JP4078337B2 - Codebook creation method and apparatus for use in vector quantization, and recording medium - Google Patents

Description

  The present invention relates to a method and apparatus for creating a code book used in vector quantization, and further relates to a recording medium storing a program for performing these processes and a code book having a predetermined data structure.

  Conventionally, various data compression techniques have been proposed. Among them, a technique called “vector quantization” is well known as one of data compression algorithms that can perform decompression processing of compressed data very easily. This algorithm has been known for a long time in the field of signal processing, and in particular has been applied to data compression or pattern recognition of image signals and audio signals.

  In this vector quantization, several pixel patterns (codes) of a certain size (for example, a 4 × 4 pixel block) are prepared, and each is given a unique number or the like (this set is called a “codebook”) ). Then, for example, blocks of the same size (for example, 4 × 4 pixels) are sequentially extracted from the image data of a two-dimensional array, and the most similar pattern is found from the code book, and the pattern number is assigned to the block. Perform data compression. In vector quantization, a data string in one block corresponds to one vector.

  On the reception side or decompression side of the compressed data encoded in this way, the original image can be reproduced simply by taking out the pattern corresponding to the number for each block from the code book. Therefore, on the decompression side, if the code book is received or held in advance, no special operation is required, and the original image can be reproduced with very simple hardware.

  In order to execute the vector quantization as described above, it is necessary to create a code book. The quality of the reproduced image to be reproduced is closely related to the quality of the code book to be used due to the characteristics of vector quantization. Therefore, for example, when performing data compression of an image, in order to obtain a high-quality reproduced image while maintaining a high compression ratio, it is a problem how to create a high-performance codebook. ing.

  Conventionally, several methods have been known as codebook optimization methods, including Kohonen's self-organizing map method. In these methods, the code book is optimized by performing appropriate mathematical processing using a sample image or the like. However, all the conventional optimization techniques have a problem that the obtained code book is a useful code book only for the data used in the optimization.

  That is, for example, a codebook optimized using image data of a person's face is the best codebook for the image used for the optimization, but is not necessarily best for other images. Not necessarily a codebook. Therefore, for example, if the code book is used for image data of another person's face and data compression is performed, the image quality of an image reproduced from the compressed data is lowered.

  Furthermore, for images that are included in the same “human face” category as the images used for optimization, images with different classifications such as landscapes and characters can be used, even if a relatively good image quality is obtained as a playback image. As a result, the image quality often deteriorates. That is, since the patterns included in the code book are completely different depending on the image, there is a problem that the code book becomes less versatile.

  Further, the conventional optimization technique has a problem that it is not possible to know what pattern is included in the optimized code book.

The present invention has been made to solve such a problem, so that a highly versatile code book can be realized for pattern data whose data values change rapidly, for example, character image data. This is the first purpose.
The second object of the present invention is to make it easy to recognize what pattern is included in an optimized codebook.

According to the present invention, there is provided a method of creating a codebook used in vector quantization, which comprises a set of code vectors, which is a data string having at least one or more data. In the data sequence that constitutes the vector, the ratio of the maximum value or the minimum value that each data of the data sequence can take is set, and the maximum value is set in any data that constitutes the data sequence according to the set ratio Alternatively, the code vector data string is created by assigning a minimum value and assigning the minimum value or the maximum value to data other than the data to which the maximum value or the minimum value is assigned. Here, an intermediate value may be assigned to predetermined data according to the arrangement state of the data to which the minimum value is assigned.
An apparatus for creating a code book used in vector quantization according to the present invention comprises a set of code vectors, which is a data string having at least one or more data. In the apparatus for creating a code book used in vector quantization, the code book In the data sequence constituting the vector, ratio setting means for setting the proportion of the maximum value or the minimum value that each data of the data sequence can take, and any data constituting the data sequence according to the set ratio Code vector creating means for assigning the maximum value or the minimum value to the data, and creating a data string of the code vector by assigning the minimum value or the maximum value to data other than the data to which the maximum value or the minimum value is assigned, respectively; It is provided with. Here, the code vector creating means may assign an intermediate value to predetermined data according to the arrangement state of the data to which the minimum value is assigned.
A computer-readable recording medium according to the present invention is a recording medium that includes a set of code vectors, which are data strings having at least one or more data, and stores a program of a procedure for creating a code book used in vector quantization. In the data sequence constituting the code vector, the ratio of the maximum value or the minimum value that can be taken by each data of the data sequence is set, and the data sequence is configured according to the set ratio. The computer executes the procedure for creating the data sequence of the code vector by assigning the maximum value or minimum value to the data and assigning the minimum value or maximum value to data other than the data to which the maximum value or minimum value is assigned. It is characterized by having recorded a program for making it happen.
Another computer-readable recording medium according to the present invention is a codebook used in vector quantization, which is a codebook used in vector quantization, which is a set of code vectors that is a data string having at least one data. The maximum value or the minimum value is assigned to any data constituting the data string according to the ratio of the maximum value or the minimum value that each data of the data string can take in the data string constituting the data string, and the maximum value Alternatively, one code is assigned by assigning a minimum value or a maximum value to data other than the data to which the minimum value is assigned, and assigning an intermediate value to predetermined data according to the arrangement state of the data to which the minimum value is assigned. A codebook having a data structure as a vector is recorded.

  According to the present invention, in the data string constituting the vector, the ratio of the maximum value or the minimum value that each data string can take is set, and any data constituting the data string according to the set ratio Since one code vector is created by assigning the minimum value or the maximum value to data other than the data to which the maximum value or the minimum value is assigned, and at least the data value A highly versatile codebook can be easily created for pattern data in which the abruptly changes, for example, character image data. It is also possible to easily recognize what pattern is included in the created codebook.

  As described above, according to the present invention, a highly versatile codebook that can deal with various data including monotonously changing data and rapidly changing data can be obtained by giving a desired value to the code vector data. Can be realized. By using the code book created according to the present invention, it is possible to realize a highly versatile data compression / decompression system that can reproduce high-quality images for various images even at a high compression rate. Become.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
(First embodiment)
In the first embodiment, an apparatus and method for creating a code book for a face image, for example, will be described. FIG. 1 is a functional block diagram illustrating a configuration example of the code book creation device according to the present embodiment, and FIG. 2 is a flowchart illustrating the operation of the code book creation device according to the present embodiment. FIG. 3 is a conceptual diagram for explaining a method of creating a face image code book according to the present embodiment.

  In general, most of the face image has a pattern in which pixel values change smoothly as a whole, the change is monotonous, and the amount of change is very small. Therefore, in this embodiment, as a typical code pattern of the face image code book, a pattern that changes monotonously in eight directions is created as shown in FIG.

  That is, for example, a set of pixel values each composed of a block of 4 pixels × 4 pixels is defined as one vector (hereinafter referred to as a code vector) in the code book, and the edge portion of the block (upper and lower left and right sides and four corner points) ) Is used as a starting point to create a pattern in which pixel values (for example, luminance values) in the block gradually change. Here, as an example, a code vector in units of 4 × 4 pixels is created, but the size is not limited to this.

  In FIG. 3A, pattern 1 (indicated by a circled number in the figure) is such that the luminance value is A → B → C → D with the left side of the block as the starting point and the vertical column as one unit toward the right side. The pattern gradually increases. For example, as shown in FIG. 3B, the degree of change is changed by 20 brightness values for each pixel. That is, if the luminance value of the leftmost column in the block is K, the luminance values of the second, third, and rightmost columns are K + 20, K + 40, and K + 60, respectively.

  In the present embodiment, the luminance value is set to 20 as the degree of change, and the pattern is generated by changing the degree to various values such as 10, 5, 2, and 1, for example. Furthermore, the luminance value K of the pixel that is the starting point of the change is given as a minimum value and a maximum value that can be taken, and is distributed in an appropriate manner therebetween. As a result, the pattern in which the luminance value gradually changes from the left to the right in the block as in the pattern 1 is also given variations. Note that the step width of the gradually changing luminance value is not limited to the above-described example.

  Here, only the pattern 1 in FIG. 3A has been described in detail, but the same applies to the other patterns 2 to 8 (indicated by circled numbers in the figure). By the way, pattern 2 is a pattern in which the luminance value gradually increases toward the left side starting from the right side of the block, starting from the right side of the block, and pattern 3 is 1 unit starting from the bottom side of the block. As a pattern in which the luminance value gradually increases toward the upper side, and the pattern 4 is a pattern in which the luminance value gradually increases toward the lower side with the upper side of the block as the starting point and one horizontal row as one unit.

  Pattern 5 starts with the upper left corner of the block as a starting point, and the luminance value gradually increases toward the lower right corner with one diagonal column as one unit. Pattern 6 starts with the upper right corner of the block as a starting point. The pattern in which the brightness value gradually increases toward the lower left corner with 1 as the unit. Pattern 7 starts with the lower left corner of the block as the starting point, and the brightness value gradually increases toward the upper right corner with 1 row as one unit. The pattern 8 is a pattern in which the luminance value gradually increases toward the upper left corner, starting from the lower right corner of the block as a starting point, with one row as one unit. In these patterns 5 to 8, the luminance value gradually changes over seven stages.

Next, the functional configuration and operation of the apparatus for creating a code vector as shown in FIG. 3 will be described using the block diagram shown in FIG. 1 and the flowchart shown in FIG.
In FIG. 2, first, in step S1, for example, blocks of 4 × 4 pixel units are classified into several groups (number of groups N). For example, when the code vector of the pattern 1 shown in FIG. 3A is created, the vertical column is divided into four groups A to D as one group.

  When this grouping is performed, first, the start point setting unit 1 in FIG. 1 is used to set where in the block the start point is set. For example, when creating the pattern 1 in FIG. 3A, four pixels (group A) on the left side in the block are set as the starting points. The start point setting unit 1 may be configured by an input device such as a keyboard and a mouse, for example, and the user may set an arbitrary start point, or the eight patterns 1 to 8 shown in FIG. May be configured so that the apparatus itself automatically sets the starting point.

  The grouping unit 2 performs grouping according to where the start point setting unit 1 sets the start point. For example, when creating pattern 1 or pattern 2 in FIG. 3A, grouping is performed with one column in the vertical direction as one group, and when creating pattern 3 or pattern 4, one column in the horizontal direction. Are grouped as one group. Further, when creating patterns 5 to 8, grouping is performed with one column in an oblique direction as one group.

  Next, in step S2, the increment setting unit 3 in FIG. 1 is used to set the increment (step) of the luminance value between adjacent groups. The increment setting unit 3 may also be configured by an input device such as a keyboard or a mouse, for example, and the user may set an arbitrary increment, or the apparatus automatically sets a predetermined default value. You may comprise as follows. In step S2, a process of initializing a count value i for counting the number of processed groups to 0 is also performed.

Next, in step S3, the start point brightness value setting unit 4 in FIG. 1 sets the start point brightness value in the block. Here, when giving the luminance value of the starting point, for example,
0 (maximum luminance value (= 255) − (N−1) × step)
Are divided into n equal parts, and each equal value is set as the luminance value of the starting point. Here, although an equal value is given, it is not always necessary to be an equal value as long as it is within the range of 0 to (255− (N−1) × step). Further, although a plurality of luminance values are given by calculation here, the user may arbitrarily input the luminance value of the starting point within the above range based on his / her own judgment.

  When necessary information such as the luminance value of the starting point and the luminance increment is set in this way, the information is given to the code book creation unit 5 of FIG. The code book creation unit 5 creates one code vector by executing the processing from step S4 onward in FIG. That is, since the luminance value that is the starting point in one code vector has been set up to the processing in step S3, in the next step S4, the processing is moved to the group adjacent to the starting point, and the count value i is increased by one. .

  In step S5, it is determined whether or not the count value i is smaller than the value of the number of groups N. If the count value i is smaller, it means that an unprocessed group remains, and the process proceeds to step S6. A luminance value calculated by (luminance value of starting point group + i × step) is given. Further, the count value i is incremented by one, the process returns to step S5, and the same process is repeated until there is no unprocessed group.

  When luminance values have been given to all groups, a code vector having a pattern in which the luminance value gradually increases from group to group from the start point to the end point is generated. In this case, the process proceeds to step S7, and one code vector generated at that time is stored in the code book data memory 6 of FIG.

  As described above, when the process of the flowchart shown in FIG. 2 is completed once, one code vector is created and stored in the code book data memory 6. By performing such processing a plurality of times while changing the position of the starting point in the block, the luminance value of the starting point, and the increment value, the codes of the eight patterns 1 to 8 as shown in FIG. A plurality of vectors are generated with different luminance values. A set of these plural code vectors becomes a code book for face images. The code book size is, for example, 512.

  FIG. 4 illustrates a vector quantization (VQ) process performed on five face images using a code book of size 512 created by the method of the present embodiment, and then restored using the same code book. It is a figure which shows the PSNR (Peak signal to noise ratio) characteristic of each image. For reference, codebooks were individually optimized using Kohonen's self-organizing map method for each image, and each image was used for VQ processing and then used for VQ. The PSNR characteristics of each image restored using the same codebook are also shown.

  As is clear from the results of FIG. 4, the PSNR characteristic obtained when using one codebook created by the method of the present embodiment is obtained when five codebooks created by Kohonen's algorithm are used. The PSNR characteristics are comparable. As described above, according to the present embodiment, even with only one codebook, the PSNR characteristic when the face image is optimally created by the Kohonen algorithm is obtained, which is extremely versatile. It is confirmed that a highly functional codebook can be created.

  As described above, according to the first embodiment, it is possible to obtain a highly versatile code book that can correspond to various face images with only one code book. In this embodiment, each code vector is created according to a predetermined calculation pattern, so that it is easy to recognize what pattern is included in the optimized code book. You can also

In the above embodiment, a pattern in which the luminance value gradually increases from the start point to the end point in the block is generated. On the contrary, a pattern in which the luminance value gradually decreases may be generated. good.
In the present embodiment, the maximum value of the degree of change is 20. However, 20 is not necessarily used, and an arbitrary value such as 40 or 10 may be appropriately given. Also, the codebook size need not be 512.

(Second Embodiment)
In the second embodiment, an apparatus and method for creating a codebook for landscape images, for example, will be described. FIG. 5 is a functional block diagram illustrating a configuration example of the code book creation device according to the present embodiment, and FIG. 6 is a flowchart illustrating the operation of the code book creation device according to the present embodiment. FIG. 7 is a conceptual diagram for explaining a method for creating a codebook for landscape images according to the present embodiment.

  When landscape images were examined, it was found that there were many patterns in which the pixel values changed monotonously as in the typical pattern of the face image, but many patterns with very large changes were included. Therefore, in this embodiment, a code vector having a pattern in which the luminance value gradually changes in the block and the dynamic range of the luminance value is used over almost the entire range is created.

  That is, a pattern in which the luminance value gradually changes with one of the four corners of the block as the starting point and the diagonal side as the end point is used, and the inside of the block is defined, for example, divided into five regions. On the other hand, the dynamic range of luminance values consisting of 256 levels is defined by dividing it into five ranges. Then, a landscape pattern is created by appropriately giving the luminance value of each region in the block within each level range. In this embodiment, as such a landscape pattern, a pattern in which luminance values gradually change in four directions starting from the four corners of the block is created.

  FIG. 7A shows a pattern in which the luminance value gradually increases toward the lower right corner, starting from the upper left corner of the block. As shown in FIG. 7A, a block of 4 pixels × 4 pixels is divided into five areas A to E in an oblique direction. On the other hand, as shown in FIG. 7B, the dynamic range of luminance values from 0 to 255 is defined by being divided into five ranges corresponding to the five regions A to E.

  In this example, the luminance value 0 to 63 for the region A, the luminance value 63 to 95 for the region B, the luminance value 95 to 159 for the region C, the luminance value 159 to 191 for the region D, and the region E. In this case, luminance values 191 to 255 are defined. Then, as a luminance value of each of the areas A to E in the block, a value for a level range assigned to each of the areas is appropriately selected and given to create a landscape pattern.

Next, the functional configuration and operation of the apparatus for creating a code vector as shown in FIG. 7 will be described using the block diagram shown in FIG. 5 and the flowchart shown in FIG.
In FIG. 6, first, in step S11, for example, blocks of 4 × 4 pixel units are classified into several groups (group number N). For example, when the code vector having the pattern shown in FIG. 7A is created, each pixel in the block is divided into five groups A to E.

  When this grouping is performed, first, the start point setting unit 11 in FIG. 5 is used to set where in the block the start point is set. For example, when creating the pattern of FIG. 7A, the upper left corner of the block is set as the start pixel. The start point setting unit 11 may be configured by an input device such as a keyboard or a mouse, for example, and the user may set an arbitrary start point, or a device that sequentially creates patterns that change in four directions. May be configured to automatically set the starting point. The grouping unit 12 performs grouping according to where the start point setting unit 11 sets the start point.

Next, in step S12, the dynamic range dividing unit 13 of FIG. 5 is used, and N dynamic ranges of luminance values (S ₀ , S ₁ ,. _N-1 ). In step S12, a process of initializing a count value i for counting the number of processed groups to 0 is also performed. When block grouping and dynamic range division are performed in this way, such information is provided to the code book creation unit 14 of FIG.

The code book creation unit 14 creates one code vector by executing the processing from step S13 onward in FIG. That is, in step S13, with respect to (the group A in the example of FIG. 7) group of the starting point, to set one of luminance values belonging to the level range S ₀ of the luminance corresponding to the group. In the next step S14, the processing is moved to the group adjacent to the starting point (group B in the example of FIG. 7), and the count value i is incremented by one.

In step S15, it is determined whether or not the count value i is smaller than the value of the number of groups N. If the count value i is smaller, it means that an unprocessed group remains, and the process proceeds to step S16. Any luminance value belonging to the luminance level range S _i corresponding to the group is set. Further, the count value i is incremented by one, the process returns to step S15, and the same process is repeated until there is no unprocessed group.

  When luminance values have been given to all groups, a code vector having a pattern in which the luminance value gradually increases from group to group from the start point to the end point is generated. In this case, the process proceeds to step S17, and one code vector generated at that time is stored in the code book data memory 15 of FIG.

  As described above, when the process of the flowchart shown in FIG. 6 is completed once, one code vector is created and stored in the code book data memory 15. By performing this process multiple times while changing the position of the start point in the block and the luminance value given to the pixels in each area, multiple code vectors of patterns that change in four directions are generated with different luminance values. Will be. A set of these code vectors is a code book for landscape images.

  Although not shown here, when the code book created in this embodiment is applied to several types of landscape images as in the first embodiment, Kohonen's algorithm is applied to each landscape image. Thus, a characteristic almost equivalent to the PSNR characteristic of the codebook optimally prepared by the above method was obtained, and a relatively good PSNR characteristic was obtained.

  As described above, according to the second embodiment, it is possible to obtain a highly versatile code book that can correspond to various landscape images with only one code book. Further, in the present embodiment, individual code vectors are created according to a predetermined pattern determined in advance, so that it is easy to recognize what pattern is included in the optimized code book. You can also.

In the above embodiment, a pattern in which the luminance value gradually increases from the start point to the end point in the block is generated. On the contrary, a pattern in which the luminance value gradually decreases may be generated. good.
In this embodiment, the area in the block is divided into five areas. However, this is only the code vector divided into five areas, and does not need to be five.

(Third embodiment)
In the third embodiment, for example, an apparatus and method for creating an image code book containing characters will be described. FIG. 8 is a functional block diagram illustrating a configuration example of the code book creation device according to the present embodiment, and FIG. 9 is a flowchart illustrating the operation of the code book creation device according to the present embodiment. FIG. 10 is a conceptual diagram for explaining a method for creating a code book for character images according to the present embodiment.

  Examining the images with letters, it was found that most of the patterns were very varied. In addition, most of the changes in the luminance value are those that fully use the 256 level dynamic range. Therefore, in this embodiment, four typical pattern vectors are created as code vectors for character images as shown in FIG.

  As shown in FIG. 10A, the first pattern is 1 × 1, 1 × 2, 1 × 3, 1 with respect to a background white pattern (maximum luminance value pattern) of 4 × 4 pixels. This is a pattern in which seven types of black patterns (patterns of minimum luminance values) of × 4, 2 × 2, 2 × 3, and 2 × 4 are appropriately embedded. As shown in FIG. 10B, the second pattern is a pattern in which black dots are embedded in a diagonal line in the block.

  As shown in FIG. 10C, the third pattern is a pattern in which black dots are embedded so as to cross. In addition, the fourth pattern is a pattern in which a black dot line is embedded in the middle as shown in FIG. Further, in these patterns, the pixel value adjacent to the black point is set to an intermediate value depending on whether the black point in the block is linear or not.

Next, the functional configuration and operation of the code vector generating apparatus as shown in FIG. 10 will be described with reference to the block diagram shown in FIG. 8 and the flowchart shown in FIG.
In FIG. 9, first, in step S21, for example, the proportion of pixels of black spots (minimum luminance values) in a 4 × 4 pixel unit block is determined.

  The ratio of black spots in the block is performed by the minimum luminance ratio setting unit 21 in FIG. The minimum luminance ratio setting unit 21 may be configured by an input device such as a keyboard or a mouse, for example, and the user may set an arbitrary ratio, or the apparatus automatically sets various patterns in order. You may comprise so that it may set automatically.

  Next, in step S22, the maximum / minimum luminance value setting unit 22 of FIG. 8 is used to give a black point (minimum luminance value) to a predetermined pixel position in the block according to the set black dot ratio, A white point (maximum luminance value) is given to the pixel position. Next, in step S23, it is determined whether or not the black spots in the block are linear. If not linear, the code vector created by the processing up to step S22 is stored in the codebook data memory 24 of FIG.

  On the other hand, if the black dot occupancy in the block is linear, the process proceeds to step S24, and the luminance value changing unit 23 in FIG. 8 is used to set the luminance value of the white dot pixel adjacent to the black dot pixel to the intermediate color. Change to the brightness value. Then, the code vector created in this way is stored in the code book data memory 24 of FIG.

  As described above, when the process of the flowchart shown in FIG. 9 is completed once, one code vector is created and stored in the codebook data memory 24. By performing such processing a plurality of times while changing the proportion of black spots in the block, the positions where the black spots are given, and the like, a plurality of code vectors having various patterns are generated. A set of these plural code vectors becomes a code book for character images.

  Although not shown here, when the code book created in this embodiment is applied to several types of character images as in the first embodiment, Kohonen's algorithm is applied to each character image. Thus, a characteristic almost equivalent to the PSNR characteristic of the codebook optimally prepared by the above method was obtained, and a relatively good PSNR characteristic could be obtained.

  As described above, according to the third embodiment, it is possible to obtain a highly versatile code book that can correspond to various character images with only one code book. In the present embodiment, individual code vectors are created in accordance with predetermined rules, so that it is easy to recognize what patterns are included in the created codebook. You can also.

  In the above embodiment, the patterns occupied by the black dots in the block are divided into seven patterns of 1 × 1, 1 × 2, 1 × 3, 1 × 4, 2 × 2, 2 × 3, and 2 × 4. However, there is no need for seven types.

(Fourth embodiment)
In the fourth embodiment, each code book created by the method shown in the first to third embodiments is combined into one code book of size 1024.

  FIG. 11 illustrates a vector quantization (VQ) process performed on 19 types of images A to S including a face image, a landscape image, and a character image using a code book of size 1024 according to the present embodiment. It is a figure which shows the PSNR characteristic of each image which decompress | restored it using the same codebook. For reference, codebooks were individually optimized using Kohonen's self-organizing map method for each image, and each image was used for VQ processing after each codebook was used for VQ. The PSNR characteristics of each image restored using the same codebook are also shown.

  As is clear from the results of FIG. 11, the PSNR characteristic obtained when using one codebook created by the method of the present embodiment is obtained when 19 codebooks created by Kohonen's algorithm are used. It is comparable to the PSNR characteristics obtained. As described above, according to the present embodiment, even with only one codebook, a characteristic almost equal to the PSNR characteristic when each image is optimally created by the Kohonen algorithm is obtained, and the change is monotonous. It is confirmed that high-quality images can be reproduced for various images including images that change or images that change rapidly, and a highly versatile codebook can be created.

  In this embodiment, the code book size is 1024. However, the code book size is not necessarily 1024. For example, the code book can be appropriately determined as 128 or 512 according to the application. Further, although the three types of code books are combined into one, appropriate two types of code books may be combined into one.

  In each of the first to fourth embodiments described above, an example of generating a code vector in units of 4 pixels × 4 pixels has been shown. However, the code vector need not be 4 pixels × 4 pixels, and vector quantization It is not necessary that the target of the image data.

(Other embodiments of the present invention)
Note that the code book creation device of the present embodiment described above is configured by a microcomputer system including a CPU, a ROM, a RAM, and the like, and the operation is realized according to a work program stored in the ROM or the RAM. In this case, a program for realizing the function of each function block of the codebook creation apparatus may be supplied from the outside to a computer via a recording medium, and the function block may be operated according to the program.

  In this case, as a recording medium for storing such a program, for example, a floppy (registered trademark) disk, hard disk, optical disk, magneto-optical disk, CD-ROM, CD-I, CD-R, CD-RW, DVD, zip, magnetic A tape or a non-volatile memory card can be used.

  In addition, the functions of the above-described embodiments are realized by executing a program supplied by a computer, and the program is jointly operated with an OS (operating system) or other application software running on the computer. When the functions of the above-described embodiment are realized, or when all or part of the processing of the supplied program is performed by a function expansion board or a function expansion unit of the computer, the function of the above-described embodiment is realized. Such a program is included in the embodiment of the present invention.

It is a functional block diagram which shows the structural example of the code book creation apparatus for face images by 1st Embodiment. It is a flowchart which shows operation | movement of the code book creation apparatus for face images by 1st Embodiment. It is a conceptual diagram for demonstrating the creation method of the code book for face images by 1st Embodiment. It is a figure which shows the comparison result of the PSNR characteristic of the code book produced by 1st Embodiment, and the code book optimized by the technique of Kohonen's self-organization map individually. It is a functional block diagram which shows the structural example of the code book creation apparatus for landscape images by 2nd Embodiment. It is a flowchart which shows operation | movement of the code book preparation apparatus for landscape images by 2nd Embodiment. It is a conceptual diagram for demonstrating the creation method of the code book for landscape images by 2nd Embodiment. It is a functional block diagram which shows the structural example of the code book preparation apparatus for character images by 3rd Embodiment. It is a flowchart which shows operation | movement of the code book preparation apparatus for character images by 3rd Embodiment. It is a conceptual diagram for demonstrating the creation method of the code book for character images by 3rd Embodiment. It is a figure which shows the comparison result of the PSNR characteristic of the code book produced by 4th Embodiment, and the code book optimized by the technique of Kohonen's self-organization map individually.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Start point setting part 2 Grouping part 3 Increment setting part 4 Start point luminance value setting part 5 Code book creation part 6 Code book data memory 11 Start point setting part 12 Grouping part 13 Dynamic range division part 14 Code book creation part 15 Code book data Memory 21 Minimum luminance ratio setting unit 22 Maximum / minimum luminance value setting unit 23 Luminance value changing unit 24 Codebook data memory

Claims (7)

In a method of creating a codebook that is composed of a set of code vectors, which is a data string having at least one or more data, and is used in vector quantization,
In the data sequence that constitutes the code vector, the ratio of the maximum value or the minimum value that can be taken by each data of the data sequence is set, and the arbitrary data that constitutes the data sequence according to the set ratio is set as described above. A vector quantization characterized by creating a data string of the code vector by assigning a maximum value or a minimum value and assigning a minimum value or a maximum value to data other than the data to which the maximum value or the minimum value is assigned, respectively. How to create a codebook for use in   2. The method for creating a codebook used in vector quantization according to claim 1, wherein an intermediate value is assigned to predetermined data in accordance with an arrangement state of data to which the minimum value is assigned. In an apparatus for creating a codebook, which is composed of a set of code vectors, which is a data string having at least one data, and is used in vector quantization,
In the data string constituting the code vector, a ratio setting means for setting a ratio occupied by the maximum value or the minimum value of each value of the data string,
According to the set ratio, the maximum value or the minimum value is assigned to arbitrary data constituting the data string, and the minimum value or the maximum value is assigned to data other than the data to which the maximum value or the minimum value is assigned. A code book creating apparatus used in vector quantization, comprising:   4. The code book used in vector quantization according to claim 3, wherein the code vector creating means assigns an intermediate value to predetermined data according to an array state of the data to which the minimum value is assigned. apparatus. A recording medium comprising a set of code vectors, which is a data string having at least one data, and storing a program of a procedure for creating a code book used in vector quantization,
In the data sequence that constitutes the code vector, the ratio of the maximum value or the minimum value that can be taken by each data of the data sequence is set, and the arbitrary data that constitutes the data sequence according to the set ratio is set as described above. A method for causing a computer to execute a procedure for creating a data string of the code vector by assigning a maximum value or a minimum value and assigning a minimum value or a maximum value to data other than the data to which the maximum value or the minimum value is assigned, respectively. A computer-readable recording medium on which a program is recorded.   6. The computer according to claim 5, further comprising a program for causing the computer to further execute a procedure of assigning an intermediate value to predetermined data in accordance with an arrangement state of the data to which the minimum value is assigned. A readable recording medium.   A code book that is composed of a set of code vectors, which is a data sequence having at least one or more data, and is used in vector quantization, and each data of the data sequence can take in the data sequence constituting the code vector According to the ratio of the maximum value or the minimum value, the maximum value or the minimum value is assigned to any data constituting the data column, and the data other than the data to which the maximum value or the minimum value is assigned is assigned a minimum value or A computer-readable recording of a codebook having a data structure in which one code vector is assigned by assigning a maximum value and assigning an intermediate value to predetermined data according to the arrangement state of the data to which the minimum value is assigned. Possible recording media.

Images