JPH0865671A - Compressing device, expanding device and compression/ expansion processing system - Google Patents

Abstract

PURPOSE: To improve compressibility by reducing a data amount by calculating the error between any specified predictive value and one component of second transformed data. CONSTITUTION: Orthogonal transformers 11, 21 and 31 orthogonally transform first and second color difference data expressing color difference and respectively calculate first and second transformed data. A predictor 80 predicts one component of the second transformed data correlated with one component of the first transformed data from one component of the first transformed data. Besides, a subtracter 90 calculates the error between one component of the second transformed data and the output from the predictor 80. Then, first quantizing parts 23 and 24 encode the first transformed data and calculate first compressed data. Further, second quantizing parts 33 and 34 calculate second compressed data while containing error encoded data for which the error calculated by the subtracter 90 is encoded. Namely, the predictive value for one component of the second transformed data is calculated from one component of the first transformed data, the error between this predictive value and one component of the second transformed data is calculated, the data amount is reduced and afterwards, the second compressed data are provided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compression device, a decompression device, and a compression / decompression processing system capable of improving the compression rate in image data compression.

[0002]

2. Description of the Related Art FIG. 18 is a block diagram showing a compression process of a conventional compression method. As shown in FIG. 18, first, the original image data 1 is input to the color converter 70, and predetermined color signal conversion (see the following equations 1, 2, and 3) is performed, and Y data, U data, and V data are obtained. Is obtained (see FIG. 7). Here, the Y data is data representing luminance, and the U data and V data are data representing color difference. The V data output from the color converter 70 in FIG. 18 is divided into blocks by the block divider 30 (see FIG. 8), and one V data is output in the order of predetermined numbers (see FIG. 10) described in the small blocks. Each of them is sent to the DCT converter 31 in FIG. DC in Figure 18
The T-transformer 31 performs a two-dimensional DCT transformation on the input small block data to transform it into a DCT coefficient (see Equation 4 described later).

The DCT coefficient is divided into a DC component and four classes by the class divider 32 in FIG. 18 (see FIG. 11). The vector quantizer 34 in FIG. 18 generates a 15-dimensional vector from the data group of the class 1 (see FIG. 11) output from the class divider 32 (see FIG. 12).
), Class 2, class 3 and class 4 (see FIG. 11).
16-dimensional vectors are generated from the respective data groups (see FIG. 13) (see FIG. 13). Then, the vector generated from the class 1 is quantized into a quantized representative vector that minimizes the distortion amount among all the quantized representative vectors of the codebook 301 in FIG. 18 (see the following Equation 12), and its index is set to 8 Output as bit data. The codebooks 302 to 30 are similarly applied to the classes 2, 3, and 4 as well.
4 is output and an 8-bit index is output. In this way, the vector quantizer 34 outputs a total of 32 bits as the input compressed data of four classes.

On the other hand, the DC component output from the class divider 32 in FIG. 18 is quantized by the scalar quantizer 33 in FIG. 18 and becomes 11-bit data. In the scalar quantizer 33 in FIG. 18, if the input DC component value is −10
When it is smaller than 24, it is quantized to -1024, and if 1
When it is larger than 023, it is quantized to 1023. Therefore, the compressed data for a small block of V data has a total of 43 bits including the output from the scalar quantizer 33 and the vector quantizer 34 in FIG. The process of compressing the Y data and U data output from the color converter 70 in FIG. 18 is the same as the process of compressing the V data.

In FIG. 18, 3, 4, and 5 are compressed data, 10 is a block divider, 11 is a DCT transformer (discrete cosine transform), 12 is a class divider, 13 is a scalar quantizer, and 14 Is a vector quantizer, 20 is a block divider, 21 is a DCT transformer, 22 is a class divider,
23 is a scalar quantizer, 24 is a vector quantizer, 1
01-104, 201-204 are codebooks.

[0006]

In the conventional example shown in FIG. 18, the DC component of the DCT coefficient of V data is converted into Y data or U data.
Like the DC component of the DCT coefficient of the data, it is encoded into 11 bits. DCT of Y data, U data, V data
In order to encode the DC component of the coefficient into the number of bits smaller than 11 bits, the value of the DC component must be quantized to a large extent. Then, the visual image quality is significantly deteriorated.

In view of the above problems, it is an object of the present invention to provide a compression device, a decompression device, and a compression / decompression processing system that can improve the compression ratio without degrading the image quality.

[0008]

According to a first aspect of the present invention, there is provided a first means for solving the problems by orthogonally transforming the first color difference data and the second color difference data representing the color difference, respectively. An orthogonal transformer that obtains two transformed data, a predictor that predicts one component of the second transformed data correlated with one component of the first transformed data, and one predictor of the second transformed data.
A subtractor for obtaining an error between the component and the output from the predictor, a first quantizer for encoding the first converted data to obtain first compressed data, and the error obtained by the subtractor And a second quantizer that obtains the second compressed data by including the error coded data that is coded.

The problem solving means according to claim 2 of the present invention is
A decompressor for decompressing the first compressed data and the second compressed data compressed by the compression device according to claim 1, wherein the first compressed data is dequantized to dequantize the first compressed data. A first dequantization section for obtaining data, a second dequantization section for dequantizing the second compressed data to obtain second dequantized data, and one component of the first compressed data Alternatively, a predictor that predicts a predicted value of one component of the second transformed data from one component of the first dequantized data, and the second dequantized data from the second dequantization unit. An adder for adding the predicted value of one component of the second converted data obtained by the predictor to obtain added data, and the first dequantized data from the first dequantization unit. And an inverse orthogonal transformer that inversely orthogonally transforms the addition data from the adder, respectively. .

The problem solving means according to claim 3 of the present invention is
A compression device that compresses the first color difference data and the second color difference data that represent color difference, and a decompression device that decompresses the data compressed by the compression device, wherein the compression device is the first
Of the first color difference data and the second color difference data by orthogonally converting the color difference data of the first color conversion data and the second color difference data of the second color difference data, and a second component correlated with one component of the first color conversion data. Predictor for predicting one component of the converted data, a subtracter for obtaining an error between the one component of the second converted data and the output from the predictor, and the first converted data to encode the first converted data. A first quantizer for obtaining compressed data and a second quantizer for obtaining second compressed data by including error coded data obtained by encoding the error obtained by the subtracter, The decompressor dequantizes the first compressed data to obtain first dequantized data, and a first dequantization unit dequantizes the second compressed data to dequantize the second dequantized data. A second dequantization unit for obtaining encoded data; A predictor for predicting a predicted value of one component of the second transformed data from a component or one component of the first dequantized data, and the second dequantization from the second dequantization unit. The second data obtained by the predictor
An adder for adding the predicted value of one component of the converted data to obtain addition data, the first dequantized data from the first dequantization unit, and the addition data from the adder. And an inverse orthogonal transformer that performs inverse orthogonal transformation.

[0011]

In the first and third aspects of the present invention, the predicted value of one component of the second converted data is calculated from the one component of the first converted data, and the predicted value and the one component of the second converted data are calculated. Since the second compressed data is obtained after the data amount is reduced by obtaining the error between and, the compression rate can be dramatically improved.

According to claim 2 and claim 3 of the present invention, when the second compressed data having the improved compression ratio as in claim 1 is restored, the first compressed data is converted into the second converted data. Since the predicted value of the component is obtained, and the predicted value is added to the second dequantized data and then the inverse orthogonal transform is performed, the first color difference data and the second color difference data before compression are almost completely formed. It is possible to restore with, and it is possible to prevent deterioration of image quality due to improvement of compression rate.

[0013]

【Example】

<Structure> The compression / expansion processing system according to the embodiment of the present invention is
In the field of color image data compression, U data (first color difference data) and V data (second color difference data)
DC of one data (U data) is utilized by utilizing the correlation between both DC components obtained by two-dimensional DCT conversion
DC component of V data is predicted from the component
The object of the present invention is to improve the compression rate without deteriorating the image quality by the method of coding the prediction error instead of coding the component.

FIG. 1 is a block diagram showing a compression apparatus of a compression / expansion processing system of this embodiment, and FIG. 2 is a block diagram showing the same expansion apparatus. Note that, of the reference numerals shown in FIGS. 1 and 2, the same reference numerals are given to the elements having the same functions as those of the conventional example.

In FIG. 1, 1 is original image data, 3 and 4 are compressed data, 10 is a block divider, and 11 is DCT.
Transformer (discrete cosine transformer: orthogonal transformer), 12 is a class divider, 13 is a scalar quantizer, 14 is a vector quantizer, 20 is a block divider, 21 is a DCT transformer (discrete cosine transform) Device: orthogonal transformer), 22 is a class divider, 23 is a scalar quantizer, 24 is a vector quantizer, 30 is a block divider, 31 is a DCT transformer (discrete cosine transformer: orthogonal transformer), 32 is a class divider, 33 is a scalar quantizer, 34 is a vector quantizer, 70 is a color converter, 80 is a predictor, 90 is a subtractor,
Reference numerals 101 to 104, 201 to 204, and 301 to 304 are codebooks. Here, the scalar quantizer 23
The vector quantizer 24 is referred to as a first quantizer, and the scalar quantizer 33 and the vector quantizer 34 are referred to as a second quantizer.

The original image data 1 is image data in which the number of pixels in the horizontal direction and the number of pixels in the vertical direction are both multiples of 8, as shown in FIG. 5, and one pixel of them is as shown in FIG. It is expressed in RGB three primary colors.

The color converter 70 uses the original image data 1
7 is performed by color signal conversion as shown in the following equations 1 to 3, and FIG.
Y data, U data (first color difference data), and V data (second color difference data) are obtained.

[0018]

[Equation 1]

[0019]

[Equation 2]

[0020]

(Equation 3)

Here, the Y data is data representing luminance, and the U data and V data are data representing color difference. The numbers of vertical and horizontal directions of Y data, U data, and V data in FIG. 7 are set equal to the numbers of vertical and horizontal pixels of the original image, respectively.

The block dividers 10, 20, 30 in FIG.
Have the same function as each other, and the Y data, U data and V data output from the color converter 70 are shown in FIG.
The block is divided into a plurality of blocks of 8 rows and 8 columns (hereinafter, simply referred to as small blocks).

The DCT converters 11, 21, 31 have the same functions as each other, and are given by the following equation 4 so that energy is concentrated on a specific frequency component with respect to given small block data. The orthogonal transform (cosine transform, that is, two-dimensional DCT transform) is performed to obtain the DCT coefficient having the same data structure as in FIG. In addition, the data from the DCT converter 21 related to U data is
The converted data, V data and the data from the DCT converter 31 are referred to as second converted data.

[0024]

[Equation 4]

However, the coefficients C (i) and C (j) of the equation 4 are divided into cases as in the following equation 5.

[0026]

(Equation 5)

The class dividers 12, 22, and 32 have the same functions as each other, and the DCT converters 11 and 2 are provided.
As shown in FIG. 11, the DCT coefficient obtained by
It divides into components and four classes.

The scalar quantizers 13 and 23 approximate (quantize) the DC component at discrete levels and convert it into 11-bit data. The value of the input DC component is -10.
When it is smaller than 24, it is quantized to -1024, and if 1
When it is larger than 023, it is quantized to 1023. Further, the scalar quantizer 33 approximates (quantizes) the error between the DC component and the predicted value at a discrete level and converts it into 7-bit data.

The vector quantizers 14, 24 (included in the first quantizer), 34 (included in the second quantizer) are output from the class dividers 12, 22, 32. The data group of class 1, class 2, class 3, and class 4 in FIG. 11 is approximated (quantized) at a discrete level, and a 15-dimensional vector as shown in FIG. A 16-dimensional vector as shown in FIG. 13 is generated from the data groups of class 2, class 3, and class 4, respectively. Then, the generated vectors are assigned to the codebooks 101 to 104, 201 to 204, 301 to
After being quantized into the quantized representative vector in 304, the corresponding index is output as a part of the compressed data (first compressed data, second compressed data, etc.). here,
In particular, the scalar quantizer 33 (included in the second quantizer) includes the error coded data obtained by coding the error found by the subtracter 90 described later to find the second compressed data. Become.

The predictor 80 substitutes the DC component of the input U data into the variable u of the following equation 6 to calculate and output the predicted value of the DC component of the V data.

[0031]

(Equation 6)

Coefficients α and β in the equation 6 are prediction coefficients set based on the original image data 1 in advance, and the variable v is V.
It represents the predicted value of the DC component of the data. A method of setting the coefficients α and β of the equation 6 will be described later in detail.

The subtractor 90 outputs a value (prediction error value) obtained by subtracting the prediction value which is the input from the predictor 80 from the DC component of V data. The output of the subtractor 90 is given by the following equation 7.

[0034]

(Equation 7)

The variable vi in the equation 7 is the DC component of the i-th small block of V data, and the variable ui is the i component of U data.
It is the DC component of the second small block. And αui + β
Represents the predicted value which is the output of the predictor 80 in FIG. However, the i represents the number described in the small block in FIG.

As shown in FIG. 3, the codebook 101 is composed of 256 15-dimensional vectors and indexes of those vectors. The index of the codebook 101 is an integer value starting from 0 and ending at 255 as shown in FIG. The codebook 201 and the codebook 301 in FIG. 1 have the same structure as the codebook 101. On the other hand, the codebook 102 in FIG. 1 is composed of 256 16-dimensional vectors and indexes (8 bits) of these vectors, as shown in FIG. The index of the codebook 102 is an integer value starting from 0 and ending at 255 as shown in FIG. In addition, the code book 103 in FIG.
Codebook 104, codebook 202, codebook 203, codebook 204, codebook 302,
The codebook 303 and the codebook 304 have the same structure as the codebook 102. Here, each of the codebooks 101 to 104, 201 to 204, 301 to 30
The vector in 4 will be hereinafter referred to as a quantized representative vector. Codebooks 101 to 104, 20 in FIG.
1 to 204 and 301 to 304 are optimized in advance for the original image data 1.

2, 6 and 7 are block combiners, 40 is an IDCT converter (inverse orthogonal converter), 41
Is a block generator, 42 is a searcher (inverse vector quantizer), 50 is an IDCT transformer (inverse orthogonal transformer), 51 is a block generator, and 52 is a searcher (inverse vector quantizer:
First dequantizer), 60 is an IDCT transformer (inverse orthogonal transformer), 61 is a block generator, 62 is a searcher (inverse vector quantizer: second dequantizer), 71 is color converter,
Reference numeral 81 is a predictor, and 91 is an adder.

The searcher (inverse vector quantizer) 42,
52 has the same function as each other, and the block generators 41 and 51 and the predictor 81 first take out 11 bits from the beginning of the data (first compressed data etc.) in the compressed data storage files 3 and 4. After that, the data following this is extracted 4 times by 8 bits. That is, the first 8-bit data (first 8-bit data), the subsequent 8-bit data (second 8-bit data), the subsequent 8-bit data (third 8-bit data), and The following 8-bit data (fourth 8-bit data) is taken out. Then, the indexes corresponding to the respective indexes are assigned to the codebooks 101 to 104 and 201 to 101 in FIG.
When a matching index is found by searching from the indexes of 204, the quantized representative vector corresponding to the index is output.

The searcher (inverse vector quantizer)
In the reference numeral 62, the adder 91 first extracts 7 bits from the beginning of the data (second compressed data) in the compressed data storage file 5, and then extracts the subsequent data 4 times by 8 bits. That is, the first 8-bit data (first 8-bit data), the subsequent 8-bit data (second 8-bit data), the subsequent 8-bit data (third 8-bit data), and The following 8-bit data (fourth 8-bit data) is taken out. Then, the index matching each of these is searched from the indexes of the codebooks 301 to 304 in FIG. 2, and when the matching index is found, the quantized representative vector corresponding to the index is output.

The block generators 41 and 51 have the same functions as each other, and the compressed data storage files 3 and 4 are provided.
11 bits from the beginning of the compressed data in D of the DCT coefficient
A class 1 data group shown in FIG. 12 is generated from the first vector output from the searchers (inverse vector quantizers) 42 and 52 as the C component, and the second vector and the third vector are generated. Vector, from the fourth vector to FIG.
The data groups of class 2, class 3, and class 4 shown in are generated respectively. Then, the small block data (first dequantized data or the like) shown in FIG. 11 is organized from the generated four classes and the previously input DC component.

The block generator 61 takes out the data in which the first 7 bits of the compressed data in the compressed data storage file 5 become 11 bits through the adder 91 as a DC component of the DCT coefficient, and A class 1 data group shown in FIG. 12 is generated from the first vector output from the searcher (inverse vector quantizer) 62, and the second vector, the third vector, and the fourth vector shown in FIG.
The data groups of class 2, class 3, and class 4 shown in are generated respectively. Then, the small block data (second dequantized data) shown in FIG. 11 is organized from the generated four classes and the previously input DC component.

The IDCT converters 40, 50 and 60 have functions equivalent to each other, and the block generator 4
Inverse cosine transform, that is, inverse 2
A three-dimensional DCT transformation is performed.

[0043]

[Equation 8]

However, the coefficients C (i) and C (j) of the equation 8 are classified according to the case of the equation 5.

The block synthesizers 6, 7 and 8 have the same functions as each other, and the IDCT converters 40 and 5 have the same functions.
The small block data given from 0 and 60 are sequentially combined at the positions of the numbers shown in the small blocks in FIG.

The color converter 71 uses the data from the block synthesizers 6, 7, and 8 as the Y data, the U data, and the V data, and uses them as image data consisting of RGB three primary colors as shown in FIG. (Decoded image data). The conversion formula of the color converter 71 is as shown in the following Expressions 9 to 11.

[0047]

[Equation 9]

[0048]

[Equation 10]

[0049]

[Equation 11]

The decoded image data 2 corresponds to the color converter 7
The decoded image data decoded in 1 is stored.

The predictor 81 reads the DC component of the U data, substitutes it into the variable u in the above equation 6 and calculates and outputs the predicted value given by the variable v.

The adder 91 adds the input value from the compressed data storage file 5 and the predicted value input from the predictor 81, and outputs this as added data.

<Operation> In the above configuration, as shown in FIG.
When the original image data 1 is input to the color converter 70 at the time of data compression, the color converter 70 configures the R, G, B constituting the original image data 1 by the conversion given by the above-mentioned equations 1 to 3.
The data is converted into Y data that is a luminance component and U data and V data that are color difference components, and Y data, U data, and V data are converted.
Output the data separately. However, the color converter 7 in FIG.
For 0, conversion to another color signal space may be performed instead of the color signal conversion given by the above equations 1 to 3. FIG.
Y data, U data, V output from the middle color converter 70
The data has the same number of pixels as the number of pixels of the original image data 1 in FIG. 1 as shown in FIG.

Here, the process of compressing Y data will be described. The Y data output by the color converter 70 in FIG.
As shown in FIG. 8, the block divider 10 in FIG.
It is divided into small blocks of 8 rows. The small blocks are sent to the DCT converter 11 block by block in the order shown in FIG. The numbers written in the small blocks represent the order in which the small blocks are sent to the converter 11 in FIG.
The DCT converter 11 in FIG. 1 applies the two-dimensional DCT converter given by the above-mentioned equation 4 to the input small block and outputs the DCT coefficient having the same data structure as in FIG. The coefficients C (i) and C (j) in the above equations 4 and 8 are as described in equation 5. DC in Figure 1
The T-transformer 11 may use another orthogonal transform instead of the two-dimensional DCT transform given by the equation (4).

The DCT coefficient output from the DCT converter 11 in FIG.
It is divided into two classes. The DC component is the data of the first row and first column of the DCT coefficient of 8 rows and 8 columns shown in FIG. Class 1 of the above four classes is composed of 15 data groups obtained by removing the DC components from the upper left 4 rows and 4 columns of the DCT coefficients in FIG. Class 2 among the above four classes is 4 at the upper right of the DCT coefficient in FIG.
It consists of 4 rows. Of the above four classes, class 3 consists of 4 rows and 4 columns at the lower left of the DCT coefficient in FIG. Of the four classes, class 4 consists of 4 rows and 4 columns at the lower right of the DCT coefficient in FIG. The DC component output from the class divider 12 in FIG. 1 is the scalar quantizer 13 in FIG.
Sent to. At the same time, the four classes output from the DCT converter 12 in FIG. 1 are sent to the vector quantizer 14 in FIG. The vector quantizer 14 in FIG. 1 generates a 15-dimensional vector from the data group of class 1 out of the four input classes and a 16-dimensional vector from the data group of class 2, class 3 and class 4, respectively. . Class 1,
The vectors generated from the class 2, class 3, and class 4 will be described as the first vector, the second vector, the third vector, and the fourth vector, respectively. here,
The correspondence between the data group of class 1 and the elements of the first vector is as shown in FIG. 12 (symbols a to o). Also, in FIG. 13, the correspondence between the data group of class 2 and the element of the second vector, the correspondence between the data group of class 3 and the element of the third vector, the data group of class 4 and the fourth by the symbols A to P in FIG. Shows the correspondence with the elements of the vector.

Next, the vector quantizer 14 in FIG.
The first vector is quantized with reference to the codebook 101. Here, first, codebook index 0
The distortion amount given by the following equation 12 is calculated from the quantized representative vector and the first vector in (1), and the value of the distortion amount and the index value 0 are stored.

[0057]

[Equation 12]

Next, the distortion amount given by the above equation 12 is calculated from the quantized representative vector at index 1 and the first vector, and is compared with the stored distortion amount value. Then, if the calculated strain amount value is smaller than the stored strain amount value, the stored strain amount value and the index value are deleted, and the calculated strain amount value and the index. The value "1" of is newly saved. If the calculated strain value is equal to or greater than the stored strain value, do nothing.
Similarly, the amount of strain given by the above equation 12 is calculated from the quantized representative vector at index 2 and the first vector, and the calculated amount of strain is smaller than the stored amount of strain. In this case, the stored strain amount value and index value are deleted, and the calculated strain amount value and index value “2” are newly stored. If the calculated distortion amount value is equal to or larger than the stored distortion amount value, nothing is done and the calculation of the distortion amount value with the quantized representative vector at index 3 is started. Hereinafter, the same is applied up to index 255. In this way, when the calculation of the distortion amounts of all the quantized representative vectors of the codebook 101 and the first vector is completed, the stored distortion amount is the smallest among all the calculated distortion amounts. The stored index is the index of the quantized representative vector that gives the minimum distortion amount.

The vector quantizer 14 quantizes the first vector into a quantized representative vector which gives the minimum distortion amount selected by the above method, and outputs the index value as compressed data of class 1. Second vector, third
Similarly, the codebook 102, the codebook 103, and the codebook 1 in FIG.
The quantized representative vector that gives the minimum distortion amount is selected from the quantized representative vectors of 04, and the index is output as compressed data.

Since an integer in the range of 0 to 255 can be represented by 8 bits, the vector quantizer 14 in FIG.
Compress the data groups of the four classes in each to 8 bits,
A total of 32 (8 × 4) bits are output. On the other hand, FIG.
The scalar quantizer 13 in the inside is the class divider 12 in FIG.
The DC component output in step 1 is quantized into 11 bits. At this time, the scalar quantizer 13 in FIG. 1 outputs − when the DC component of the input Y data is a value smaller than −1024.
Quantize to 1024 and 10 when larger than 1023
Quantize to 23. Therefore, the compressed data for one small block is 43 bits, which is the sum of the output of the scalar quantizer 13 in FIG. 1 and the output of the vector quantizer 14 in FIG. FIG. 14 shows the structure of compressed data for one small block. The 11 bits from the front of the compressed data in FIG. 14 are the DC component compressed data, followed by 8 bits of compressed data for class 1, 8 bits of compressed data for class 2, and 8 compressed data for class 3.
Bit, followed by 8 bits of compressed data for class 4.
The 43 bits of compressed data for each small block become the compressed data storage file 3 in the figure in order.

Next, the compression process of the U data output from the color converter 70 in FIG. 1 will be described. Similarly to the Y data, the U data output from the color converter 70 is divided into the DC component and four classes shown in FIG. 11 through the block divider 20, the DCT converter 21, and the class divider 22. It The four classes output from the class divider 22 in FIG. 1 are sent to the vector quantizer 24 and compressed to 32 bits in the same manner as the Y data is compressed by the vector quantizer 14. It On the other hand, the DC component output from the class divider 22 in FIG.
3 is sent to the predictor 80 at the same time. The scalar quantizer 23 encodes the DC component of the input U data into 11 bits in the same manner as the scalar quantizer 13 encodes the DC component of Y data into 11 bits. Scalar quantizer 23 and vector quantizer 2
The compressed data for the small block of U data output in 4 has the same structure as the compressed data for the small block of Y data. Therefore, the compression process of U data is
The process is the same as the Y data compression process except that the components are sent to the predictor 80. The predictor 80 in FIG.
The DC component of the data is substituted into the variable u of the above-mentioned equation 6 to calculate and output the predicted value of the DC component of the V data.

Next, the compression process of the V data output from the color converter 70 in FIG. 1 will be described. The V data output from the color converter 70 in FIG. 1 passes through the block divider 30, the DCT converter 31, and the class divider 32 in FIG. Divided into 4 classes. The four classes output by the class divider 32 in FIG. 1 are sent to the vector quantizer 34, and compressed into 32 bits in the same manner as the Y data is compressed by the vector quantizer 14. It on the other hand,
DC of V data output by the class divider 32 in FIG.
The components are input to the subtractor 90. The subtractor 90 outputs a value (prediction error value) obtained by subtracting the prediction value which is the input from the predictor 80 from the DC component of the input V data. Subtractor 9
The output of 0 is given by the equation (7). Here, in FIG.
Is a graph showing the relationship between both sides of Expression 7. Αui + β shown in FIG. 17 is a prediction value for the value vi, and δvi is a prediction error value thereof. Since the scalar quantizer 33 in FIG. 1 quantizes the prediction error value δvi having a smaller amplitude, instead of quantizing the value vi shown in FIG. 17, it can be coded with a smaller number of bits. The scalar quantizer 33 quantizes the prediction error value δvi to obtain 7-bit data. At this time, the scalar quantizer 33 quantizes to -64 if the input prediction error value is smaller than -64.
If it is larger than 63, it is quantized to 63. Thus V
The DC component of data is compressed by 4 bits less than the DC component of Y data and U data. Therefore, the V output from the scalar quantizer 33 and the vector quantizer 34
The compressed data for one small block of data is 39 bits. FIG. 15 shows compressed data for a small block of V data. 7 from the front of the compressed data in FIG.
The bits are the compressed data of the DC component of the V data, followed by 8 bits of compressed data of class 1, 8 bits of compressed data of class 2, 8 bits of compressed data of class 3, and 8 bits of compressed data of class 4. . Where, in particular,
The second compressed data obtained by the vector quantizer 34 (included in the second quantizer) includes error coded data obtained by coding the error obtained by the subtractor 90.

Next, a method for obtaining the above-described coefficients α and β of the equation 6 will be described. The number described in the small block of FIG. 10 is represented by the variable i. In addition, D of the i-th block of U data output by the class divider 22
The C component is represented by a variable ui, and the DC component of the i-th block of the V data output by the class divider 32 is represented by a variable vi. In general, since ui and vi have a correlation, u− with the variable ui on the horizontal axis and the variable vi on the vertical axis
The set of points (ui, vi) on the v plane shows a biased distribution as shown in FIG. Therefore, if the number of straight lines that optimally models this distribution is known, the value of the variable ui is used to calculate the variable v
The value of i can be predicted with a small error. Therefore, using the unknowns α and β, the optimal straight line equation for the distribution of points (ui, vi) is assumed to be Equation 6, and all points (ui, vi)
Then, α and β are set by the least square method. Point (ui, v
The squared error between i) and the point (ui, αui + β) on the straight line in FIG. 17 is given by the following equation 13.

[0064]

[Equation 13]

Here, as shown in the following equation 14, all i
For the above, the squared error is calculated and the total sum is obtained. It can be said that the straight line that minimizes the sum of squared errors is a straight line that optimally models the distribution of points (ui, vi).

[0066]

[Equation 14]

The left side S of equation 14 is the sum of squared errors,
The right side is a quadratic expression of unknowns α and β. However, s of the number 14
_i is a value obtained by Expression 13, and N represents the total number of points (ui, vi) or the total number of small blocks shown in FIG.
Therefore, α and β that minimize Equation 14 should be obtained. Α and β that minimize Equation 14 can be obtained by solving Equations 15 and 16 below in parallel.

[0068]

(Equation 15)

[0069]

[Equation 16]

Next, the decoding process of the data compressed in the compression process shown in FIG. 1 will be described. FIG. 2 is a diagram showing a decoding process. In the process of decoding the compressed data storage file 3 in FIG. 2, first, 43 bits are extracted from the compressed data storage file 3 and the 11 bits from the beginning are sent to the block generator 41 as the DC component of the DCT coefficient. On the other hand, all the remaining 32 bits are the searcher (inverse vector quantizer) 42 in FIG.
Is input to

Then, the 8 bits from the beginning of the 32 bits input to the searcher (inverse vector quantizer) 42 in FIG. 2 are respectively the first 8 bits, the second 8 bits, and the third 8 bits.
If the bit is the fourth 8 bits, the index matching the first 8 bits is searched from the indexes of the codebook 101 in FIG. 2, and if the matching index is found, the quantized representative vector of the index is output. To do. Similarly, the second 8 bits, the third 8 bits, and the fourth 8 bits are codebooks 102, 103, and 103 in FIG. 2, respectively.
When 104 is searched and an index that matches them is found, the quantized representative vector of that index is sequentially output.

The block generator 41 in FIG. 2 generates the class 1 data group shown in FIG. 12 from the first vector output from the searcher (inverse vector quantizer) 42 in FIG. The data groups of class 2, class 3, and class 4 shown in FIG. 13 are generated from the second vector, the third vector, and the fourth vector, respectively. Then, the small blocks shown in FIG. 11 are organized from the four classes thus generated and the DC components previously input.

Search device (inverse vector quantizer) 4 in FIG.
The correspondence between the elements of the first vector which is the output of No. 2 and the data group of class 1 generated from the vector is shown in FIG.
(Symbols a to o). Further, FIG. 13 shows elements of the second vector, the third vector, and the fourth vector, which are outputs of the searcher (inverse vector quantizer) 42 in FIG. 2, and the class 2 and the class generated from the elements. 3 shows the correspondence with the data group of class 4 (symbol A to
P).

In this way, the small blocks generated by the block generator 41 in FIG. 2 are input to the IDCT converter 40. The IDCT converter 40 performs the conversion given by the above equation 8 on the input small block,
The result is sent to the block synthesizer 6 in FIG. Equation 8 is the inverse transformation equation of Equation 4. The block synthesizer 6 in FIG. 2 sequentially synthesizes the input small blocks into the positions of the numbers shown in the small blocks in FIG.

The decoding process of the compressed data storage file 4 in FIG. 2 is the same as the decoding process of the compressed data storage file 3 in FIG. The predictor 81 in FIG. 2 reads the DC component of the U data, substitutes it into the variable u in Equation 6 and calculates and outputs the predicted value given by the variable v.

Next, the decoding process of the compressed data storage file 5 in FIG. 2 will be described. First, from the compressed data storage file 5 in FIG. 2, the compressed data (second compressed data, etc.) 3
9 bits are taken out, 7 bits from the beginning are input to the adder 91 in FIG. 2, and the remaining 32 bits are input to the searcher (inverse vector quantizer) 62 in FIG. The searcher (inverse vector quantizer) 62 in FIG. 2 outputs four vectors in the same manner as the searcher (inverse vector quantizer) 42.
The adder 91 in FIG. 2 adds the input value from the compressed data storage file 5 in FIG. 2 and the predicted value that is the input from the predictor 81 in FIG. 2 and outputs this as added data. After that, from the block generator 61 to the block synthesizer 8 in FIG.
Up to the above, the compressed data storage file 3 in FIG. 2 described earlier
This is the same as the decoding process of.

By repeating the above steps until there is no data in the compressed data storage files 3, 4, 5 in FIG. 2 in this way, the block synthesizer 6, block synthesizer 7, and block synthesizer 8 in FIG. As shown in FIG. 7, Y data, U data, and V data having the same data number as the original image data 1 in FIG. Color converter 7 in FIG.
Reference numeral 1 denotes a color signal converter which gives the Y data, U data, and V data created by the block synthesizer 6, block synthesizer 7, and block synthesizer 8 in FIG. To obtain a color signal component (RGB three primary colors). The color converter 71 outputs the R data, the G data, and the B data one by one in the order of RGB to become the decoded image data 2 in FIG.

As described above, in this embodiment, the U data D
Using the correlation between the DC component of the CT coefficient and the DC component of the DCT coefficient of the V data, the DC of the DCT coefficient of the V data is used.
Since the prediction error is quantized instead of the component, the compression rate can be improved without degrading the image quality.

{Modification} (1) In the above embodiment, D of the DCT coefficient of U data
The DC component of the DCT coefficient of V data is predicted from the C component and V
The prediction error for the data was calculated, but from the DC component of the DCT coefficient of the V data to the DC of the DCT coefficient of the U data
The components may be predicted and the prediction error regarding the U data may be calculated.

[0080]

According to claim 1 and claim 3 of the present invention, a predicted value of one component of the second converted data is calculated from one component of the first converted data, and the predicted value and the second converted value are calculated. Since the amount of data is reduced by obtaining the error from one component of the data, there is an effect that the compression rate can be dramatically improved.

According to claim 2 and claim 3 of the present invention, when the second compressed data having the improved compression ratio as in claim 1 is restored, the first converted data is converted to the second converted data. The predicted value of one component of the
Since it is added to the inverse quantized data of and then subjected to the inverse orthogonal transform, the first color difference data and the second color difference data before compression can be restored in almost perfect form, and the image quality deterioration due to the improvement of the compression rate The effect is that it can be prevented.

[Brief description of drawings]

FIG. 1 is a block diagram showing a compression device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a decompression device according to an embodiment of the present invention.

FIG. 3 is a diagram showing a codebook for class 1 according to an embodiment of the present invention.

FIG. 4 is a diagram showing codebooks for class 2, class 3 and class 4 according to an embodiment of the present invention.

FIG. 5 is a diagram showing an original image according to an embodiment of the present invention.

FIG. 6 is a diagram showing an original image according to the embodiment of the present invention.

FIG. 7 is a diagram showing Y data, U data, and V data in an example of the present invention.

FIG. 8 is a diagram showing a method of dividing Y data, U data, and V data into small blocks according to an embodiment of the present invention.

FIG. 9 is a diagram showing a small block according to an embodiment of the present invention.

FIG. 10 is a diagram showing an order of compressing small blocks according to an embodiment of the present invention.

FIG. 11 is a diagram showing a DCT coefficient and its dividing method according to an embodiment of the present invention.

FIG. 12 is a diagram showing correspondence between a data group of class 1 and elements of a vector according to an embodiment of the present invention.

FIG. 13 is a diagram showing correspondence between data groups of class 2, class 3 and class 4 and elements of vectors in one embodiment of the present invention.

FIG. 14 is Y data and U in one embodiment of the present invention.
It is a figure which shows the structure of the compressed data of the DCT coefficient of data.

FIG. 15 is a DCT of V data according to an embodiment of the present invention.
It is a figure which shows the structure of the compressed data of a coefficient.

FIG. 16 is a diagram showing a correlation between the DC component of the U DCT coefficient and the DC component of the V DCT coefficient according to the embodiment of the present invention.

FIG. 17 is a diagram showing an effect of the predictor in the embodiment of the present invention.

FIG. 18 is a diagram showing a compression process of a conventional example.

[Explanation of symbols]

 1 original image data 2 decoded image data 3 compressed data 4 compressed data 5 compressed data 6 block combiner 7 block combiner 8 block combiner 10 block divider 11 DCT converter 12 class divider 13 scalar quantizer 14 scalar quantizer 14 vector quantizer 20 block divider 21 DCT converter 22 class divider 23 scalar quantizer 24 vector quantizer 30 block divider 31 DCT converter 32 class divider 33 scalar quantizer 34 vector quantizer 40 IDCT converter 41 Block generator 42 Search device 50 IDCT converter 51 Block generator 52 Search device 60 IDCT converter 61 Block generator 62 Search device 80 Predictor 81 Predictor 90 Subtractor 91 Adder 101-104 Codebook 201-204 Codebook Three 01-304 codebook

Claims (3)

[Claims] 1. A first color difference data representing a color difference and a second color difference data are orthogonally transformed to obtain a first transformed data and a second transformed data, respectively.
And an orthogonal transformer for obtaining the converted data of the first conversion data, and a second component that correlates from one component of the first conversion data
A predictor that predicts one component of the converted data, a subtracter that calculates an error between the one component of the second converted data and the output from the predictor, and a first subtractor that encodes the first converted data. Compression unit that obtains compressed data of 1) and a second quantization unit that obtains second compressed data by including error coded data obtained by encoding the error obtained by the subtractor. apparatus. 2. A method for decompressing the first compressed data and the second compressed data compressed by the compression device according to claim 1, wherein the first compressed data is dequantized to dequantize the first compressed data. A first dequantization section for obtaining dequantized data of 1; a second dequantization section for dequantizing the second compressed data to obtain second dequantized data; A predictor for predicting a predicted value of one component of the second transformed data from one component of compressed data or one component of the first dequantized data; and the second from the second dequantization unit. An adder that adds the predicted value of one component of the second converted data obtained by the predictor to the inverse quantized data to obtain added data; and the first dequantization unit from the first dequantization unit. Inverse quantized data for inverse orthogonal transforming the inverse quantized data and the addition data from the adder, respectively. Decompression device and a converter. 3. A first color difference data and a second color difference data representing the color difference.
And a decompression device for decompressing the data compressed by the compression device, wherein the compression device orthogonally transforms the first color difference data and the second color difference data, respectively. An orthogonal transformer that obtains first transformed data and second transformed data, and a second component that correlates from one component of the first transformed data
A predictor that predicts one component of the converted data, a subtracter that calculates an error between the one component of the second converted data and the output from the predictor, and a first subtractor that encodes the first converted data. And a second quantizer for obtaining second compressed data by including error encoded data obtained by encoding the error obtained by the subtracter, The decompressor dequantizes the first compressed data to obtain first dequantized data, and a first dequantization unit dequantizes the second compressed data to dequantize the second dequantized data. A second dequantization unit that obtains the quantized data, and predicts a predicted value of the one component of the second converted data from one component of the first compressed data or one component of the first dequantized data A predictor and the second dequantized data from the second dequantizer An adder that adds the predicted value of one component of the second converted data obtained by the predictor to obtain added data; and the first dequantized data from the first dequantization unit. A compression / expansion processing system, comprising: an inverse orthogonal transformer that inversely orthogonally transforms the addition data from the adder.