JPH11272843A - Device and method for encoding color image, and device and method for decoding color image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To shorten time for encoding and decoding, and to perform motion compensation sufficiently in consideration of a shadow part of an image by encoding and decoding a color image without using DCT. SOLUTION: An encoding device 1 for a moving picture consisting of multicolor images is provided with a thinning-out means 2c which thins out at least UV components among YUV components where RGB components are converted, a quantizing means 2d which quantizes the Y component which is not thinned out or slightly thinned out, and an encoding means which performs encoding by compressing the UV components that are largely thinned out and the quantized Y component. When motion compensation for the moving picture is performed, motion compensation is performed for the UV components and not performed for the Y component.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an apparatus and method for encoding a color image and an apparatus and method for decoding a color image. More specifically, the present invention relates to an improvement in a method of processing a YUV component when encoding and decoding a multi-color image or a natural image.

[0002]

2. Description of the Related Art Conventionally, in personal computers and game machines,
An image called a multi-color image is used. The multi-color image is also called a representative color image or a limited color image, and as shown in FIG. 37, specific colors, that is, specific R (red), G (green), B (blue) ) Is an image in which an index is assigned to a color having a value of ()) and expressed by a limited representative color such as 16 colors or 256 colors using data of the index.

[0003] Such multi-color image data is:
If each color of R, G, and B is represented by 8 bits (256 types), a total of 24 bits are required. However, since the index itself is displayed by, for example, 8 bits, considerable compression is required. Rate. However, although the data is compressed, the amount of information is still large, so if the processing is performed without any modification, the memory capacity is increased and the communication speed is reduced, which is not practical. Therefore, the compression technique of a multi-color image is extremely important, like other image data. Since the number of colors of a multicolor image is limited, generally, lossless encoding and decoding, that is, a reversible compression technique is required.

On the other hand, in recent years, a technique using an entropy encoder and a decoder has attracted attention as one of data compression techniques. As one of the entropy coding and decoding techniques, for example, there is one using an arithmetic coding and decoding technique. The outline of this technology is described in, for example, JP-A-62-185413 and JP-A-63-7432.
No. 4, JP-A-63-76525 and the like.

FIG. 35 shows a conventional multi-color image encoding system 50 and decoding system 60 using such a technique. The encoding system 50 includes a line buffer 51 and an entropy encoder 52. The input index data, that is, the color pixel data 100A, is input to the line buffer 51 and the entropy encoder 52. As shown in FIG. 36, all of the color pixel data 100A are raster-scanned and sequentially input as pixel data in the horizontal scanning order. As a method of creating the data of the index, that is, the color pixel data 100A, a method of assigning an index in the order of the input colors is general, and as shown in FIG. For example, the colors of “1” and “2” are significantly different, and those with distant index numbers, for example, “10”
A phenomenon occurs in which the colors are similar even for “0” and “200”. In order to avoid such a phenomenon, Japanese Patent Application Laid-Open No. 5-32
As indicated by reference numeral 8142, there is a case in which serial numbers are assigned to objects having similar colors.

The line buffer 5 in the encoding system 50
Reference numeral 1 denotes reference pixel generation means that generates reference pixel data A, B, C, and D for an encoding target pixel X from the already input color pixel data 100A. That is, the line buffer 51 stores histories for n lines (often about 1 to 5 lines) when scanning an image. Then, the color pixel data 10 of the encoding target pixel X is obtained.
Each time 0A is input, a series of pixel data including the immediately preceding pixel A and the surrounding pixels B, C, and D is output to the entropy encoder 52 as reference pixel data 110.

The entropy encoder 52 is, for example,
It is formed using a technique such as arithmetic coding or Huffman coding. Then, using the reference pixel data 110 as a state signal, the target color pixel data 100A is converted into encoded data 200 and output.

On the other hand, the decoding system 60 includes a line buffer 61 and an entropy decoder 62. Here, the line buffer 61 and the entropy decoder 62 are formed so as to decode and output the input coded data 200 in a procedure completely opposite to that of the line buffer 51 and the entropy coder 52 of the coding system 50. I have.

[0009] Thus, the encoding system 50,
The decoding system 60 converts the color pixel data 100A into the coded data 2 using completely opposite algorithms.
00, and the encoded data 200 can be decoded into color pixel data 100B and output. Therefore, this system can be widely used for various applications.

On the other hand, in the case of a natural image, the color image is directly represented by RGB components. However, in a standard called JPEG or MPEG, the data is usually once converted into a YUV component for ease of data compression. Moreover, the human eye is sensitive to the luminance component (= Y component) and the color component (= U
V component) is insensitive, so that the UV component is thinned out. The Y component and the decimated UV component are each divided into blocks, and each block is subjected to discrete cosine transform (hereinafter referred to as DCT). Thereafter, the DCT coefficients are quantized, and the values are encoded by Huffman coding or the like. Note that even in the case of a multi-color image, each R
Since the values of GB are entered, it is possible to perform the same processing after forming a natural image composed of RGB.

In coding and decoding of a moving image, what is called motion compensation is known. This motion compensation focuses on the fact that a moving image is a sequence of still images, and that when looking at each of them, there is a correlation with a previous frame or a subsequent frame. Specifically, the target frame is set to 16
If the predetermined block in the frame of interest is similar to a similar block shifted one pixel to the left of the reference frame, the image data of the block in the frame of interest is encoded or It encodes or decodes data that resembles a block shifted one pixel to the left in the reference frame without decoding.

That is, in this motion compensation, instead of encoding or decoding all the data of the frame of interest, data (moving area) of only a changed portion is sent,
For the same block as the block in the reference frame, the difference information can be further reduced by further converting the motion vector into data.

[0013]

SUMMARY OF THE INVENTION A multi-color image is
Due to the limited number of colors, lossless encoding and decoding are generally required. However, there is a case where it is desired to decode a multi-color image with a natural image, although it is irreversible (= lossy). In some cases, it is desired to decode a natural image by lossy. In the case of such a request, the conventional one requires a lot of time to execute, that is, perform encoding and decoding, because the DCT is used.

When motion compensation is applied to a recent stereoscopic image (3D image), it is important to efficiently encode and decode a portion corresponding to the shadow for the overall data compression and efficient decoding. I understand that. But,
Efficient encoding and decoding of a portion corresponding to a shadow has not yet been sufficiently solved.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-described problems, and it is possible to reduce the time required for encoding and decoding by encoding and decoding a color image without using DCT. It is an object of the present invention to provide a color image encoding apparatus and method, and a color image decoding apparatus and method. In addition, other inventions,
When encoding or decoding a moving image composed of color images,
It is an object of the present invention to provide a color image encoding device and a color image decoding device and a color image decoding device and a color image decoding method capable of performing motion compensation in which a shadow portion of the image is sufficiently considered.

[0016]

In order to achieve the above object, according to the present invention, in a color image encoding apparatus, a thinning means for thinning at least a UV component of a YUV component converted from an RGB component, Quantizing means for quantizing a Y component which is not thinned out or thinned out, and coding means for compressing and coding a UV component which has been thinned out and the quantized Y component are provided. ing.

For this reason, the color image is encoded without using the DCT, and the encoding time can be reduced. Also, the Y component is quantized to reduce its value,
Since the UV component is greatly thinned out, the data amount is similarly reduced, and encoding can be performed efficiently.

According to the second aspect of the present invention, when a moving image is formed by continuously forming a still image frame composed of color pixel data, each of the frames of interest to be encoded is set to N.
When the block in the frame of interest is the same as the block in the reference frame, the motion vector of the block is converted into data as difference information, In a color image encoding apparatus that performs motion compensation to reduce the data amount of difference information, a thinning-out unit that thins out at least a UV component of a YUV component converted from an RGB component, and Y that is not thinned out or thinly thinned out. Quantizing means for quantizing the component, and encoding means for compressing and encoding the greatly thinned UV component and the quantized Y component are provided, and motion compensation is performed for the UV component. The motion compensation is not performed.

Generally, the luminance (Y
Component) greatly changes depending on how light hits. Therefore, it is not suitable for motion compensation. On the other hand, UV which is a color component
The components do not change significantly. Therefore, if motion compensation is used only for the UV component, efficient data compression becomes possible.

According to the third aspect of the present invention, in the color image encoding apparatus according to the first or second aspect, the color image is converted into a multi-color image, and the multi-color image is converted into a natural image composed of RGB components. Natural image conversion means, YUV conversion means for converting the converted RGB components into YUV components, and palletizing means for assigning an index to the largely thinned out UV components and palletizing, the encoding means, The quantized Y component and the palletized indexed UV component are encoded.

Therefore, a multi-color image can be irreversibly encoded like a natural image, and the data compression ratio can be improved. In addition, since the UV component is indexed as a set of UVs and is palletized, the compression efficiency is further improved.

In addition, according to the fourth aspect of the present invention, in the color image encoding apparatus according to the first, second or third aspect, the encoding means arranges the Y component and the UV component according to the frequency of appearance of the color. It has a color order conversion means for changing the color order, and a run length modeling means for making a run length model of the values of the components rearranged by the color order conversion means. As described above, since the image processing apparatus includes the color order conversion unit and the run-length modeling unit and performs encoding, the data compression efficiency is extremely high.

According to a fifth aspect of the present invention, in the color image encoding method, the Y component converted from the RGB component is used.
A thinning step of thinning at least the UV component of the UV component,
A quantization step of quantizing a Y component that is not thinned out or thinned out, and an encoding step of compressing and coding the UV component that has been largely thinned out and the quantized Y component. ing.

For this reason, the color image is encoded without using the DCT, and the encoding time can be shortened. Also, the Y component is quantized to reduce its value,
Since the UV component is greatly thinned out, the data amount is similarly reduced, and encoding can be performed efficiently.

According to the sixth aspect of the present invention, when a still image frame composed of color pixel data is formed continuously to form a moving image, each of the frames of interest to be encoded is set to N.
When the block in the frame of interest is the same as the block in the reference frame, the motion vector of the block is converted into data as difference information, In a color image encoding method for performing motion compensation for reducing the data amount of difference information, a thinning step of thinning at least a UV component of a YUV component converted from an RGB component, and a Y thinning that is not thinned or thinned out. A quantization step of quantizing the component, and an encoding step of compressing and encoding the significantly thinned UV component and the quantized Y component, performing motion compensation on the UV component, The motion compensation is not performed.

In general, the brightness (Y
Component) greatly changes depending on how light hits. Therefore, it is not suitable for motion compensation. On the other hand, UV which is a color component
The components do not change significantly. Therefore, if motion compensation is used only for the UV component, efficient data compression becomes possible.

According to a seventh aspect of the present invention, in the color image encoding method according to the fifth or sixth aspect, the color image is a multi-color image, and the multi-color image is converted into a natural image composed of RGB components. A natural image conversion step, a YUV conversion step of converting the converted RGB components into a YUV component, and a palletization step of assigning an index to the greatly thinned out UV components and palletizing the components are provided. The quantized Y component and the palletized indexed UV component are encoded.

Therefore, a multicolor image can be irreversibly encoded like a natural image, and the data compression ratio can be improved. In addition, since the UV component is indexed as a set of UVs and is palletized, the compression efficiency is further improved.

In addition, in the invention according to claim 8, in the color image encoding method according to claim 1, the encoding step arranges the Y component and the UV component according to the frequency of appearance of the color. It has a color order conversion step for changing the color order, and a run length modeling step for forming a run length model of the values of the components rearranged in the color order conversion step. As described above, since the encoding process includes the color order conversion process and the run length modeling process, the data compression efficiency is extremely high.

According to a ninth aspect of the present invention, in a color image decoding apparatus having a YUV component, a decoding means for decoding input target encoded data, and a quantized quantized data in the decoded data are provided. A smoothing means for smoothing the Y component, and a U which is greatly thinned compared to the Y component.
And an intermediate color processing means for processing the V component by intermediate color interpolation.

Therefore, a color image can be decoded without using the DCT, and the decoding time can be reduced. In addition, since the Y component is quantized to reduce its value and the UV component is greatly thinned out, the amount of data is similarly reduced, and decoding can be performed efficiently.

Further, according to the present invention, there is provided a decoding means for decoding input target encoded data into color pixel data comprising YUV components, and a moving picture comprising a series of still images comprising said color pixel data. And a moving image forming unit that divides a target frame to be decoded into a plurality of blocks each including N × N pixels (N is an integer of 4 or more). When a motion vector indicating the relative displacement of the block exists in the decoded data related to the block, a color image decoding apparatus that performs motion compensation for decoding the block using the same block in the reference frame to be referred to In the data decoded by the decoding means, the motion compensation is performed for the UV component, and the motion compensation is not performed for the Y component. It has to.

Generally, the brightness (Y
Component) greatly changes depending on how light hits. Therefore, it is not suitable for motion compensation. On the other hand, UV which is a color component
The components do not change significantly. Therefore, if motion compensation is used only for the UV component, efficient decoding of a moving image becomes possible.

Further, according to the eleventh aspect of the present invention, the decoding means outputs the Y component and the UV according to the frequency of appearance of the color.
Color order conversion means for rearranging and decoding the components, and run-length decoding means for run-length decoding data indicating the values of the components rearranged by the color order conversion means are provided. As described above, since the image processing apparatus has the color rank conversion unit and the run length modeling unit and performs decoding,
The data decoding efficiency becomes extremely high.

According to a twelfth aspect of the present invention, in the method for decoding a color image having a YUV component, a decoding step of decoding input target encoded data, and a step of decoding quantized data in the decoded data. A smoothing step for smoothing the Y component and an intermediate color processing step for processing the UV component, which is thinned out more than the Y component, by intermediate color interpolation are provided.

Therefore, a color image can be decoded without using the DCT, and the decoding time can be reduced. In addition, since the Y component is quantized to reduce its value and the UV component is greatly thinned out, the amount of data is similarly reduced, and decoding can be performed efficiently.

According to a thirteenth aspect of the present invention, there is provided a decoding step of decoding input target encoded data into color pixel data composed of YUV components, and a moving picture by continuously forming a still image composed of the color pixel data. And a moving image forming step of dividing the target frame to be decoded into a plurality of blocks each including N × N pixels (N is an integer of 4 or more). A color image decoding method for performing motion compensation for decoding a block using a same block in a reference frame when a motion vector indicating a relative displacement of the block exists in decoded data related to the block In the data decoded in the decoding step, motion compensation is performed for the UV component, and the motion compensation is not performed for the Y component. It has to.

In general, the brightness (Y
Component) greatly changes depending on how light hits. Therefore, it is not suitable for motion compensation. On the other hand, UV which is a color component
The components do not change significantly. Therefore, if motion compensation is used only for the UV component, efficient decoding of a moving image becomes possible.

Further, in the invention according to claim 14, in the decoding step, the Y component and the UV of the
A color rank conversion step for rearranging and decoding the components and a run length decoding step for run length decoding data indicating the values of the components rearranged by the color rank conversion means are provided. As described above, since the image has the color order conversion step and the run-length modeling step and is decoded,
The data decoding efficiency becomes extremely high.

[0040]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described with reference to FIGS. In addition, each data corresponding to the data in the prior art will be described with the same reference numerals.

FIG. 1 shows a preferred embodiment of a moving picture coding system 1 comprising multi-color pictures according to the present invention. FIG. 3 shows a moving image decoding system 3 composed of multi-color images corresponding to the encoding system 1 of FIG.
1 shows a preferred embodiment of the present invention.

The encoding system 1 includes a preprocessing unit 2
And two frame memories 1a and 1b for storing frames as still images, and an exclusive logic generation unit / motion compensation unit 1 for generating a motion vector V for performing motion compensation.
c, an entropy encoder 4 for encoding the motion vector V generated by the exclusive logic generation means / motion compensation means 1c, and a control circuit 5 for controlling the switches 5a and 5b.
An index conversion unit 7, a line buffer 10 as a peripheral pixel generation unit and a reference pixel generation unit, a Markov model generation unit 11 as a degeneration unit, an entropy encoder 12 as an entropy encoding unit, and a synthesis unit 20
And configured to convert the input data stream of the color pixel data 100A into the data stream of the encoded data 200 and output the converted data stream. The pre-processing means 2 and subsequent parts are coding means.

The pre-processing means 2 includes a natural image converting means 2a for converting the color pixel data 100A of the multi-color image into a natural image composed of RGB components, and a converted RGB image.
YUV conversion means 2b for converting the component into a YUV component;
Thinning means 2c for thinning out at least the UV component of the UV component
A quantizing means 2d for quantizing the Y component which is not thinned out or thinned out; an indexing means 2e for indexing the quantized Y component; And palletizing means 2f for palletizing.

In this embodiment, the color pixel data 100A of the multi-color image is a 256-color index image. The RGB image is a 24-bit natural image composed of 8 bits each. Further, the thinning means 2c thins out the UV component from the YUV component. This thinning-out means 2c thins out YUV to 4: 2: 2 or 4: 1: 1. It should be noted that the thinning-out may be performed at a different ratio instead of the same ratio in the vertical and horizontal directions. If necessary, the Y component may be thinned out, but the degree of the thinning needs to be smaller than that of the UV component.

The quantizing means 2d has the same configuration as the quantizing means used in the JPEG standard or the like, and reduces the value of Y by a predetermined coefficient. The quantized Y component is given an index of 0, 1, 2, etc. by the indexing means 2e. On the other hand, U
The V component is indexed as a pair of UV by the palletizing means 2f, and the value is palletized.

Exclusive logic generation means / motion compensation means 1c
Has the same function as the conventional one. Here, this operation will be briefly described.

The color pixel data 100A 'is input to one of the frame memories 1a and 1b as multi-color pixel data. This input signal is input to the exclusive logic generation means / motion compensation means 1c. When the original signal is input according to raster scan, pixel data is input to an image buffer (not shown) in order to perform pixel comparison in block units by the exclusive logic generation unit / motion compensation unit 1c. I do. In this case, the image buffer has a capacity to input image data of at least the number of vertical pixels of the block × the number of horizontal pixels of the image frame. The value of this image buffer and the past or future image data stored in the frame memories 1a and 1b are obtained by an exclusive OR XOR.
Verify whether the corresponding pixels match each other. In this verification, a block in the frame memories 1a and 1b having the same coordinates as the block in the image buffer is taken out and XOR is performed.
In addition to the comparison, the blocks around the same coordinates are also compared.

The pixel data is 1 in a multi-color image.
It consists of about 1 byte of data per pixel. Since the exclusive OR XOR is a check as to whether the pixel of the frame image data and the pixel of the image buffer are the same,
Not all bits representing one pixel are XORed, but if any one bit is different, the confirmation is terminated.

Pixel coincidence and non-coincidence are counted by a counter (not shown) in the exclusive logic generation means / motion compensation means 1c. Here, it is assumed that the counter increases every time a mismatch is detected. This counter is reset each time counting of one block of the frame memories 1a and 1b is completed. The count value immediately before the reset is stored in the count value array storage unit (not shown) inherent in the counter for each shift between the position of the block currently being compared and the block position of the frame memories 1a and 1b. Store.

At the end of the comparison of one block of the current image, the smallest value of the count value is searched in the array storage. If the value is larger than the set threshold value, it is determined that the current block is not the same or similar to the past or future block at the corresponding position. The operation performed based on the result of the determination differs between forward prediction, that is, prediction from the past to the current direction, reverse prediction, that is, prediction from the future to the current direction, and bidirectional prediction, that is, prediction from the past or future to the present.

In the forward prediction, when it is determined that they are the same or similar, the area of the past frame having the fewest mismatch points is copied to the current frame. Conversely, if they are not the same or similar, copy from the image buffer to the current frame. At this time, in portions other than the exclusive logic generation unit / motion compensation unit 1c, data corresponding to the compared block in the image buffer is encoded by the entropy encoder 12.

In the backward prediction, when it is determined that they are the same or similar, the area of the future frame having the fewest mismatch points is copied to the current frame. Conversely, if they are not the same or similar, copy from the image buffer to the current frame. At this time, in portions other than the exclusive logic generation unit / motion compensation unit 1c, data corresponding to the compared block in the image buffer is encoded by the entropy encoder 12.

In the bidirectional prediction, when it is determined that they are not the same or similar, data corresponding to the compared block in the image buffer is encoded by the entropy encoder 12 except for the exclusive logic generation means / motion compensation means 1c. When it is determined that they are the same or similar, the prediction is the same as the forward prediction or the backward prediction.

The displacement of the relative position between the area having the fewest mismatch points and the area currently being determined when it is determined that the forward, backward and bidirectional predictions are the same or similar is referred to as a motion vector V. . The motion vector V is compressed or multiplexed by an encoding procedure determined for the motion vector V.

The operation of such motion compensation will be described with reference to the drawings. For example, a frame for which prediction is not used at all is an I picture, and a frame for which prediction is
Assuming that a picture and a frame to be predicted using past and future images are B pictures, the input order is PBBPBBIBBPBBPBBI as shown in FIG. 5A. Thus, when a role is given to each frame, the encoding order is as shown in FIG. That is, for example, 8 is No. 8 No. 7 I-picture and No. 7 After 10 P pictures have been encoded, they are encoded using those data. At the time of this encoding, motion compensation is appropriately used.

The motion compensation is used when encoding a P picture and a B picture. Specifically, as shown in FIG. 6, a frame to be coded is 16 × 1
The macroblock MB is divided into 6-pixel macroblocks, and a macroblock determined to be the same as the macroblock MB is searched for from the reference frame. The reference frame is an I picture when encoding a P picture, and an I picture and a P picture when encoding a B picture.

The method of searching for the same block from the reference frame is performed by comparing the position of the corresponding macro block MB in the reference frame and each pixel of the macro block MB slightly shifted from the position, as in the conventional method. . In this embodiment, the determination as to whether or not the pixels are the same is made based on the number of pixels at corresponding positions in both macroblocks MB that are different. That is, when the number of mismatches becomes equal to or less than a predetermined value, the judging means (not shown) in the exclusive logic generating means / motion compensating means 1c determines whether the two macro blocks MB
Are determined to be the same. In this embodiment, the predetermined value for determining the macroblock MB is set to 20.

If they are determined to be the same, exclusive logic generation means /
The motion compensator 1 c generates a motion vector V in which the moving directions of both macroblocks MB are vectorized, and sends the value to the entropy encoder 12. The exclusive logic generation unit / motion compensation unit 1c has a small block generation unit (not shown) that divides the macroblock MB into four and generates four small blocks SB. Then, it is determined by the determining means whether or not the small blocks SB are the same. In this embodiment, when the number of mismatches between the pixels of the block SB and the block SB in the reference frame is 3 or less, the blocks SB are determined to be the same. It is preferable that the predetermined number serving as a criterion for judging whether or not they are the same is a value of 0 to 5 in order to prevent the disturbance of the multi-color image and perform the effective compression.

For the small block SB, the data portion for the motion vector V is not provided, and the motion vector V of the previous macro block MB is used as it is.
For this reason, the data portion for the motion vector V becomes small and does not cause an increase in data. On the other hand, the determination as to whether or not they are the same can be made strictly with a small block SB, so that there is no disturbance of the multi-color image.

For example, the macro block MB shown in FIG.
Assuming that the number of pixel mismatches between the reference frame and the macroblock MB is 18, the two macroblocks MB are determined to be the same, and the motion vector V is generated. On the other hand, the number of mismatches of the small block SB (1) is 3, and SB (2)
Is two and SB (3) is two, a small block S
It is determined that B (1), SB (2), and SB (3) are the same, and data is compressed using the motion vector V generated by the macroblock MB. However, the number of mismatches of the small block SB (4) is 11, and the pixel of the small block SB (4) is encoded without using motion compensation.

In this embodiment, the macro block MB
Of the small blocks SB is made stricter. For this reason, while the number of blocks that can use the motion vector increases, the image is not disturbed. Note that the macroblock MB
May be made stricter, and the same determination may be made less strictly for small blocks SB, or both may be of the same level.

As shown in FIG. 2, the index conversion unit 7 includes a line buffer 10 serving as a reference pixel generation unit, a Markov model generation unit 11 serving as a degeneration unit, and an entropy encoder 12 serving as an entropy coding unit. A state generator 13 for generating a state signal ST, a conversion table generator 14 for generating two types of conversion tables during prescan, a reference order table 15, and a pixel index (color symbol) 3 for encoding. Pixel index conversion table 16 for input and index conversion
And a color pallet conversion unit 17, a converted color pallet 18, and a determination unit 19 that receives the converted pixel index and outputs a predetermined order.

The line buffer 10 inputs reference pixels to the state generator 13 during prescan, and inputs reference pixels to the Markov model generator 11 during encoding.
As shown in FIG. 7A, the Markov model generation unit 11 sets the surrounding four pixels R as a reference pixel for the encoding target pixel buf [i].

[0] to R [3] are taken in to generate a Markov state signal CX.

As will be described later, the entropy encoder 12 is modeled by predictive run-length encoding and encodes an image signal into encoded code. The entropy encoder 12 may employ an arithmetic entropy encoder shown in FIG. 40 or an encoder using a Huffman code. Note that the entropy encoder 12
The details of the configuration, operation, and function of will be described later.

The state generating section 13 operates at the time of pre-scanning prior to the encoding process. As shown in FIG. 7A, when the encoding target pixel is buf [i], the surrounding reference pixels R

The states [0] to R [03] are classified as shown in FIG. For example, R

When [0] to R [3] are all the same color, the state signal S
T becomes "0" and R

When [0] to R [3] are all different colors, the state signal ST becomes "33". When the encoding target pixel buf [i] is in the leading raster, all the reference pixels R [1] to R [3] which do not appear, as shown in FIG. 0 ", that is, the most frequent pixel index is set as described later. When the encoding target pixel buf [i] is at the head of the image, R

“0” is set to all of [0] to R [3].

The pixel index is represented by the luminance component Y
And the color components UV. The luminance gradation is reduced to about 64 (or 32) by quantization. Therefore, the luminance index is calculated from YC0 to Y
There are about 64 types of C63. Also, the UV index is likely to decrease instead of staying at 256 due to thinning.

The conversion table generator 14 operates only at the time of prescan. Then, the encoding target pixel buf
When [i] matches four reference pixels (hereinafter referred to as R [4]), the frequency of the position of the matching color is calculated for each state signal, and the encoding target is included in the reference pixel R [4]. When there is no pixel buf [i], the frequency of appearance of the pixel index that does not match is counted.

For example, when the state signal ST is “3”, as shown in FIG. 9A, the encoding target pixel buf [i] is set to the reference pixel R

The number corresponding to [0] is defined as NA, and the reference pixel R
Assuming that the number matching [2] is NB, if NA> NB, the reference pixel position of the reference order table 15 to be created is as shown in FIG.
As shown in FIG.

[0] comes and R is in 1st place
[2] comes. On the other hand, when NA <NB, FIG.
As shown in the figure, R [2]

[0] comes. In this way, a reference rank table 15 as shown in FIG. 10 is created. The numbers in the reference order table 15, for example, “2” in the 0th place and “0” in the 1st place when the status signal ST is “3” are, in the example shown above, the R in the 0th place. [2] of "2" is replaced by R

"0" of [0] is shown. When the status signal ST is “0”, the order is only the 0th place and is excluded from the reference order table 15. Therefore, the table 15 is a table having 14 columns and 4 columns.

When the color of the pixel to be coded buf [i] does not exist in the reference pixel R [4], the conversion table generation unit 14 outputs the reference pixel R [4] as shown in FIG. Are counted for each pixel index. Then, a pixel index conversion table 16 as shown in FIG. 12 is generated. In addition, what is shown in FIG.
... ≧ Nn.

The color pallet conversion unit 17 converts the color pallet (for example, a color pallet as shown on the left side of FIG. 13) 9 into a converted color pallet as shown in FIG. 18 is generated. For example, if the pixel index (color symbol) YC2 is Y2 as the luminance component Y, and if the rank of the frequency that was not present in the reference pixel R [4] is the first, that is, the rank is 0, the conversion color palette At 18, the color symbol YC0 is set at the highest position.

The discriminator 19 operates at the time of encoding, and receives a pixel index converted by the pixel index conversion table 16 and a reference pixel based on the state signal ST. When the pixel index matches the reference pixel, the order is output based on the reference order table 15. If they do not match, a number obtained by adding 1 to the Markov state signal CX is added to the order of the pixel index and output. For example, when the converted pixel index is YCn and the state signal ST is “3”, the same color as the pixel index YCn is used as the reference pixel R

When it is [0], "1" is output as the rank. On the other hand, if the pixel index YCn is the reference pixel R
When [4] does not exist, since CX is “1”, “n + 1”
+1 "=" n + 2 "is output.

The entropy encoder 12 transforms the output of the discriminating section 19 into a Markov model with a Markov state signal CX and performs variable length coding. The synthesizing unit 20 synthesizes the data of the converted color palette 18, the data of the reference order table 15, and the output from the entropy encoder 12, and outputs the result as encoded data 200. In this embodiment, the encoded data 200 includes information such as the size of an image, data of a converted color palette 18, data of a reference order table 15, and finally an encoded code encoded by the entropy encoder 12. Although they are output in order, basically, they only need to be included in the encoded data 200, and may be in another order.

The luminance component Y has been described above.
The same applies to the V component. In the following description, the luminance component Y and the UV component are both processed in the same manner.

Next, the operation of the encoding system 1 having the above configuration will be described with reference to the flowcharts shown in FIGS. Since the operation of the exclusive logic generation unit / motion compensation unit 1c has already been described, here, the operation of the index conversion unit 7 and its peripheral members will be described.

The operation of the coding system 1 is divided into a prescan process and a coding process. In the pre-scan process corresponding to the pre-processing of the encoding, the pixel index conversion table 16 and the reference order table 15 are created so that appropriate encoding can be performed in the subsequent encoding process. For this purpose, first, switch 21 is connected to terminal 21a.
Connect to Then, in step S0, a count table, a reference rank table 15, and a pixel index conversion table 1 described later in the conversion table generation unit 14 are stored.
6 and initialization of various variables for encoding processing to be described later.

In step S 1, the data of the color palette 9 of the input image is stored in the array p in the color palette conversion unit 17.
Input to allet []. In step S2, the pixel index 100A 'of the input image is stored in the line buffer 1
Input to array buf [] in 0. Next, step S
In 3, four pixels around the encoding target pixel buf [i] are extracted from the line buffer 10, and the conversion table generation unit 1
4 and the array R [] of the state generation unit 13 and the like. The encoding target pixel buf [i] and the reference pixel R

[0] to R [3]
Are as shown in FIG. In step S4, based on the reference pixels extracted from the line buffer 10,
The state signal ST is generated based on the reference in FIG.

In step S5, it is checked whether or not any of the reference pixels R [4] matches the pixel to be coded buf [i]. Then, in step 6, the reference pixel R [i]
If it matches, the count table N_table_A for the reference rank table 15 in the conversion table generation unit 14
One is added to [ST + j]. Here, since j represents the position of the reference pixel, the number of times of matching is counted in N_table_A [] for each position that matches the state of the reference pixel. In the next step S7, if the pixel does not match the reference pixel, the count table N_tabl for creating the pixel index conversion table 16 in the conversion table generation unit 14
One is added to e_B [buf [i]]. That is, in the count table N_table_B [buf [i]],
A pixel that does not match the reference pixel is counted for each pixel index 3.

At step S8, for all the pixel data, step S3 to step 6 or step 7
The process up to is repeated. Count table N_tabl
In e_B [], the frequency of appearance is counted for each pixel index 100A. In step S9, a pixel index conversion table 16 for converting a code having a higher frequency of occurrence into an ascending code is created based on the table. According to this table 16, the pixel index 10 having a high appearance frequency
If the index value is converted to a smaller value for 0A and entropy encoding is prepared according to the characteristic, the most efficient entropy compression can be performed for any image.

In step S10, a reference rank table 15 describing the reference rank is created. That is, in the count table N_table_A [], since the frequency is counted for each reference pixel position that matches the reference pixel status signal ST and the encoding target pixel buf [i], the ascending order can be added in descending order of frequency. If so, the most efficient compression can be expected.

After the reference rank table 15 and the pixel index conversion table 16 are created by the pre-scan process, the encoding process shown in FIG. 15 is executed. In executing this, first, the switch 2 is switched to the terminal 21b side.

The reference order table 15 created in the prescan process is necessary for the decoding process. Therefore, in order to add to the stream of the encoded data 200 as a part of the header information of the compressed stream, step S
At 11, it is output to the synthesis unit 20. In step S12, the pixel index 100 is used by using the pixel index conversion table 16 created in the prescan process.
A 'is converted. The pixel index 100A ′ that does not match the reference pixel by this conversion becomes a code in ascending order in descending order of the appearance frequency.

Along with the conversion of the pixel index 100A ', it is necessary to convert the color palette so that the decoding side reproduces the correct color. Therefore, step S13
Then, using the pixel index conversion table 16, the color palette conversion unit 17 creates the conversion color palette 18. The converted color palette 18 is input to the synthesizing unit 20 and added as a part of the header information of the compressed stream composed of the encoded data 200. In step S14, the same operation as in step S3 of the prescan process is performed. That is, the reference pixel R [4] is extracted from the line buffer 10 and input to the Markov model generation unit 11 and the state generation unit 13.

In the next step S15, the same operation as in step S4 of the prescan process is performed. That is, the state generator 13 generates the state signal S based on the reference shown in FIG.
T is output to the reference ranking table 15. In step S16, a Markov state signal CX is generated. The Markov state signal CX is a value representing the number of colors of the reference pixel,
It is generated from the state of the reference pixel based on the table of FIG. Although not specified in the flow of this encoding process, this Markov state signal CX is also used as an encoding condition for Markov modeling when performing entropy encoding.

In step S17, the pixel to be encoded bu
It is checked whether or not f [i] matches the reference pixel R [4]. Specifically, the state signal S in the reference order table 15
The value in the column corresponding to T is changed from 0 to a value equal to CX, and it is checked whether or not the value matches the encoding target pixel buf [i]. If it matches the reference pixel, the process immediately proceeds to step S18; otherwise, the process proceeds to step S19.

If the pixel matches the reference pixel, the rank at that time is subjected to entropy coding (step S18). If it does not match the reference pixel, the encoding target pixel buf
[I] is directly entropy coded. However, (CX
Since (+1) codes are assigned to the code in the case where the reference pixel coincides with the reference pixel, (CX + 1) is added and encoded (step S19). In step 20, the processing from step S14 to step S18 or step S19 is repeated for all the pixel data.

In the above-mentioned flow, the entropy coding is performed sequentially with the determination of the order in step S18 or S19. The Markov state signal CX corresponding to the order of all pixels is temporarily stored in a buffer. Alternatively, the encoding described later may be performed collectively. For example, when a prediction run-length encoding method described later is adopted, future symbols after the encoding target symbol are encoded collectively.

In this embodiment, motion compensation is not performed on the Y component, which is a luminance component, and the UV component of the color component is not corrected.
Motion compensation is performed only for the component.

In coding a P picture or a B picture, when it is determined that a predetermined macro block MB does not perform motion compensation, the macro block MB
Are picked up and coded continuously. For example, FIG.
As shown in FIG. 1 to
No. When No. 7 does not perform motion compensation, the coding is No. 1 → No. 4 → No. 2 → No. 6 → No. 7
→ No. 5 → No. 3 is performed in order. Therefore, effective compression is possible even if the memory capacity is reduced.

For example, as shown in FIG. 1,
No. 4 and No. 2 macro blocks MB. 1 → No. No. 2 is encoded. N of macro block MB of 2
o. To effectively compress the pixels adjacent to 1
No. At least one on the right end of one macroblock MB
An extra memory for storing a column is required and access is troublesome. However, no. 1 → N
o. No. 4 is encoded. No. 4 macro block MB No. 4 The latest appearance table described above can be used as it is for encoding pixels in a portion adjacent to 1, and no extra memory is required, and the access time is shortened.

Next, a decoding system 3 for a moving image composed of multi-color images corresponding to the encoding system 1 will be described.

The decoding system 3 includes a switch 6
a control circuit 6 for switching between a, 6b, and 6c, a motion compensating unit 3a, and a correcting unit 9 for correcting the Y component and the UV component
, Two frame memories 3b and 3c, an index conversion unit 8, a line buffer 30 as a peripheral pixel generation unit and a reference pixel generation unit, a Markov model generation unit 31 as a degeneration unit, and an entropy decoding unit. It includes an entropy decoder 32 and a separation unit 33, and is formed to convert a data stream of input coded data 200 into a data stream of color pixel data 100B and output the data stream. Note that the configuration before the correction means 9 becomes the decoding means.

The correction means 9 comprises a smoothing means 9a for performing a smoothing process on the Y component and an intermediate color processing means 9b for processing the UV component by intermediate color interpolation.
The smoothing process is a known processing method performed by a filter having a predetermined coefficient. For example, there is a filter or the like that has a total of 9 squares in 3 rows and 3 columns, and has a coefficient of 0.2 in the center and 0.1 in the others. In addition, the intermediate color interpolation is, for example, a correction in which an intermediate value is set to an intermediate value between values of pixels on both sides thereof, and various conventionally known correction methods can be adopted. Note that, for a portion exactly at the center of the four pixels, the average value of the four pixels or the intermediate value of the two pixels in the specific direction is used.

In this embodiment, the smoothing means 9a also performs an edge preserving process. This edge preservation processing is to rotate a fixed area including a target pixel, for example, a rectangular area by 360 degrees around the pixel, and assign the pixel to a color having the smallest change rate. This makes it possible to reliably preserve edges that are likely to be lost by the smoothing process.

The algorithm of the decoding system 3 is as follows.
The decoding means is an algorithm completely opposite to the coding means, but is totally different from the algorithm of the coding system 1 as a whole. Therefore, the color pixel data 10
The bit configuration and data stream of 0A are slightly different from the bit configuration and data stream of color pixel data 100B. Note that the motion compensation unit 3a
The operation is basically the reverse of that of the exclusive logic generation means / motion compensation means 1c of the encoding system 1 described above, and includes a judgment means 3d for judging the presence or absence of the motion vector V inside.

The operation of the motion compensator 3a is as follows. That is, based on the information on the motion vector V, the motion vector V and information on whether or not to use the motion vector V are obtained. When the motion vector V is used, the data in the frame memories 3b and 3c is decoded. When the motion vector V is not used, the data decoded from the entropy decoder 32 is directly used as the decoded data.

When the motion vector V is used, data is transferred from past data in the forward direction and from future data in the backward direction. In the case of bidirectional transmission, either the past or the future is selected and transferred. On the other hand, when the motion vector V is not used, the data from the entropy decoder 32 is output as decoded data, and in the case of forward and reverse directions, the data is stored in the frame memories 3b and 3c as current frame data. .

The index conversion unit 8 is connected to a line buffer 30 serving as reference pixel generation means, a Markov model generation unit 31 serving as degeneration means, and an entropy decoder 32 serving as entropy decoding means. A state generating unit 34 to generate, a reference order table 35,
And a pixel index (color symbol) generator 36.

[0098] The line buffer 30
A plurality of pixels can be stored in the same manner as the line buffer 10. Then, the value is input to the Markov model creation unit 31, the state generation unit 34, and the like as peripheral pixels,
The reference order table 35 and the signals ST and CX are created.

The entropy decoder 32 decodes the encoded code in the input encoded data 200 by using the Markov state signal CX in a procedure reverse to that of the entropy encoder 12. Note that the entropy decoder 32
Must be formed to perform its operation with an algorithm completely opposite to that of the entropy encoder 12. Therefore, when an arithmetic encoder is used as the entropy encoder 12, the entropy decoder 32 needs to be formed as an arithmetic decoder having a similar configuration.
If a Huffman encoder is used as the entropy encoder 12, it must be configured as a Huffman decoder having the same configuration.

The separating unit 33 converts the data of the converted color palette 18 from the data stream of the encoded data 200,
This is to separate the data of the reference rank table 35 from the data. Then, the synthesizing unit 20 of the encoding system 1 is provided therein.
The reverse function is set, and the entropy encoder 12
The encoded code encoded by the entropy decoder 32
Output to

The state generator 34 has the same configuration and function as the state generator 13 in the encoding system 1 and
The state signal ST is output based on the criteria shown in the table shown in FIG. The reference rank table 35 is configured by data for the reference rank table 35 separated by the separation unit 33,
The table is similar to the reference order table 15 in the encoding system 1. The pixel index generation unit 36 outputs the pixel index 6 from the data of the predetermined order and the Markov state signal CX generated by the encoding system 1 and each data of the reference order table 35.

Next, the operation of the decoding system 3 thus configured will be described with reference to the flowchart shown in FIG.

First, in step S30, the reference pixel and various variables for predictive run length (PRLC) decoding described later are initialized. Next, the data of the reference rank table 35 added to the header of the compressed stream of the encoded data 200 by the separating unit 33 is extracted, and the reference rank table 3
5 is generated (step S31). In step S32, the data of the converted color pallet 18 added to the header of the encoded data 200 to be a compressed stream is similarly extracted and output by the separation unit 33.

Next, at step S33, the reference pixel R
[4] is extracted. This is step S of the encoding process.
It is exactly the same as 14. In step S34, a Markov state signal CX is generated. This is exactly the same as step S16 of the encoding process. Step S
At 35, a decoding order X is obtained by entropy decoding by PRLC.

In step S36, the pixel index generator 36 determines whether or not the rank X is equal to or less than the value of the Markov state signal CX. If the rank X is less than or equal to the value CX,
This indicates that the pixel index to be decoded exists in the reference pixel, and the process proceeds to step S37. Otherwise, the process proceeds to step S38. In step S37, the pixel index 100B of the reference pixel located at the position indicated by the rank X existing in the state segmented by the state signal ST is output.

In step S38, if the decoding order X does not indicate a reference pixel, the decoding order X itself becomes the pixel index 100B to be decoded. However, (CX
Since +1) codes are assigned to the reference pixels,
It is necessary to subtract (CX + 1) from the decoding order X. In step S39, steps S33 to S3 are performed until decoding of all pixel indexes 100B is completed.
7 or step S38 is repeated. Then, when decoding of all pixel indexes 100B is completed,
The decoding operation ends.

Next, the encoding performed by the entropy encoder 12 in the encoding system 1 described above will be described. In this embodiment, what is called a predicted run length (hereinafter referred to as PRLC) is employed. The PRLC is an improvement of the well-known run-length coding. The outline of the algorithm will be described with reference to FIGS.
A basic encoding method and the like that form the basis of C will be described with reference to FIGS. 21 and 22.

The basic algorithm of this encoding is basically the same as that of a conventionally well-known QM coder.
A binary bit string is to be compressed. First, as an initial value, either “0” or “1” is determined as a dominant symbol, and the number run of which the symbol is predicted to be continuous is set. If the appearance probability of the input sequence is unknown, run is preferably set to 1. Then, the encoding is performed according to the following rules. Note that the number run corresponds to the number of predicted bits.

As shown in FIG. 18, all the sequences of interest indicated by run are predicted to be dominant symbols, and when the prediction is successful, "0" is output as a codeword and the coding of this sequence is completed. . If it deviates, "1" is output, and the next division encoding step is executed.

If the prediction is incorrect, the sequence of interest is divided into the first half sequence and the second half sequence as shown in FIG. 19, and if the first half is all dominant symbols, "0" is used as a codeword.
When the inferior symbol exists in the first half sequence and the latter half is all superior symbols, "10" is used as a codeword, and when the inferior symbol exists in both the former and latter half, "1" is used.
1 ". The sequence in which the inferior symbol exists is further divided (see FIG. 20). The sequence in which the inferior symbol exists is divided as much as possible, and the above-described division encoding step is performed. repeat.

Note that the division need not necessarily be two equal divisions, but may be non-uniform divisions or three or more divisions. In addition, when the prediction is successful, a superior symbol is output instead of “0”. When the prediction is incorrect, an inferior symbol is output instead of “1”. "May be output.

The above is the basic algorithm of data encoding which is the premise of this encoding. In order to follow the change of the appearance probability of the input sequence and improve the encoding efficiency,
The following processing may be added.

That is, when the sequence predicted by run hits a predetermined number of times consecutively, for example, twice, run is increased twice or the like. If the prediction continues to be correct, the prediction range is further expanded. On the other hand, the sequence predicted by run includes the inferior symbol twice or more, and the run is halved. If the prediction continues to be missed, it is further increased by a factor of two. Then, when run is 1 and it is the inferior symbol, the subsequent input sequence is inverted. That is, the dominant symbol is changed.

When a less-probable symbol exists in the latter half of the sequence predicted by run, run is immediately changed to 1 /
It may be reduced to 4 or the like. This is because, when the inferior symbol exists in the latter half, it is determined that the following sequence includes many inferior symbols. For this reason, when the inferior symbol exists only in the first half sequence of the sequence predicted by run, a value larger than when the inferior symbol exists in the second half, for example, run may be multiplied by １ ／.

The entropy encoder 12 of this embodiment
21 and FIG. 2 described below.
2 is an improved version of the encoding process. First, the encoding process will be described. The encoding process before improvement includes an encoding main routine of FIG. 21 and an encoding subroutine of FIG. The encoding subroutine in FIG. 22 performs recursive reading of a function that calls the same subroutine from the subroutine.

First, each step of the encoding main routine of FIG. 21 will be described. The encoding target is an input sequence composed of binary bit strings. First, setting of the initial value run of the prediction and selection of the dominant symbol (“0”
Or “1”) (step S50). Next, 0 is substituted for the local variable ofs and run is substituted for the width (step S51). Here, ofs is a pointer of the array A defined in advance for encoding, and indicates a prediction start bit position. Therefore, the initial value is 0. width is o
A value indicating how many bits are to be predicted from the bit position indicated by fs. In this case, the initial value of prediction run is substituted. Then, input bits are written from A [ofs] to A [width-1] of the array A defined in advance (step S52). Then, from A [ofs] to A [w
If all elements of [idth-1] are superior symbols, the process proceeds to step S54. If at least one inferior symbol is included, the process proceeds to step S55.

If the prediction is correct, a signal "0" is output as a code word per prediction, and the encoding of the sequence taken into array A is completed (step S54). On the other hand, if the prediction is incorrect, a mispredicted signal "1" is output as a code word (step S55). Then, it is detected whether the width is 1 or more (step S56). If the width is equal to or less than 1, the image cannot be further divided, so that the process proceeds to step S58 without going to the encoding subroutine of step S7.
On the other hand, if the width exceeds 1, the encoding subroutine in FIG. 22 is called (step S57).

In step S58, the prediction run is reset and, if necessary, the dominant symbol is changed. That is, in this step S58, basically, if the prediction is correct, the run is increased, and if it is off, the run is reduced. If the prediction continues to shift a predetermined number of times even when run is reduced, the dominant symbol is changed. It should be noted that various methods can be adopted as to how to evaluate a hit or a wrong prediction. For example, it is possible to adopt a method in which the run is reduced immediately when the prediction is incorrect, or the run is reduced for the first time when the prediction is incorrectly performed two or more times. Furthermore, a method can be employed in which the degree of run reduction differs between when only the first half sequence or the second half sequence is deviated and when both are deviated. It is also possible to adopt a method of drawing a predetermined probability table with an encoded bit sequence and setting the next prediction run.

If the primary prediction deviates in the encoding main routine, an encoding subroutine shown in FIG. 22 is called in step S57. Arguments to be passed to the encoding subroutine are ofs and width. Hereinafter, each step of the encoding subroutine will be described.

In the encoding subroutine, since the prediction is performed separately for the former half sequence and the latter half sequence, the range of the prediction is halved (step S60). That is, the width received as an argument from the parent routine is halved. Then, in the next step S61, the first half sequence (A [o of the array)
fs] to A [ofs + width-1]) are all checked symbols. If all the symbols are superior, the process proceeds to step S62. If at least one inferior symbol exists, the process immediately proceeds to step S64.

If all the first half sequences are dominant symbols, "0" is output as a codeword (step S62). Then, the pointer ofs indicating the head position of the first half series is set to wi.
dth is added to change to indicate the head position of the latter half series. Further, when the first half sequence is all dominant symbols, since the inferior symbol always exists in the second half sequence, there is no need to output the code word “1” indicating that the prediction of the second half sequence has been lost. Therefore, step S7 described later
0 is skipped and the process proceeds to step S71.

On the other hand, when the inferior symbol exists in the first half sequence, "1" is output as the code word (step S6).
4). Next, it is checked whether the width exceeds 1 (step S65). If the value is equal to or less than 1, the image cannot be further divided, so that the call of the child encoding subroutine (step S66) is skipped, and the process proceeds to step S67. If the width is 2 or more, it is necessary to further divide the sequence into two and encode each of them. The child encoding subroutine for that purpose is called (step S66). The child encoding subroutine is exactly the same as the encoding subroutine shown in FIG. That is,
Here, the same routine (function) is recursively called.

When the encoding of the first half sequence is completed by recursive calling of the encoding subroutine, the width set in step S60 is added to the pointer ofs indicating the start position of the first half sequence, and the start position of the second half sequence is indicated. It is changed (step S67). Thereafter, it is checked whether or not all the latter half sequences (from A [ofs] to A [ofs + width-1] in the arrangement) are all dominant symbols (step S68). If all symbols are superior, step S69
Proceed to. If at least one inferior symbol exists, the process immediately proceeds to step S70. If all of the latter half sequences are dominant symbols, "0" is output as a codeword (step S69).

On the other hand, when the inferior symbol exists in the first half sequence, "1" is output as the code word (step S7).
0). Then, it is checked whether the width exceeds 1 (step S71). If less than 1,
Since the image cannot be further divided, the step S72 of executing the child encoding subroutine is skipped, and the process returns to the encoding process of the next sequence of interest. If the width of the latter half sequence is 2 or more, the sequence is further divided into two and encoded. Therefore, the same child encoding subroutine as the encoding subroutine shown in FIG. 22 is called (step S72). By the recursive call of this encoding subroutine, encoding of the latter half part series is executed.

Next, a specific example of the above-described encoding process will be described. That is, as a specific example of the encoding, the initial value of prediction run is set to 8, the dominant symbol is set to “0”, and “0” is set to “0”.
A case where an input bit represented as 0001001 "is encoded will be described.

First, in step S52 of the encoding main routine of FIG.

The above input bits are input from [0] to A [7]. In step S53, A

It is determined whether all of [0] to A [7] are “0”. In the case of the above example, since "1" is included in the bit string, the process proceeds to step S55, and "1" is first output as a codeword.
Subsequently, in step S56, the size of the width is checked. Since the width is 8 at this time, the process proceeds to the encoding subroutine (step S57).

In the encoding subroutine, first, in step S
At 60, the width is set to 1/2, ie, 4. Then, in step S61, the first half of the input bit, that is, A

[0]
To A [3] are all 0. In this case, since all are "0", the process proceeds to step S62, and "0" is output as a code word. Thus, the encoding of the first half sequence is completed. Subsequently, step S63 is executed, and the process proceeds to the encoding of the latter half part sequence.
Obviously, the latter half series includes "1". Therefore, unless the width is not less than 1 in step S71, the latter half sequence must be further divided and encoded. Therefore, the encoding subroutine is called again in step S72 as a child process. As a pre-process for that purpose, in step S63, the width is added to ofs, and the ofs is set to the head position of the latter half series, as described above.

In step S72, ofs and width are set.
Is called as an argument, and the child encoding subroutine is called. In step S72 of executing the child encoding subroutine,
First, step S6 of the encoding subroutine shown in FIG.
At 0, the width is further halved and changed to 2. In the next step S61, it is checked whether or not both the first half series, that is, A [4] and A [5] are both "0". In this case, since A [4] is "1", the next step S64
And outputs "1" as a code word. Then, in step S65, it is determined that the width exceeds 1, and the grandchild process is called in step S66. In the grandchild encoding subroutine, first, in step S60, the width is
Becomes 1. Since A [4] is "1", the process moves from step S61 to step S64, and outputs a code word "1". In step S65, since the width is 1 or less,
Step S66 is skipped.
To 5. Since A [5] is "0", step S6
From 8, the process proceeds to step S <b> 69 to output a code word “0”.

Next, the process exits from the grandchild encoding subroutine and returns to step S67 of the child encoding subroutine. Ofs of child encoding subroutine is 4, width
Is 2, the value ofs is changed to 6 in step S67. Therefore, in step S68, A [6] and A
[7] will be checked. In this case, A [7]
Is "1", the flow shifts to step S70, where the code word "1"
Is output. Then, the grandchild encoding subroutine is called again in step S72. In the grandchild encoding subroutine, since A [6] is "0", the code word "0" is output in step S62. Since the width is 1, the process skips step S72 and returns to the child encoding subroutine.

The process that has returned to the child encoding subroutine further returns to the encoding main routine, and resets the predicted run and the dominant symbol in step S58. In this example, the primary prediction was missed, but the first half was hit in the secondary prediction, so run is changed from 8 to 4, and the process of continuously setting the dominant symbol to "0" is performed.
Note that the setting of the predicted run may be changed, for example, when the prediction run is missed twice.

By such an encoding process, the input bit “000000101” is changed to “101101”.
The encoded sequence is 0 ". Therefore, in this case, the input sequence of 8 bits is compressed to 7 bits.

In the decoding process, an input codeword is decoded by an algorithm reverse to that of the encoding process. That is, the decryption process also
It consists of a decoding main routine and a decoding subroutine, and performs decoding by an algorithm opposite to the encoding.

As described above, in the encoding process shown in FIGS. 21 and 22 and the decoding process performed using the algorithm reverse to the encoding process, the compression ratio is at the same level as that of a conventional QM coder. Yes,
The encoding time and the decoding time are greatly reduced. However, in the encoding process and the decoding process, a bit string determined by the number of predicted bits, run, is encoded at a time. Therefore, in order to increase the compression ratio, if the maximum value of run is set to be large, the number of run There is a problem that a buffer for storing the minute bits becomes large.

This problem becomes more serious when the Markov state signal CX is obtained from Markov modeling, because the buffer becomes very large. That is, if run is assumed to be n, n buffers are required for the buffer, and if the Markov model of m states is further performed, the buffer is required for each state, so that the capacity becomes n × m bits. This capacity becomes a size that cannot be ignored as the value of run increases.

In the encoding process shown in FIGS. 21 and 22 and the decoding process using an algorithm reverse to the encoding process, each processing time is equal to QM
Although it is much shorter than the coder, it performs encoding and decoding by recursively calling the encoding and decoding subroutines, and this recursive call of the subroutine has time. I have.

Therefore, in the present embodiment, it is possible to reduce the size of the decoding buffer and to further reduce the encoding and decoding time while making use of the encoding and decoding processes shown in FIGS. A data encoding method and the like are adopted. Hereinafter, the basic concept of the operation of the index conversion units 7 and 8 in the embodiment of the present invention will be described.
This will be described with reference to FIGS.

First, the entropy encoder 12 used in the embodiment of the present invention will be described with reference to FIG.

The entropy encoder 12 is an entropy encoding device, and includes a bit string decomposition section 22 for inputting a binary bit string to be encoded, and a prediction bit length r
Coding table unit 2 with a built-in coding table for each un
3, a stream generation unit 24 for temporarily buffering the variable-length code input from the encoding table unit 23 to output a fixed bit width, and a state transition table described later. And the like.

Here, a state transition unit 26 having a state transition table and an encoding condition generated by a Markov model or the like are input to the encoding control unit 25, and the signal of the current state is converted into a state for each condition. There is provided a state storage section 27 which is provided to the transition section 26, receives a signal of the next state after encoding, and stores the state of the encoding.

Therefore, when performing conditional coding such as the Markov model, the condition is used as an index and
The corresponding state is taken out from the state storage unit 27, and the state is transited according to a state transition table shown in FIG.
If the next state is stored in the original address of the state storage unit 27 again, the state can be managed for each condition. Therefore, parameters such as the predicted bit length run can be individually set for each condition. In the case of using Markov models, the bit string decomposing unit 22 requires a plurality of buffers to be switched for each encoding condition. However, in this embodiment, not a number of predicted bit lengths run but a smaller fixed delimiter Since the coding is performed stepwise with the number of bits p, the capacity of the buffer does not increase so much, which is suitable for practical use.

The bit string decomposing unit 22 includes an encoding control unit 25
, A signal RUN indicating a prediction bit length run and a signal SW indicating a dominant symbol are input. Here, the signal RUN takes a value from 1 to n (n is the maximum predicted bit length). When the value of the signal SW is “0”, “0”
Is the dominant symbol, and when it is "1", "1" is the dominant symbol, but vice versa.

Further, the bit string decomposing unit 22 outputs to the encoding table unit 23 a signal DECNUM of the number of bits to be decoded and a signal pattern DECPATN which is a pattern of the bits to be decoded. Signal DECNUM
Is the total number of 4 bits and the number of superior symbols that have continued up to that point when a 4-bit pattern including inferior symbols appears in the input bit string. When the signal RUN is less than “4”, the same value as the signal RUN is output. This is because the number of delimiter bits p is set to “4” in this embodiment.

As described above, the bit string decomposing unit 22
When the dominant symbol specified by the signal SW continues for the input bit string for the number of bits specified by the signal RUN, that is, when the prediction is successful, the signal DECNU
The value of the signal RUN is set as the signal pattern DECPAT.
"0" is output as N.

The encoding table unit 23 is shown in FIG.
7, the table to be used is selected by a table number instruction signal TABLE from the encoding control unit 25. The encoding table unit 23 searches a predetermined table based on the signal DECNUM and the signal pattern DECPTN from the bit string decomposing unit 22, and searches for a predetermined compressed bit string DEC.
The BIT, its bit length LENGTH, and FAIL indicating a prediction hit or miss are output. In addition, the signal TABLE
Has a one-to-one relationship with the signal RUN.

In the encoding table of FIG. 24, the table number is “0” and the value of signal RUN is “1”, that is, run
Is "1". As shown in FIG. 24, when run is "1", there are two types of signals. That is, the number of bits to be decoded is one,
There are two types of signal patterns, "0" and "1". This 2
Compressed bit string DECBIT
And two types of signals corresponding to the combination of the bit length LENGTH and the flag FAIL indicating a column out of prediction. For example, when the signal DECNUM is “1” and the signal pattern DECCATN is “0”, a prediction hit occurs, the flag FAIL becomes a hit signal “0”, the compressed bit string DECBIT becomes “0”, and the bit length LE
NGTH is "1".

The encoding table of FIG. 25 shows a case where the table number is “1” and run is 2. The signal pattern DECPTN and the compressed bit string DECBIT of each encoding table both indicate signals input from the right to the left. In the case of FIG. 25, the signal forms are four types. The number of bits to be decoded is all two, and the signal pattern DECPTN at that time is “0”.
0, 10, 10, 01. When the signal pattern DECPTN is "00", both of the symbols are the predominant symbols, so that the prediction was successful, and the flag F
AIL becomes "0" of the hit signal, and the compressed bit string DECBIT at that time becomes "0", and the bit length L
ENGTH becomes “1”.

On the other hand, when the signal pattern DECPTN is "1"
When the value is "0", the inferior symbol "1" is included, which means that the prediction has been missed.As a result, the flag FAIL becomes "1" of the missed signal, and the compressed bit string DECBIT becomes
"1" comes first. Next, the first half of “10” is “0”
, And the second bit of the compressed bit string DECBIT becomes “0” and becomes “01”. Here, since the prediction is initially unsuccessful, there is "1" in the latter half. For this reason, this “01” is used as it is in the compressed bit string DECBIT.

When the signal pattern DECPTN is "01", the inferior symbol "1" has been entered, which means that the prediction was incorrect. As a result, the flag FAIL becomes the departure signal “1”, and the compressed bit string DECBIT first comes “1”. Next, since the first half of “01” is “1”, the prediction is again lost, and the compressed bit string D
The second of the ECBIT is "1". Signal pattern DE
Since the latter half of CPUTN “01” is “0”, it is a prediction hit and the third of the compressed bit string DECBIT is “0”. That is, the signal pattern DECPTN
The compressed bit string DECBIT corresponding to “01” is “0”.
11 ". The bit length LENGTH is" 3 ".
Becomes Similarly, the signal pattern DECPTN “1”
The compressed bit string DECBIT for "1" is "111".
Becomes FIG. 20 shows a correspondence table of the above four types of signals.

Similarly, if the table number is “2” and r
FIG. 26 shows the correspondence relationship between 16 types of signals whose un is “4”, and FIG. 27 shows the correspondence relationship of a total of 31 types of signals whose table number is “3” and run is “8”. When the run in FIG. 26 is "4", the value of run is the same as the number of delimiter bits p, so that only the same relationship as in FIGS. 24 and 25 is obtained. In the case of "8", the number of delimiter bits is larger than p (p = 4 in this embodiment), so that the table is slightly modified.

Next, the contents of the encoding table of FIG. 27 slightly different from the other tables will be described. In this encoding table, the number of bits DECNUM of the signal to be decoded is:
There are "8" and "4". In the case of “8”, the first half is all “0000” and “4”
Indicates a case where the run is "8" and the inferior symbol "1" comes in the first half. The number of bits DECNUM (hereinafter simply referred to as DECNUM) of a signal to be decoded is “8”, and the signal pattern DECPTN
(Hereinafter simply referred to as DECPTN) is "0000"
Indicates that it is “00000000”, which means that the prediction was successful, and the flag FAIL (hereinafter simply referred to as FA
IL) is "0" of the hit signal. Then, the compressed bit string DECBIT (hereinafter simply referred to as DECBIT
) Is “0”, and the bit length LENGTH (hereinafter simply referred to as LENGTH) is “1”. DEC
When NUM is "8" and DECPTN is "1000", it indicates "10000000", which means that the prediction has been lost, and FAIL becomes "1" of the disconnection signal. Then, "1" comes at the first of the DECBIT.
Next, the first half “0000” becomes a prediction hit, and DEC
"0" comes at the second of the BIT. At this time, since the inferior symbol “1” naturally comes to the latter half “1000”, the DECBIT for the four signals in the latter half is:
No particular problem occurs.

The first half “00” in the second half “1000”
Is per prediction, and the third DECBIT is “0”
Becomes At this time, since the inferior symbol “1” naturally comes to the latter half “10”, DECBIT for the two signals in the latter half is not particularly generated. Then, the first half “0” of this second half “10” becomes a prediction hit, and 4
The third DECBIT is "0". In this case, it is natural that there is a “1” at the end, and especially DECBIT
Does not occur. Therefore, DECBIT corresponding to DECPTN “1000” is “0001”. And
LENGTH is "4". This corresponds to the second state from the top of the table with the table number “3” in FIG.

Such a relationship also applies to the third to sixteenth encoding tables in FIG. On the other hand, from the 17th to the third from the top of the table number "3"
Up to the first, DECNUM is "4", which is similar to that of table number "2" in FIG. That is, FIG.
In the second to sixteenth encoding tables of the encoding table No. 1, “1” which is a misprediction when the run is viewed as “8”
Are added first, and the DECNU in FIG.
It is the same as that of M “4”. The code output from the encoding table unit 3 is a variable length code specified by LENGTH.

In this embodiment, run is 16
And a table with 32 run. The two tables have the same concept as the table number “3” of which run is 8 shown in FIG. 27, and a description thereof will be omitted.

The stream generating section 24 temporarily buffers the input variable length code, and outputs the buffer after arranging it to a fixed bit width determined by the output transmission path.

The basic operation of the encoding control unit 25 is as follows.
N (hereinafter simply referred to as RUN), the bit string decomposing unit 2
2 to indicate the bit extraction method, and at the same time, the signal TAB
The coding table is selected by LE (hereinafter simply referred to as TABLE). Then, the RUN and TABLE for the next encoding are set by the FAIL fed back from the encoding table unit 23. In this embodiment, stepwise encoding using the number p of delimiter bits is introduced. Therefore, when encoding is performed with a certain prediction bit length run, this encoding control may be performed at an intermediate stage as necessary. The unit 25 needs to memorize.

The specific operation of the encoding control unit 25 is as follows.
This is based on the state transition table shown in FIG. The operation of the state transition table will be described by taking a case where prediction hits are continued as an example. Here, the initial state is SS1. First, in the state SS1, run is “1”,
TABLE is “0”. Therefore, the encoding table of table number “0” shown in FIG. 24 is used. When the prediction is successful, the dominant symbol is “0”, which means that the bit string input from the input bit string
Only “0” of “1”, which is the number of CNUMs, is sent to the encoding table unit 23, and FAIL “0” and DE are determined based on the table of table number “0” (= table in FIG. 24).
CBIT “0” and LENGTH “1” are output. Then, the FAIL “0” is transmitted to the encoding control unit 25.

The coding control unit 25 finds a FAIL “0” in SS1 based on the state transition table of FIG. 28, and selects the state SS0 as the next state (see the state transition table of FIG. 28). Third). At this time, since the signal SW becomes "0", there is no inversion of the symbol, and "0" becomes the dominant symbol as it is. Also in the state SS0,
As a result of the same operation, the first state transition table is selected,
State SS3 is the next state. This means that the prediction has been made twice.

In this state transition table, if prediction is successful twice, run is doubled. That is, the seventh and eighth states SS3 from the top are set, and the run becomes "2". If the prediction continues to be hit in this way, that is, if the input bit string continues to be “0” in this case, the run increases to “2” “2” “4” “4” “8” “8” Go.
On the other hand, if the prediction continues to be incorrect, the same r
It becomes smaller with un. That is, the run becomes smaller as “8” “8” “6” “6” “4” “4” “2” “2”. And when run is "1",
If the prediction is incorrect, the signal SW is inverted.

The rules for the operation of such a state transition table can be summarized as follows.

(1) When prediction with the same prediction bit length run is hit twice consecutively, the prediction bit length run is doubled.

(2) When the prediction with the same prediction bit length run is missed twice consecutively, the prediction bit length run is reduced by half.
Multiply.

(3) If the prediction bit length run is 4 or less, the encoding is performed once.

(4) When the predicted bit length run is 8, and the DECN
When UM = 4, the encoding is performed twice.

(5) At this time, the state transits to the state SS5 and
The prediction bit length run is set to “4”, and the latter bit is encoded.

The inversion of the signal SW means that when this value is 1, the signal SW is inverted.

It should be noted that the state transition table shown in FIG.
Indicates only up to “8”, but in this embodiment, since run is set to a maximum of “32”, run “1”
6 "and run" 32 "are similarly created, though not shown. In addition, as the state transition table, run
May be "64" or more. Further, if the hit or miss continues twice, the run may not be increased or decreased, but may be changed each time, or may be three or more times, or various patterns may be adopted. Also, as such an encoding table, only an encoding table having a small number of bits may be prepared, and if the number of bits is large, for example, 16 bits or more, the encoding table may not be provided.

Next, the operation of the entropy encoder 12 having the above configuration will be described using a specific example.

For example, if the prediction continues and run = 1
6 and “00000001000011110”
When encoding a bit string input in the form of “0...”, It is divided by a 4-bit delimiter bit number p, and first, up to “0000100100” is encoded. This is because the inferior symbol “1” does not exist in the first delimiter bit number (= first 4 bits) portion, but appears in the second (= next 4 bits). And
Next, “0011” is encoded, and finally “1100” is encoded.

Therefore, the DECNUM output from the bit string decomposing unit 22 is “8”, “4”, and “4”. On the other hand, DECPTN is “0100” “0011” “1”
100 "(in this case, all patterns are assumed to be input from the left numerical value). Under such conditions, DECBIT firstly becomes" 1 ", which is a disappointment when run = 16. Next, "00000100"
Is RUN "8", TABLE "3", DECNUM
Since it is “8”, it corresponds to the fifth from the top shown in FIG. 27 (note that in the case of each numerical value shown in FIG. 27, it is input from the right end side), and DECBIT is “1”.
Therefore, DECBIT and LENGTH “6” of “110100” (the numerical values are output in order from the left end) and “1” are output from the encoding table unit 23 to the stream generation unit 24. You.

On the other hand, in the state transition table in the encoding control section 25, if the state SS6 is DECNUM “8”, the FA
It becomes IL "1", and the next state is state SS7.
Becomes Then, the next “0011” is run = 8 and D
Since ECNUM is "4", the table with the table number "3" (= the encoding table in FIG. 27) is adopted, and the nineteenth table from the top corresponds to the DECN of "11011".
UM and LENGTH "5" are output from the code table 3.

As for the last “1100”, the previous state is run “8” and DECNUM “4” in the state SS7, and
Since FAIL has become “1” (the lowest state in the current state code “0111” in FIG. 28), the state SS5 is adopted. Therefore, run “4” and TABLE “2” are used, and the encoding table with the table number “2” shown in FIG. 26 is used. In the table with the table number “2”, the fourth from the bottom corresponds, and the DECBIT of “11110” and LENGTH “5” are output from the encoding table unit 23 to the stream generation unit 24. In addition,
In the state SS5, the FAIL becomes “1”, so that the process proceeds to the state SS2. That is, for the next input bit string, the encoding table of FIG. 25 where run = 2 is used.

To summarize the above, the input bit string “000”
0010000111100 "becomes" 110100 ",
That is, it is encoded as three compressed bit strings “11011” and “11110”. It is assumed that the input bit string and the three compressed bit strings are input and output from the head side. It should be noted that this point is different from each of the encoding tables of FIGS.
That is, in each encoding table, each value of the display is:
Input and output are performed sequentially from the right end of the display.

The run is initially “16”, but while the encoding is performed step by step with the four-bit delimiter bit p, the run becomes “2”, and the run is Will be encoded with a prediction bit length run of “2”.

On the other hand, in the basic process which is the basis of the present embodiment shown above, the same input bit string “0000010” is used.
000111100 ", first, run = 1
Attention is paid to the mispredicted "1" at 6 and then the first 8 bits, and the second mispredicted "1" comes, and the first 4
Paying attention to bit “0000”, “0” per prediction is 3
Come th. Then, since it is certain that the inferior symbol will come in the latter 4 bits “0100”, it is immediately divided into two and the first 2 bits “01” are noted. For this reason, the unpredicted "1" comes fourth. Next, this is 2 more
Attention is paid to “0” in the first half, and “0” per prediction comes fifth. Then, since the inferior symbol is certain in the latter half “1”, immediately notice the latter two bits “00”,
“0” per prediction comes in sixth.

The above first eight bits are coded as “110100”. This is exactly the same as the coded bits according to the present embodiment. When the subsequent 8 bits are advanced in a similar manner, these become the same as the encoded bits according to the present embodiment. The difference between the basic process that is the basis of the present embodiment and the present embodiment is not the coded bits themselves, but the following: 1.How to separate coding, 2.How to change prediction bit table, 3. There are three points in using the encoding table.

That is, in the improved embodiment, the number of delimiter bits p smaller than run
(In this embodiment, p = 4), and the encoding is temporarily divided up to the delimiter where the inferior symbol exists. In the above example, the input bit sequence of 16 bits is 3
The encoding is divided into two. Also, in the present embodiment, the predicted bit length run is “2” for the next input bit string, whereas the basic process concept is
There is only one disappointment, and run remains at “16”. Furthermore, in the concept of the basic process of the present embodiment, the encoding is performed by recursively calling the encoding subroutine. However, in the improved encoding method of the present embodiment, the encoding table, An encoding table is used for each bit length run.

The above three points produce a great effect when they are used at the same time, but they are sufficiently effective when used alone. For example, if the method of stepwise encoding of the first point is adopted, not only the buffers such as the buffers in the bit string decomposing unit 22 and the stream generating unit 24 can be reduced, but also the compressed bit string can be obtained by Markov modeling described later. In such a case, the capacity of the buffer can be reduced.

The change of the predicted bit length run at the second point is particularly effective when the characteristics of the input bit sequence change from the middle. In the above example, when the prediction continues to be run and run = 16, next, when a bit string whose property has changed drastically, that is, “000000010000111100” including many inferior symbols comes, the improved embodiment of the present embodiment is performed. In the encoding method, run is set to “2” according to the property, the probability of matching the property of the subsequent input bit string is high, and the compression ratio is high. However, when processed by the basic process, the run remains at “16”, which has a high probability that it does not match the properties of the next input bit string. In addition, the improvement of the compression ratio
Specifically, it is about 0.5% to several% compared to before the improvement. However, at present when each program software and the like has a large capacity, such a slight improvement in the numerical value cannot be ignored. It has become.

As for the encoding table of the third point, although the memory capacity for the encoding table is slightly increased as compared with the encoding by the recursive call of the subroutine, the encoding speed is extremely high.

Next, the entropy decoder 32 of this embodiment will be described with reference to FIG.

The entropy decoder 32 includes a stream cutout section 41 for inputting a stream of an encoded signal, a decoding table section 42 containing a plurality of decoding tables corresponding to a prediction bit length run, and a decoded bit stream. And a decode buffer unit 43 that outputs a predetermined symbol and a state conversion unit 26 of the entropy encoder 12.
And a decoding control unit 44 having the same state transition table as the state transition table. Here, the decryption control unit 4
4, a state transition unit 45 having a state transition table and a decoding condition generated by a Markov model or the like are input, and a signal of the current state is given to the state transition unit 45 for each condition, and after decoding, the next state , And a state storage unit 46 for storing the state of the decoding condition.

The decoding buffer 43 receives decoding conditions and is individually managed for each condition. Therefore, in the case of conditional decoding such as a Markov model, a very large buffer is required. However, the entropy decoder 32 of the present embodiment
In this case, since the decoding is performed stepwise as will be described later, each buffer can sufficiently cope with a small buffer, and the decoding buffer unit 43 does not require a large capacity even in conditional decoding such as a Markov model.

The stream cutout unit 41 calculates the number of decoded bits from the decoding table
H, the decoded bits are discarded based on the value, and the head of the undecoded bits is the lowest bit of the code word signal CODE (hereinafter, simply referred to as CODE) to be decoded, which becomes encoded data. Cut out the stream so that it is at the bit (or most significant). It should be noted that the LENGTH is evaluated and the decoded bits are discarded because the discard instruction DECREQ (hereinafter simply referred to as DE
CREQ). Also, CO
DE is transmitted in units of 8 bits.

The decoding table section 42 is shown in FIGS.
Each decoding table as shown in FIG.
Are switched and used by a table number indicating signal TABLE (hereinafter simply referred to as "TABLE") output from the CPU. Then, the decoding table section 42 outputs the next signal. That is, (1) a signal LE indicating how many bits have been decoded
NGTH (hereinafter simply referred to as LENGTH), which corresponds to LENGTH in the entropy encoder 12;
(2) A signal FAIL (hereinafter simply referred to as F
AIL), and F in the entropy encoder 12
AIL equivalent, (3) decoded bit pattern signal DECPTN (hereinafter simply referred to as DECPTN)
Which corresponds to DECPTN in the entropy encoder 12, and (4) a signal D indicating the number of bits of the decoding result.
An ECNUM (hereinafter simply referred to as DECNUM) which is equivalent to DECNUM in the entropy encoder is output.

The decoding table of run = 1 shown in FIG. 30 describes each output corresponding to two types of CODE of "0" and "1". This decoding table corresponds to ru in FIG.
What corresponds to the encoding table of n = 1, and which corresponds to DECBIT in the encoding table is CODE in this decoding table. Run = shown in FIG.
Similarly, the decoding table of 2 corresponds to the encoding table of run = 2 in FIG. FIG. 32
In the decoding table of run = 4 shown in FIG.
= 4, and the run table shown in FIG.
The decoding table of 8 corresponds to the encoding table of run = 8 in FIG. Each numerical value in each decoding table is input and output from the right end side of each numerical value, similarly to the encoding table. Also, tables of run = 16 and run = 32 are prepared as decoding tables.

The decode buffer unit 43 has DECPATN and DECNUM of 4 bits or less (in this embodiment).
Are directly stored, and the decode buffer 43
PANTREG inside (hereinafter simply PANTREG)
And number register NUMREG (hereinafter simply NUMRE)
G). When the output of the decode buffer unit 43 has a q bit width, the decode buffer unit 4
3 subtracts q from NUMREG stored each time decoded data is output. When NUMREG becomes smaller than q, DECREQ is activated, and a request for decoding new data is issued. NU
When MREG is 5 or more, a dominant symbol determined by the signal SW is output as a decode output. On the other hand, NUMRE
When G becomes 4 or less, the value of PATHREG is output.

For example, the fifth CODE from the top in FIG.
When “00101” is input to the decoding table unit 42, DECNUM = 8 and DECCATN = “0010”
Is input to the decode buffer unit 43. At this time, assuming that the signal SW is “0” and the output is performed in units of 2 bits (this corresponds to q = 2), the first two outputs may output the superior symbols. In this case, since SW = 0, the dominant symbol is “0”. Therefore, "0000" is output. When these 4 bits are output, NUMREG is 4 (= 8−4). Then, in the next cycle, the value of PATHNREG is
Output in order. That is, "0100" is output from the left end side of this display.

The state transition unit 45 of the decoding control unit 44 has the same state transition table as the state transition unit 26 of the encoding control unit 25. Then, the initial value of the state is SS1, and F
The next transition destination is determined by AIL and DECNUM,
When DECREQ is active, transition is made to that transition destination.

The entropy decoder 32 configured as described above operates according to an algorithm reverse to that of the entropy encoder 12 described above. The entropy decoder 32 is controlled by the output state of the decode buffer 43. That is, when NUMREG of the decode buffer unit 43 becomes smaller than the output bit width q, DECREQ is output from the stream cutout unit 41.
Is output to the decoding control unit 44. The stream cutout unit 41 discards the decoded bits by the LENGTH according to the DECREQ.

In the above example, when run = 8 and CODE “0010”
In the case of 1 ", NUMREG changes from" 8 "to" 4 ",
It falls from "4" to "2" and from "2" to "0". When the value falls from "2" to "0", DECREQ occurs. Since LENGTH is “5”, C
Discard the decoded 5 bits from the ODE. For this reason,
In the CODE in the stream cutout unit 41, the undecoded bits come to the lowest order or the highest order to prepare for the next decoding.
On the other hand, in the decryption control unit 44, run = 8, DECNUM
= 8 and FAIL = 1, transit to state SS7.
Therefore, TABLE = 3 corresponding to run = 8 is output to the decoding table unit 42.

As a result, the decoding table section 42
A decoding table of run = 8 with the table number “3” is prepared. Then, from the input CODE to LEN
GTH, DECNUM, DECPTN and FAIL
Is determined and output. For example, if the beginning of the CODE is “0”, it is determined that the CODE is “0”,
LENGTH = 1, DECNUM = 8, DECPTN
= “0000” and FAIL = “0”. on the other hand,
When the CODE is “01011”, since the beginning of the CODE is “1”, the CODE is not yet determined, and even if the next “1” is 3
Neither the fourth “0” nor the fourth “1” is determined. However, at the stage when the fifth “0” is entered, it is determined that it is “01011”. With this confirmation, LENGTH
= 5, DECNUM = 4, DECPTN = “010
0 "and FAIL =" 1 "are output. In this way, decoding is performed sequentially.

The entropy decoder 32, like the entropy encoder 12, has a buffer capacity reduced by stepwise decoding as compared with the decoding based on the basic process described above. Long run
Has various advantageous effects of improving the decoding speed by the modified decoding table.

In the entropy decoder 32,
It is similar to the entropy encoder 12 in that the state transition is managed individually for each condition. However, in the case of the entropy decoder 32, the decode buffer unit 43 must also be individually managed as described above. For this reason, the decode buffer unit 43 includes NUMREG, PAT
Registers equivalent to NREG are provided for the number of possible conditions, and are switched according to decoding conditions.

In the above description, 2
The prediction run-length encoding method has been described by taking the bit string of the value as an example. However, the rank code output from the determination unit 19 is multi-valued data, and it is actually necessary to convert the multi-valued data into a binary sequence. For example, dividing into bit planes,
Each bit plane may be encoded using this predictive run-length encoding method. Alternatively, encoding may be performed for each plane starting from the most significant bit using this predictive run-length encoding method, and the subsequent lower bits may be directly output to the stream when "1" appears.

In this embodiment, as a method of applying this PRLC to a multilevel sequence, the PRLC is divided into a level plane instead of a bit plane. More specifically, since the symbols are 8 bits, the symbols are divided into 256 level planes, the input symbols are divided into groups, and the group numbers are encoded by this predictive run-length encoding method. That is, as shown in FIG.
Grouping is performed, and first, a decision bit indicating whether the input symbol is a group number 0 or other than 0 is encoded by this PRLC. If the input symbol is 0, the encoding of this symbol is completed.
A decision bit indicating whether the bit is a value other than 1 is encoded by the predictive run-length encoding method.

In this way, the decision bits are encoded by PRLC until the group number is determined. If the determined group number is 2 or more, the necessary additional bits are directly added to determine the symbol in the corresponding group. Output to stream. However, each group judgment bit is
Each group is treated as an independent sequence, and the current status code is individually managed and encoded. In this method, when the group number is determined, the higher-order determination bits are not encoded, so that the processing speed is improved.

In this embodiment, the number of pixel indices 100A of the pixel to be compressed is 256 in addition to 256.
Since (CX + 1) is added, a maximum of 260 items are targeted, and the order is 259. As a measure against this,
The rank 255 is assigned to an escape code, and for ranks of 255 or more, after encoding this escape code, the difference from 255 is directly output to the compressed stream as 3-bit additional bits. Such a measure can be applied even when the pixel index 100A to be compressed slightly exceeds 256 colors.

The index palette of the multi-color image in this embodiment is shared between different frames, and the data amount for the palette does not increase. That is, the index palette of the moving image is used as a common palette in the moving image stream. Further, in the encoding using the inter-frame correlation, a common palette is used within the range of the inter-frame correlation. For example, a palette is provided for each I-picture (GOP), or in another range.

Also, in the entropy encoder 12 of this embodiment, when encoding, the compression ratio is improved by appropriately setting conditions based on past encoded sequences and the like. That is, PRLC encoding is executed by performing condition classification based on the Markov state signal CX. Specifically, the Markov state signal CX has four states and one
Since eight group determination bits are individually encoded in one state, a total of 32 state current state codes are individually managed and encoded.

Although the above-described embodiment is an example of a preferred embodiment of the present invention, the present invention is not limited to this embodiment, and various modifications can be made without departing from the gist of the present invention. . For example, the color pixel data 100A may be configured to encode and decode n-bit (n is an integer of 2 or more) color pixel data 100A.

Also, instead of multi-color images, RGB
It can be applied to a natural image having a component or a YUV component directly. In that case, the natural image conversion means 2a becomes unnecessary, or the natural image conversion means 2a and the YUV conversion means 2b become unnecessary. In the above-described embodiment, the case of a moving image using motion compensation has been described. However, the preprocessing unit 2 and the correcting unit 9 can be applied to a moving image or a still image without motion compensation. Further, instead of palletizing the UV components, the UV components thinned out by the thinning means 2c may be indexed, or may be coded by the direct coding means.

In performing the encoding and decoding, the method before improvement shown in FIGS. 18 to 22 may be employed, or a conventional arithmetic code type entropy encoder and entropy decoder may be used. Also, instead of performing pre-scan as in the above-described embodiment, Japanese Patent Laid-Open No.
The technology disclosed in Japanese Patent No. 6041 may be used. That is, the color ranking data obtained by rearranging the indexes in the order of appearance frequency may be used, or the latest appearance table obtained by performing the moving process of moving the latest one to the top may be used.

Furthermore, when the head moving process is used, the appearing color symbol may not always be brought to the head, but may be moved in the head direction according to a predetermined rule. In this way, when a color symbol with a low appearance probability appears, the color symbol does not immediately move to the head, so that a short code is not assigned to a symbol with a low appearance probability. For this reason, even if an object having a low appearance probability appears, the encoding efficiency and the decoding efficiency do not deteriorate.

The double judgment based on the macro block MB and the small block SB can be applied to the motion compensation of an image other than a multi-color image, for example, a natural image. Also, the method of picking up macroblocks MB for which motion compensation is not performed is to collect vertically when frame scanning is horizontal, and to collect horizontally when frame scanning is vertical. preferable.

The blocks for performing motion compensation have various sizes such as 8 × 8 pixels, 4 × 4 pixels, and 32 × 32 pixels in addition to the generally employed 16 × 16 pixels. However, considering the amount of data for a motion vector, it is preferably 4 × 4 pixels or more, and considering effective motion detection, it is preferably 64 × 64 pixels or less.

Further, in the above-described embodiment, the encoding and decoding are performed by the system configured as hardware, but may be entirely performed by software. When processing by software, the processing procedure is programmed and installed in a storage medium such as a CD-ROM, and the storage medium is inserted into hardware such as a personal computer or the like, and the program is transmitted or received via a network or the like. Alternatively, the data may be imported to a hard disk such as a personal computer and processed.

[0207]

As described above, the color image encoding apparatus and method according to the present invention can reduce the encoding time by encoding a color image without using DCT. According to another aspect of the present invention, when a moving image composed of a color image is encoded, motion compensation can be performed by sufficiently considering a shadow portion of the image.

In the color image decoding apparatus and method according to the present invention, decoding time can be reduced by decoding a color image without using DCT. When decoding a moving image composed of a color image, it is possible to perform a motion compensation in which a shadow portion of the image is sufficiently considered.

[Brief description of the drawings]

FIG. 1 is a diagram showing a color image encoding apparatus and an encoding system employing the method according to the present invention.

FIG. 2 is a diagram showing a configuration of an index conversion unit and its periphery in the encoding system of FIG. 1;

FIG. 3 is a diagram showing a color image decoding apparatus and a decoding system employing the method according to the present invention.

FIG. 4 is a diagram showing a configuration of an index conversion unit and its periphery in the decoding system of FIG. 3;

5A and 5B are diagrams for explaining an operation result of the motion compensation unit of the encoding system in FIG. 1; FIG. 5A is a diagram showing types and input orders of pictures (frames); FIG. 3 is a diagram showing an encoding order.

FIG. 6 is a diagram showing macroblocks and small blocks employed in the encoding system of FIG. 1 and the decoding system of FIG. 3 and their relationships.

7A and 7B are diagrams for explaining an arrangement of reference pixels adopted as a Markov model in the encoding system and the decoding system in FIGS. 1 and 3, where FIG. 7A illustrates a normal reference pixel, and FIG. FIG. 4C is a diagram illustrating a case of a leading raster, and FIG. 4C is a diagram illustrating a case of a leading pixel.

FIG. 8 shows a state of a reference pixel employed in the encoding system and the decoding system shown in FIGS. 1 and 3, and a state signal ST.
FIG. 7 is a diagram illustrating a relationship between the number of colors, a number of colors, and a Markov state signal CX.

9A and 9B are diagrams for explaining a function of generating a reference order table of a conversion table generation unit employed in the encoding system of FIG. 1; FIG. 9A illustrates a frequency when a pixel to be encoded matches a reference pixel; (B) and (C) are diagrams illustrating the order of reference pixels when the respective frequencies have a predetermined relationship.

FIG. 10 is a diagram showing an example of a reference order table employed in the encoding system and the decoding system of FIGS. 1 and 3;

11 is a diagram for explaining a pixel index conversion table creation function of a conversion table generation unit employed in the encoding system of FIG. 1; FIG. 11 is a diagram illustrating the frequency with which a pixel corresponding to an input pixel index is not present in a reference pixel; FIG.

12 is a diagram for explaining a function of creating a pixel index conversion table of a conversion table generation unit employed in the encoding system of FIG. 1, and is a frequency at which a pixel corresponding to an input pixel index is not present in a reference pixel; FIG.

13 is a diagram for explaining conversion from an input color palette to a conversion color palette in the encoding system of FIG. 1;

FIG. 14 is a flowchart showing an operation at the time of prescan in the encoding system of FIG. 1;

FIG. 15 is a flowchart showing an operation during an encoding process in the encoding system of FIG. 1;

16A and 16B are diagrams for explaining a method of encoding a macroblock determined not to perform motion compensation in the encoding system of FIG. 1; FIG. 16A illustrates an example of a position of a macroblock not performing motion compensation; (B) is a partially enlarged view of FIG.

FIG. 17 is a flowchart showing a decoding process in the decoding system of FIG. 3;

FIG. 18 is a diagram for explaining an outline of an algorithm used in an entropy encoder and an entropy decoder employed in the present invention.
FIG.

FIG. 19 is a diagram illustrating a state where the attention sequence in FIG. 18 is divided.

FIG. 20 is a diagram showing a state in which the first half attention sequence in FIG. 19 is further divided.

21 is a flowchart for explaining an encoding process in the entropy encoder shown in FIGS. 18 to 20, and is a flowchart showing an encoding main routine.

FIG. 22 is a flowchart for explaining an encoding process in the entropy encoder shown in FIGS. 18 to 20, and is a flowchart showing an encoding subroutine.

FIG. 23 is a block diagram illustrating a configuration of an entropy encoder employed in an embodiment of the present invention.

24 is a diagram illustrating an encoding table in an encoding table unit of the entropy encoder of FIG. 23, and is a diagram illustrating a table in a case where a prediction bit length is “1”.

25 is a diagram illustrating an encoding table in an encoding table unit of the entropy encoder in FIG. 23, and is a diagram illustrating a table in a case where a prediction bit length is “2”.

26 is a diagram illustrating an encoding table in an encoding table unit of the entropy encoder in FIG. 23, and is a diagram illustrating a table in a case where a prediction bit length is “4”.

27 is a diagram illustrating an encoding table in an encoding table unit of the entropy encoder of FIG. 23, and is a diagram illustrating a table in a case where a prediction bit length is “8”.

FIG. 28 is a diagram showing a state transition table in a state transition unit of the entropy encoder of FIG.

FIG. 29 is a block diagram illustrating a configuration of an entropy decoder employed in an embodiment of the present invention.

30 is a diagram illustrating a decoding table in a decoding table unit of the entropy decoder in FIG. 29, where a prediction bit length is “1”;
It is a figure showing the table in the case of.

FIG. 31 is a diagram showing a decoding table in a decoding table section of the entropy decoder of FIG. 29, where the prediction bit length is “2”;
It is a figure showing the table in the case of.

FIG. 32 is a diagram showing a decoding table in a decoding table section of the entropy decoder of FIG. 29, where the prediction bit length is “4”;
It is a figure showing the table in the case of.

FIG. 33 is a diagram illustrating a decoding table in a decoding table section of the entropy decoder in FIG. 29, where the prediction bit length is “8”;
It is a figure showing the table in the case of.

FIG. 34 is a diagram for describing an example in which an algorithm employed in the present invention is applied to the multi-level sequence data described in the present embodiment, and is a diagram illustrating a state where input symbols are divided into a plurality of groups. .

FIG. 35 is a block diagram of a conventional multicolor image encoding system and decoding system.

FIG. 36 is an explanatory diagram of reference pixel data with respect to conventional encoding target pixel data.

FIG. 37 is a diagram for explaining an index of a multi-color image used in the related art and the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Encoding system 1c Exclusive logic generation means / motion compensation means 2 Preprocessing means 2a Natural picture conversion means 2b YUV conversion means 2c Decimation means 2d Quantization means 2e Indexing means 2f Palletizing means 3 Decoding system 4 Entropy encoder 7, 8 index conversion unit 9 correction means 9a smoothing means 9b intermediate color processing means 12 entropy encoder 100A color pixel data to be encoded 100B decoded color pixel data 200 encoded data

Claims (14)

[Claims] 1. An apparatus for encoding a color image, comprising:
Thinning means for thinning at least the UV component of the YUV component converted from the B component, quantizing means for quantizing the Y component which is not thinned or thinned, the UV component which is greatly thinned and the quantum Encoding means for compressing and encoding the converted Y component. 2. When forming a moving image by continuously forming a still image frame composed of color pixel data, a target frame to be encoded is divided into a plurality of blocks each composed of N × N pixels (N is an integer of 4 or more). When the block in the frame of interest is the same as the block in the reference frame, the motion vector of that block is converted into data as difference information, and the color image of the color image for which motion compensation is performed to reduce the data amount of the difference information In the encoding device, thinning means for thinning out at least the UV component of the YUV component converted from the RGB component, quantization means for quantizing the Y component which is not thinned out or thinned out, Encoding means for compressing and encoding the UV component and the quantized Y component, performing the motion compensation on the UV component, Encoding device of a color image, characterized in that does not perform the motion compensation for the serial Y component. 3. A natural image converting means for converting the color image into a multi-color image and converting the multi-color image into a natural image composed of RGB components, a YUV converting means for converting the converted RGB components into YUV components, Palletizing means for assigning an index to the large-thinned UV component and palletizing the UV component, wherein the encoding means comprises: the quantized Y component and the palletized indexed UV component. The color image encoding apparatus according to claim 1, wherein the color image is encoded. 4. The encoding means according to claim 1, wherein said encoding means rearranges the Y component and the UV component according to the frequency of appearance of the color, and encodes the values of each component rearranged by said color order converting means into a run-length model. And a run length modeling means for converting the run length into a run length.
Or a color image encoding device according to 3. 5. A method for encoding a color image, comprising the steps of:
A thinning step of thinning out at least the UV component of the YUV component converted from the B component; a quantization step of quantizing the Y component that is not thinned out or thinned out; An encoding step of compressing and encoding the converted Y component. 6. When forming a moving image by continuously forming a still image frame composed of color pixel data, a target frame to be encoded is divided into a plurality of blocks each composed of N × N pixels (N is an integer of 4 or more). When the block in the frame of interest is the same as the block in the reference frame, the motion vector of that block is converted into data as difference information, and the color image code for performing motion compensation to reduce the data amount of the difference information A thinning step for thinning out at least the UV component of the YUV component converted from the RGB components, a quantization step for quantizing the Y component that is not thinned out or thinned out, and a highly thinned U component.
A coding step of compressing and coding the V component and the quantized Y component, performing the motion compensation on the UV component, and not performing the motion compensation on the Y component. Image encoding method. 7. A natural image conversion step of converting the color image into a multi-color image and converting the multi-color image into a natural image composed of RGB components, a YUV conversion step of converting the converted RGB components into YUV components, A palletizing step of assigning an index to the largely-thinned UV component and palletizing the UV component, wherein the encoding step includes the quantized Y component and the palletized indexed UV component. 7. The color image encoding method according to claim 5, wherein 8. The encoding step includes: a color rank conversion step of rearranging the Y component and the UV component according to the frequency of appearance of the color; and a run length model by rearranging the values of the components rearranged by the color rank conversion step. And a run-length modeling step for converting the run-length into a run-length model.
Or a method for encoding a color image according to 7. 9. A decoding apparatus for decoding a color image having a YUV component, wherein decoding means decodes input target encoded data, and smoothing means performs a smoothing process on a quantized Y component in the decoded data. And the above Y
A color image decoding apparatus, comprising: intermediate color processing means for processing, by intermediate color interpolation, UV components thinned out more than components. 10. The input target encoded data is YUV
Decoding means for decoding the image data into color pixel data composed of components, and moving image forming means for forming a moving image by continuously forming still images composed of the color pixel data. When a frame is divided into a plurality of blocks each composed of N × N pixels (N is an integer of 4 or more), and a motion vector indicating a relative displacement of the block exists in decoded data relating to a block in the frame of interest, In a color image decoding apparatus that performs motion compensation by decoding the block using the same block in a reference frame to be referenced, the motion compensation is performed on the UV component in the data decoded by the decoding unit. A color image decoding apparatus, wherein the motion compensation is not performed for the Y component. 11. The color order conversion means for rearranging and decoding the Y component and the UV component according to the frequency of appearance of a color, and a value of each component rearranged by the color order conversion means. 11. The color image decoding apparatus according to claim 9, further comprising: run length decoding means for run length decoding data. 12. A method for decoding a color image having a YUV component, comprising: a decoding step of decoding input encoded data to be inputted; and a quantized Y in the decoded data.
A method for decoding a color image, comprising: a smoothing step of performing a smoothing process on a component; and an intermediate color processing step of processing a UV component, which is significantly thinned compared to the Y component, by an intermediate color interpolation. 13. The input target encoded data is YUV
A decoding step of decoding into color pixel data composed of the components, and a moving image forming step of forming a moving image by continuously forming a still image composed of the color pixel data. When a frame is divided into a plurality of blocks each composed of N × N pixels (N is an integer of 4 or more), and a motion vector indicating a relative displacement of the block exists in decoded data relating to a block in the frame of interest, In a color image decoding method for performing motion compensation by decoding the block using the same block in a reference frame to be referred to, in the data decoded by the decoding step, the motion compensation is performed for the UV component. And performing the motion compensation on the Y component. 14. The decoding step shows a color order conversion step of rearranging and decoding the Y component and the UV component according to the frequency of appearance of the color, and shows the values of the components rearranged by the color order conversion step. 14. The color image decoding method according to claim 12, further comprising a run-length decoding step of run-length decoding data.