4), (3,3), (6,2),
(7, 2). As shown in the Huffman table of FIG. 12, the position where the 12-bit code exists is that (Run, Size) is (2,4), (3,3), (6,2), (7,
2).

The lower part of FIG. 7 shows changes in JPEG values when an escape code is added, corresponding to the upper part. As shown in the lower part of FIG. 7, the size 0, the position of the run 1, the 0x1 in the two-dimensional Huffman table
By inserting 0 at the beginning of the range to which a 12-bit code is assigned in the values table, an escape code having a code length of 12 bits is generated. However, this new one
The insertion of the two-bit code causes the subsequent codes to be shifted one by one from the originally set positions, so that the code length assignment is changed for some codes. FIG.
In the lower stage, one 12-bit code, 0x72, is shifted from the original 15-bit code allocation range.

Further, the value of the element of the array values in the position corresponding to the escape area is replaced with 0.
Also, the number of 16-bit codes included in the escape area is subtracted from the number of 16-bit codes in the array bits.

The bits and va modified by the above procedure
The two-dimensional Huffman code table used in the present invention can be newly generated by applying the JPEG Huffman code generation procedure to the two arrays of lues.

As a result, for a code word of less than 12 bits, a code which is not changed from the original JPEG code assignment is generated. Some codewords having a code length of 12 bits or more may have a code length longer than the original JPEG code bit length.

FIG. 8 shows some r_limit and s_limit.
It shows the escape area set by the combination of the limit and the code assignment for the combination of run and size outside the escape area. The code length in the escape area is 0, indicating that no code exists.

FIG. 8A shows r_limit = 3, s_l
The case where limit = 3 is shown. The escape area is r
≧ 3 and s ≧ 3, and the code length in these escape areas is 0. Size 0, Run 1 position 12
Bit code escape code 0x01 is assigned. Numerical value "12" shown below 4 at the upper left corner indicates this. This r_limit = 3, s_limit =
As a result of setting the escape area to 3, the code of the used area is reduced to 59 symbols.

FIG. 8B shows r_limit = 4, s_l
The case where limit = 4 is shown. The escape area is r
≧ 4 and s ≧ 4, and the code length in these escape areas is 0. As in the case of FIG. 8A, a 12-bit escape code 0x01 is assigned to the position of size 0 and run 1. Further, in the case of this escape area setting, as described in FIG.
What was originally a 12-bit code in the array values is replaced with a 12-bit code escape code.
Array value because it was inserted into the 12-bit code area of
A code 0x72 which is expelled in the 15-bit area of s occurs. This code 0x72 has the position (run, size) in the two-dimensional Huffman table as described above, (7, 2).
It is. Therefore, as shown in FIG. 8B, the code at the position where (run, size) is (7, 2) is replaced with 15 bits. This r_limit = 4, s_limit
As a result of setting the escape area with t = 4, the code of the used area is reduced to 79 symbols.

FIG. 8C shows r_limit = 5, s_l
The case where limit = 5 is shown. The escape area is r
≧ 5 and s ≧ 5, and the code length in these escape areas is 0. As in the case of FIG. 8A, a 12-bit escape code 0x01 is assigned to the position of size 0 and run 1. Also in the case of this escape area setting, the code at the position where (run, size) is (7, 2) is replaced with 15 bits as in the case of FIG. 8B. As a result of setting the escape area where r_limit = 5 and s_limit = 5, the sign of the used area is 9
Reduced to 7 symbols.

FIG. 8D shows that r_limit = 6 and s_l
The case where limit = 6 is shown. The escape area is r
≧ 6 and s ≧ 6, and the code length in these escape areas is 0. As in the case of FIG. 8A, a 12-bit escape code 0x01 is assigned to the position of size 0 and run 1. Also in the case of this escape area setting, the code at the position where (run, size) is (7, 2) is replaced with 15 bits as in the case of FIG. 8B. As a result of setting the escape area where r_limit = 6 and s_limit = 6, the sign of the used area is 1
It is reduced to 13 symbols.

The above is a procedure for setting an escape code and an escape area without changing a code having a code bit length of less than 12 bits in the JPEG Huffman code table. However, if a representative set of images to be encoded can be determined, and the appearance frequency of a combination of run and size can be determined based on these preliminary encoding experiments, a code table based on this can be designed. I do not care.

(3.2) Method of shortening the maximum code length By the way, code words used in JPEG having the same code word length have their bit patterns determined so that codes can be identified by several lower bits. Have been. For example, FIG.
As shown in the second JPEG Huffman table, there are a number of codewords having different run / sizes having the same 16-bit code length. An identification bit for identifying codes having a 16-bit code length is required.

Although the number of identification bits needs to be longer as the number of codes having the same code length is larger, the number of identification bits is smaller as the number of codes having the same code length is smaller. Therefore, if the number of Huffman codes can be reduced, the number of bits required to identify each code can be reduced. As a result, the maximum code length can be reduced.

There are a total of 125 16-bit codes in the JPEG two-dimensional Huffman table shown in FIG. 12, and the lower 7 bits are used to identify code words.

FIG. 9 shows a two-dimensional Huffman table (JPEG luminance) for the original JPEG luminance component and a table provided with various sizes of escape areas according to the present invention. , The bit length required for mutual identification of codewords, and JPE
FIG. 11 is a diagram comparing the relationship between the number of bits that can be reduced among the lower 7 bits of a 16-bit G code.

As can be understood from FIG. 9, in the JPEG luminance, the 16-bit code is 125, but in the table of the present invention in which various sizes of escape areas are provided, they are 22, 42, 60 and 75, respectively. Accordingly, the number of 16-bit codes is reduced, and the number of bits required for identification is also reduced along with the reduction except for the case where r_limit = 6 and s_limit = 6.

FIG. 10 shows an example in which codes are assigned in consideration of the reduction in the number of identification bits described above with reference to FIG. FIG. 10A shows s_limit =
4, code assignment when r_limit = 4. This is similar to FIG. 8B described above. By setting the escape area, the number of 16-bit codes becomes 42. At this time, in order to identify the codewords from each other, it suffices to have 6 bits. For this reason, the original JP
This means that the code length can be reduced by 1 bit with respect to the EG 16-bit code.

Further, as shown in FIG. 10B, s_lim
Consider setting an escape area where it = 4 and r_limit = 3. At this time, since the total of the 15-bit code and the 16-bit code is 31, only 5 bits are required for codeword identification. As a result, the 15-bit code and 1
All of the 6-bit codes can be 14-bit codes that are 2 bits shorter than the original maximum code length of 16 bits.

In order to reflect these code bit number allocation changes in actual code allocation, first, s_
Create an array values in which an escape area of limit = 4 and r_limit = 3 is set, and then create an array bi
The number of 15-bit codes and the number of 16-bit codes of ts are added and set as the number of 14-bit codes. Thereafter, the code can be generated in the same procedure as described above. FIG. 10B shows that s_limit = 4 and r_limit =
This is an example in which the escape area is enlarged and the maximum code length is shortened by setting 3.

(4) Operation of Encoding Device and Decoding Device of the Present Invention The operation of the two-dimensional Huffman coding device of the present invention described above with reference to FIG. 2 will be described below. In FIG.
The AC components of the quantized coefficients are input to the zero determination unit 11 in the zigzag scan order shown in FIG. Here, the coefficient that is zero is sent to the run counter 12, and the coefficient whose value is non-zero is sent to the size determination unit 13.

The run counter 12 counts the number of consecutive zero coefficients input. This count value is output to the run comparison unit 14 when the size determination unit 13 detects a non-zero coefficient. At this time, the count value is initialized to zero.

On the other hand, the size judging section 13 judges which of the plurality of groups classified by the magnitude of the input non-zero coefficient belongs to the magnitude of the value, and determines and outputs the group number. . Also, an additional bit indicating which of the numerical values in the group the amplitude value corresponds to is output.

The run comparing section 14 compares the input run information with a predetermined threshold value r_limit, and
If it is greater than or equal to t, the logic value 1 is output to the AND element 16. The size comparing unit 15 compares the input group number with a predetermined threshold s_limit, and
If it is greater than or equal to t, a logical value 1 is output. AND element 16
Outputs the logical product of the outputs of the run comparison unit 14 and the size comparison unit 15 to the code assignment unit 17.

The code allocating section 17 switches the coding operation according to the escape information input. If the escape information has the logical value 1, that is, if the combination of the run and the size to which the code is to be assigned is within the escape area, a codeword indicating the escape area is output. Outputs zero coefficient. When the escape information has a logical value of 0, it outputs input run information, a code word corresponding to the group number, and additional bits. Thus, the operation of the two-dimensional Huffman coding according to the present invention is completed.

Next, the operation of the two-dimensional Huffman decoding apparatus shown in FIG. 3 will be described. The buffer 21 temporarily stores input code data. This is because the length of the Huffman code cannot be known unless it is actually decoded, and it is necessary to hold the data until one decoding operation is completed and the head of the next code is detected. When the decoding of all the code data stored inside is completed, the buffer 21
The next code data is read out and stored.

The bit cutout unit 22 collectively reads out the bit data of the maximum length of the code word from the code start position in the buffer stored therein. A portion corresponding to the maximum length of the Huffman code from the beginning of the cut-out bit data having the maximum length is input to the code conversion table 23.

The code conversion table 23 includes an input bit string,
Then, it is determined which of the preset Huffman codes the partial sequence including the head of the input bit sequence matches. Further, it outputs a code length, an additional bit length, and run information corresponding to the Huffman code for which a match has been detected. If the decoded code is an escape code, escape information is further output. Note that the detected total value of the Huffman code length and the additional bit length is input to the bit cutout unit 22 and updates the code head position in the buffer stored therein.

The reconstruction unit 24 reads out a specific bit field in the bit string cut out by the bit cutout unit 22 based on the escape information, additional bits, and run information output from the code conversion table 23, and To restore.

By the above operation, the two-dimensional Huffman decoding apparatus of the present invention can decode the code data coded by the two-dimensional Huffman coding apparatus of the present invention.

With the above settings, the circuit size was estimated by applying the present invention to an actual LSI design. As a result, the circuit size of the two-dimensional Huffman decoding unit can be reduced by about 25% as compared with that of the JPEG system. is there.

As described above, the present invention has been described based on the recommended Huffman code for the AC component of the JPEG luminance signal. However, the present invention is also applicable to the same JPEG color difference signal Huffman code. In the present invention, the Huffman code has been described as an example, but another variable length code other than the Huffman code may be used.

[0106]

As described above, according to the present invention, in the two-dimensional Huffman coding method of the JPEG system, which is a coding standard of still images, codes which are hardly used in actual image coding are occupied. Heading the area,
This part was used as an escape area to reduce the number of codes. The determination of the escape area can be easily performed simply by comparing the run and the size with the two thresholds.

In the JPEG recommended Huffman table, frequently used code words are not changed in the bit allocation because they are outside the escape area. Therefore, general-purpose encoding performance for various images possessed by the JPEG Huffman table is secured.

Further, as a result of reducing the number of code words, the number of bits for identifying codes can be reduced.
As a result, the maximum code length can be reduced.

As a specific effect of the present invention, r_li
When mit = 4 and s_limit = 3, the total number of codewords is reduced from 162 to 67. At this time, the maximum code length of the Huffman code can be reduced from 16 bits to 14 bits.

By applying the configuration of the present invention, the circuit size of the two-dimensional Huffman decoding unit in the actual LSI circuit scale is 25 times smaller than that of the conventional JPEG system.
% Can be reduced.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating setting of an escape area in a two-dimensional Huffman table according to the present invention.

FIG. 2 is a diagram illustrating a configuration of a two-dimensional Huffman encoding device according to the present invention.

FIG. 3 is a diagram illustrating a configuration of a two-dimensional Huffman decoding device according to the present invention.

FIG. 4 is a diagram showing a correlation between a quantization scaling factor and generation of a two-dimensional Huffman code.

FIG. 5 is a diagram illustrating a relationship between a coefficient position of a transform coefficient and a code length in the encoding method of the present invention.

FIG. 6 is a diagram illustrating a method of assigning a code length based on bits (bits) and values (values).

FIG. 7 is a diagram illustrating addition of an escape code and change of bit allocation in the encoding method of the present invention.

FIG. 8 shows r_limi in the encoding method of the present invention.
It is a figure explaining the relation of the combination of t and s_limit and an escape area.

FIG. 9 shows s_limi in the encoding method of the present invention.
a combination of t, r_limit and the number of codes to be reduced,
FIG. 7 is a diagram illustrating the number of bits that can be reduced.

FIG. 10 shows r_limi in the encoding method of the present invention.
It is a figure explaining code bit number allocation at the time of t = 4 and s_limit = 3.

FIG. 11 is a diagram illustrating a transform coefficient matrix and a zigzag scan order.

FIG. 12 is a diagram for explaining a JPEG luminance component Huffman table.

FIG. 261 is a diagram for describing the code length of two-dimensional Huffman coding in H.261.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 Zero determination part 12 Run counter 13 Size determination part 14 Run comparison part 15 Size comparison part 16 AND element 17 Code allocation part 21 Buffer 22 Bit cutout part 23 Code conversion table 24 Reconstruction part

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Yasuharu Sakurai 430 Nakai-cho, Ashigamikami-gun, Kanagawa Prefecture Green Tech Inside Fuji Xerox Co., Ltd. F-term (reference) 5C059 MA00 MA23 MC11 ME02 ME08 ME15 ME17 PP01 PP04 SS28 TA17 TC43 TD11 UA05 5C078 AA04 AA09 BA57 CA01 DA00 DA01 DA11 DB05

Claims (14)

[Claims] 1. A run that is a continuous length of zero of a quantized coefficient obtained by quantizing a transform coefficient obtained by performing an orthogonal transform on an image signal, and a size indicating a magnitude of a non-zero quantized coefficient. And an image signal encoding method for performing encoding by combining the additional information for representing the value of the non-zero quantized coefficient, wherein a predetermined run threshold or more, and a predetermined size threshold or more. It is determined whether or not a combination of the run and the size of the quantization coefficient is included in the escape area defined by the combination of the run and the size, and the combination of the run and the size of the quantization coefficient is included in the escape area. When it is determined that the image signal is included in the image signal, an escape code indicating the escape area, a run value, and a value of a non-zero quantization coefficient are combined and encoded. If it is determined that the combination of the run and the size of the quantization coefficient based on is not included in the escape area, a codeword predetermined corresponding to the combination of the run and the size, and the non-zero An image signal encoding method characterized in that encoding is performed in combination with additional information indicating a value of a quantization coefficient. 2. The encoding according to claim 1, wherein the encoding is performed using a code table in which code words corresponding to a plurality of combinations of runs and sizes are predetermined. The escape area is a rectangular area in the code table. 2. The image signal encoding method according to claim 1, wherein: 3. A code table, wherein a code word corresponding to a combination of run and size included in the escape area is deleted from the code table, and a code word indicating the escape area is added to the code table. 3. The image signal encoding method according to claim 2, wherein 4. A code word set in the code table is set with an identification bit for identifying code words having the same code length, and a code deleted from the code table by setting the escape area. 4. The image signal encoding method according to claim 3, wherein the number of identification bits is set to be reduced according to the number of words. 5. The image signal encoding method according to claim 1, wherein the escape area is set as a combination area of a run and a size of a quantization coefficient having a low frequency of occurrence. 6. When it is determined that the combination of the run and size of the quantization coefficient is included in the escape area, an escape code indicating the escape area, a run value, and a non- The image signal encoding method according to any one of claims 1 to 5, wherein a value of a zero quantization coefficient is assigned to each fixed code length area, and encoding is performed by combining them. 7. The length of a code word to be assigned to the escape code is a value of the run when it is determined that the code word is included in the escape area from a maximum code length set for a quantization coefficient outside the escape area. 7. The image signal encoding method according to claim 6, wherein a code length is longer than or at least equal to a code length obtained by subtracting each fixed code length indicating a value of a non-zero quantization coefficient. 8. The image signal according to claim 1, wherein an escape area in which the run threshold is set to 4 and the size threshold is set to 3 is set, and a maximum code length is reduced by 2 bits. Encoding method. 9. A zero determining means for performing an orthogonal transform on an image signal and quantizing the obtained transform coefficient to determine that the quantized coefficient is zero, Means for counting runs indicating the continuous length of the quantized coefficient determined to be zero by the means; and determining a size indicating the size of the quantized coefficient determined to be non-zero by the zero determining means. A non-zero coefficient determining unit, and a code allocating unit that codes a code word corresponding to the combination of the run and the size and additional information indicating a value of the non-zero quantization coefficient. A run comparing means for comparing the run with a preset run threshold value; a size comparing means for comparing the size with a preset size threshold value; Escape information determining means for determining, when the size comparing means determines that the size is equal to or greater than the size threshold value, is a combination of a run and a size included in an escape area; and Code word allocating means for switching a code allocating operation in response to the output of, when the escape information determining means determines that a combination of run and size is included in the escape area, it is an escape area. When the escape information determination unit determines that the combination of run and size is not included in the escape area, the run is performed. Codewords and non-zero quantization corresponding to combinations of size and size Picture signal encoding apparatus characterized by having a codeword allocation means for encoding additional information representing the value of the number. 10. The image signal according to claim 9, wherein the escape information determination unit is configured by a logical operation element that outputs a logical product of comparison results of the run comparison unit and the size comparison unit. Encoding device. 11. The code word allocating means has a code table for obtaining a predetermined code word corresponding to a combination of a run and a size, wherein the code table is equal to or more than the run threshold value and the size threshold value. The above-mentioned area is set as an escape area, the escape area does not have a corresponding code word, and the code table is a table having an escape code indicating an escape area. Item 11. The image signal encoding device according to Item 9 or 10. 12. An image signal encoding method according to claim 9, wherein an escape area having a run threshold value of 4 and a size threshold value of 3 is set, and the maximum code length is shortened by 2 bits. Method. 13. Input code data, read out bit data of a predetermined length from the input code data, decode a variable-length code portion located at the head of the read bit data, and determine whether it is an escape area. Information indicating whether or not it is the escape area, performing code conversion to output the length of the additional bit, the run, and the entire length of the variable length code and the information accompanying the variable length code, Performing a quantization coefficient reconstruction for reading information from the bit data and restoring a quantization coefficient based on the length of the additional bit, the run, and the entire length of the variable length code and the accompanying information. A video signal decoding method. 14. A storage means for inputting code data and temporarily storing input code data, a bit cutout means for reading bit data of a predetermined length from said storage means, and a variable length located at the head of said bit data. Code conversion means for decoding a code part, and information indicating whether or not the area is an exceptional area, an additional bit length, a run, and outputting the entire length of the variable length code and information accompanying the variable length code; Information indicating whether the area is an exception area, an additional bit length, a run, and information based on the entire length of the variable-length code and the accompanying information. An image signal decoding apparatus, comprising: reconstruction means for restoring.