JPH0877360A - Image data compressing method - Google Patents

Abstract

PURPOSE: To reduce the calculated variable required for compressing image data. CONSTITUTION: An input image is divided into blocks in block dividing processing S1, and the average value of picture elements is calculated in average value calculating processing S2. First distribution is calculated in first distributed calculating processing S3, and an input color component distribution vector is calculated in distributed vector constituting processing S4. An input color component difference vector is calculated in input color component difference vector constituting processing S5, and the synthetic value of respective picture elements is calculated in color component synthetic value calculating processing $6. Second distribution is calculated in second distribution calculating processing S7, and a first color component converting vector and a first synthetic color component are calculated in selecting processing S8. A residual vector is calculated in residual vector constituting processing S9, and a maximum vector is calculated in maximum vector calculating processing S10. A second color component converting vector is calculated in second color component converting vector constituting processing S11, and a second synthetic color component is calculated in second synthetic color component calculating processing S12. An average value, first and second synthetic color components and first and second color component converting vectors are outputted in output processing S13.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image data compression method used in an image storage device for compressing and storing data such as color still images.

[0002]

2. Description of the Related Art Conventionally, techniques in such a field include:
For example, some documents were described in the following documents. Reference: IEICE Technical Report IE92-74 (1992-11), Japan, Shinichi Sakaida, "Data compression of images using correlation between three color signals by principal component analysis" P.63-68 In the conventional image data compression method described in, when a color image is divided into blocks of n × n (n; any natural number) pixels, a strong correlation is observed between the color components in each block. This is the focus. That is, the correlation existing between the color components is removed by the principal component analysis method which is a method of multivariate analysis, and the image in the block is represented by the principal component which is one or two composite color components. this is,
The image data is compressed by reducing the principal component having a small variance. Therefore, even when the image is restored using only the first main component that is one composite color component, a reproduced image with relatively little deterioration from the original image can be obtained.

The specific processing procedures (1) to (4) performed for each block will be described below. (1) A variance / covariance matrix for a plurality of color components expressing an input image is obtained. (2) The eigenvalue problem of the variance / covariance matrix obtained in (1) above is solved, and its eigenvalue and eigenvector are obtained. (3) Color coordinate conversion is performed by taking the product of the color component vector having each color component as an element and the conversion matrix having the eigenvector obtained in (2) above as a row vector. However, the color component vector is standardized by subtracting the average value of each color component before conversion. The converted color components have no correlation with each other. Here, among the color components after conversion, the component having the largest variance is referred to as the first principal component, and the component having the second largest variance is referred to as the second principal component. (4) The following data D1 to D3 are stored as compressed data. D1; only the first principal component, or the first and second principal components D2; transformation matrix D3; average value of each color component in the input image

[0004]

However, the conventional image data compression method has the following problems. That is, since the matrix eigenvalue problem must be solved for each block in the image, there is a problem that the amount of calculation becomes large. Especially in high-definition images such as printed images,
Since the number of pixels is large, the amount of calculation becomes enormous. The present invention provides an image data compression method that efficiently reduces the correlation between color components without solving the eigenvalue problem of a matrix in order to solve the problem of the amount of calculation required for the color coordinate conversion described above. .

[0005]

In order to solve the above problems, the present invention provides a block dividing process for dividing an input image into a plurality of blocks, and an input color component of pixel data of pixels forming the blocks. An average value calculation process for obtaining an average value and a first variance calculation process for obtaining a first variance, which is a variance for each input color component of pixel data of pixels forming the block, are performed. Further, a dispersion vector constructing process for constructing an input color component variance vector obtained by normalizing a vector having the first variance as an element, and a value of each of the input color components of each pixel data in the block An input color component difference vector is obtained by subtracting the average value corresponding to the input color component from each of the input color component difference vectors, and an input color component difference vector configuration process that configures an input color component difference vector having the elements as the elements, The color component composite value calculation process for obtaining the composite value of the input color component of each pixel in the block by taking the inner product of the input color component dispersion vector and the input color component difference vector, and the dispersion of each composite value A second variance calculation process for obtaining a certain second variance is performed. Then, a maximum variance is obtained from the first variance and the second variance, and a value corresponding to the maximum variance is calculated from the values of the input color component differences and the composite value. A unit vector in which each pixel is selected as a first composite color component and has the same number of elements as the number of color components of the input color component, and only one element value is 1 and the remaining element values are 0, or A selection process for determining a first color component conversion vector when converting the input color component dispersion vector to the first composite color component is performed.

Next, the input color component difference vector is calculated by subtracting the product of the value of the first composite color component and the first color component conversion vector from the input color component difference vector. Residual vector configuration processing that configures a residual vector that represents an error when approximated for each pixel, and a norm of the residual vector is calculated for each pixel, and a maximum vector that maximizes the norm is obtained. A maximum vector calculation process and a second color component conversion vector construction process for normalizing the maximum vector to obtain a second color component conversion vector;
A second composite color component calculation process for calculating a second composite color component for each pixel by taking an inner product of the input color component difference vector and the second color component conversion vector,
Finally, an output process of outputting the average value, the first composite color component value, the first color component conversion vector, the second composite color component value, and the second color component conversion vector as compressed data is performed.

[0007]

According to the present invention, since the image data compression method is configured as described above, the input image is divided into a plurality of blocks by the block division processing. Next, an average value calculation process obtains an average value for each input color component of the pixels forming the block based on the pixel data of the input image, and further, a first variance calculation process determines the input image of the input image. The first variance, which is the variance for each input color component of the pixels in the block, is obtained based on the pixel data. Next, the variance vector configuration process configures an input color component variance vector obtained by normalizing the vector having the first variance as an element for each block. By the input color component difference vector construction processing, the input color component difference is obtained by subtracting the respective average values corresponding to the respective input color components from the values of the respective input color components of the pixels in the block, The input color component difference vector to be formed is configured for each pixel.
Further, the color component composite value calculation process calculates the inner product of the input color component variance vector and the input color component difference vector to obtain the composite value of the input color components of each pixel in the block. In addition, the second variance, which is the variance of each composite value, is obtained by the second variance calculation process. By the selection processing, the maximum variance is obtained from the first variance and the second variance, and the value corresponding to the maximum variance is calculated from the values of the input color component differences and the composite value. A unit that is selected for each pixel and becomes the first composite color component for each pixel, has the same number of elements as the number of color components of the input color component, and has only one element value of 1 and the remaining element values of 0. A first color component conversion vector for converting the vector or the input color component dispersion vector into the first composite color component is determined.

On the other hand, in the residual vector construction processing, the product of the value of the first composite color component and the first color component conversion vector is subtracted from the input color component difference vector to obtain the first composite color component difference vector. A residual vector representing an error when approximated by color components is configured for each pixel. Next, the norm of the residual vector is calculated for each pixel by the maximum vector calculation process, the maximum vector having the maximum norm is obtained, and the maximum vector is normalized by the second color component conversion vector configuration process. And becomes a second color component conversion vector. Further, the second composite color component calculation process calculates the second composite color component for each pixel by taking the inner product of the input color component difference vector and the second color component conversion vector. Finally, by the output processing, the average value, the value of the first composite color component, the first color component conversion vector, the value of the second composite color component, and the second color component conversion vector are output as compressed data. That is, in the present invention, the first composite color component and the second composite color component
The composite color component can be obtained by a relatively simple calculation. Therefore,
The above problems are solved.

[0009]

FIG. 2 is a functional block diagram of an image data compression / decompression device for carrying out the image data compression method according to the embodiment of the present invention. This image data compression / decompression device has a color coordinate converter 10. The color coordinate converter 10 includes a block divider 11 and a color component converter 12. The block divider 11 includes red, green, blue, cyan, magenta,
Input image data IN having a plurality of color components such as yellow and black
The input color component D11 is generated by dividing into a plurality of blocks each having × n (n; any natural number) pixels. The output side of the block divider 11 is connected to the input side of a color component converter 12 that generates the compressed data D12 in block units by utilizing the correlation between the color components of the input color component D11. The color component converter 12 includes a first composite color component generator 12a and a second composite color component generator 12b, and has an input color component D11.
Is input to both sides. The first composite color component generator 12a receives the input color component D11 and generates the first composite color component pi and the first color component conversion vector Lp. The second composite color component generator 12b inputs the input color component D11, the first composite color component pi, and the first color component conversion vector Lp to generate the second composite color component qi and the second color component conversion vector Lq. To do. The first composite color component pi, the first color component conversion vector Lp, the second composite color component qi, and the second color component conversion vector Lq are adapted to be added at the output side of the color component converter 12. A storage device 20 for storing the compressed data D12 is connected to the output side of the color component converter 12, and a color coordinate inverse converter 30 is connected to the output side thereof. Color coordinate inverse converter 30
Is a color component reverser 31 that generates a reproduction color component D31 in block units from the compressed data D12 stored in the storage device 20, and a reproduction color component D31 arranged at the same position as the input image data IN. And a block arranging device 32 for generating data D32. Each of these functional blocks is composed of, for example, a digital circuit such as an adder or a multiplier.

Next, an image data compression method and the like using the image data compression / decompression device of FIG. 2 will be described. FIG. 1 is a flowchart of an image data compression method showing an embodiment of the present invention. The processing contents of the color coordinate converter 10 in FIG. 2 will be described with reference to FIGS. 1 and 2. The present embodiment is an image data compression method for an input image having an arbitrary number of color components of two or more colors. As a typical example, an image having three input color components is input. The case will be described. In the block division processing S1 of FIG. 1, the block divider 11 of FIG. 2 divides the input image data IN into blocks of n × n pixels, and outputs the three input color components D11 of the input image data IN to each block. Color component converter 12 for each
To enter. The color component converter 12 receives the three input color components D
11 (Three color input color components are a, b, and c, respectively)
To generate the first and second composite color components in block units. The processing procedure (S2 to S13) will be described below. First, the first composite color component generator 12a in the color component converter 12
The processing procedure (S2 to S8) will be described. In the average value calculation process S2, the average value E (x) of the input color component x of each pixel in each block (where x
= A, b, c) is calculated for each input color component.

[0011]

[Equation 1] In the first variance calculation process S3, the variance v of the input color component x of each pixel in each block is calculated using the following equation (2).
_x (however, x = a, b, c) is calculated for each input color component.

[0012]

[Equation 2] In the dispersion vector construction process S4, the input color component dispersion vector F is obtained for each block using the following equation (3).

[0013]

(Equation 3) In the input color component difference vector construction process S5, the input color component difference vector X _i for each pixel is obtained for each block using the following equation (4).

X _i = [a _i −E (a) b _i −E (b) c _i −E (c)] (4) where i = 1, 2, ..., N × n In the color component composite value calculation process S6, the composite value y _i of the input color component is obtained for each pixel in the block using the following equation (5). This combined value y _i is the input color component difference vector X
_It means a weighted sum of the elements that make up _i . y _i = F · X _i ^T (5) where i = 1, 2, ..., N × n ·; product of matrix operations T; transposition of vector In the second distributed calculation process S7, The variance v of the input color component with respect to the composite value y is calculated for each block using the equation (6).
_{Find y} .

[0015]

[Equation 4] In the selection process S8, the first composite color component p _i and the first color component conversion vector Lp are determined using the following equation (7). That is, each variance v _x (x = a, b, c) and variance v _y
The maximum variance is calculated from among the values, and the values of the respective input color component differences {a _i -E (a)}, {b _i -E (b)}, {c _i -E
(C)} and the composite value y _i , the value corresponding to this maximum variance is selected for each pixel, and the selected value becomes the first composite color component p _i corresponding to each pixel, and , The input color component difference vector X _i of each pixel is set to the first composite color component p.
The first color component conversion vector Lp corresponding to the vector to be converted to _i is determined. When v _a > v _y , v _a > v _b , v _a > v _c p _i = a _i −E (a) v _b > v _y , v _b > v _a , v _b > v _c p _{i = b i -E (b) v} c> v y, v c> v a, v c> v b p i = y i However, when other than _{p i = c i -E (c} ) above when, i = 1, 2, ..., When n × n v _a > v _y , v _a > v _b , v _a > v _c Lp = [1 0 0] v _b > v _y , v _b > v _a , v when _{b> v c Lp = [0} 1 0] v c> v y, v c> v a, v c> v when _b Lp = [0 0 1] when other than the above Lp = F · · · ( 7) However, p _i ; the value of the first composite color component in each pixel in the block Lp; the first color component conversion vector Next, the processing procedure of the second composite color component generator 12b (S9-
S12) will be described below.

In the residual vector construction process S9, the residual vector U _i is obtained using the following equation (8). The residual vector U _i represents an error when the input color component difference vector X _i is approximated by the first combined color component p _i . _{U i = X i -p i Lp} ··· (8) where, i = 1,2, ···, n × n Here, the _{U i = [r i s i} t i]. In the maximum vector calculation process S10, _i that maximizes the norm ‖U _i ‖ of the vector U _i representing the magnitude of the vector U _i expressed by the following equation (9) is obtained. ‖U _i ‖ = (r _i ² + s _i ² + t _i ² ) ^1/2 (9) Now, when i = m, it is assumed that ‖U _i ‖ becomes maximum. That is, ‖U _m ‖ = max‖U _i ‖ (i = 1, 2, ..., N × n).

In the second color component conversion vector construction processing S11, the second color component conversion vector Lq is obtained using the following equation (10). The second color component conversion vector Lq is a vector that converts the input color component difference vector X _i into the second composite color component q _i .

(Equation 5) In the second combined color component calculation process S12, the following (11)
The second combined color component q _i at each pixel in the block is calculated using the formula.

Q _i = Lq · X _i ^T (11), where i = 1, 2, ..., N × n T; transpose of vector; product of matrix operation In output process S 13, The data d1 to d3 are stored in the storage device 20 as compressed data D12 with accuracy as required.
Save to. d1; first composite color component p _i and second composite color component q _i (where i = 1, 2, ..., N × n) d2; first color component conversion vector Lp and second color component conversion Vector Lq d3; average value E (x) of input color component x (where x = a,
b, c) FIG. 3 is a flowchart showing the operation of the color coordinate inverse converter 30 in FIG. The processing contents of the color coordinate inverse converter 30 in FIG. 2 will be described with reference to FIGS. 3 and 2. The color component inverse converter 31 uses the compressed data D stored in the storage device 20.
The reproduction color component D31 is generated from 12 in block units.
The processing procedure (S14 to S16) will be described below.

In the reproduction value calculation process S14 of FIG. 3, the reproduction value z (x) (where x = a, b, c) of the input color component is obtained for each pixel using the following equation (12). Z _i = p _i L _i + q _i Lq + K (12) where, Z _i = [z (a _i ) z (b _i ) z (c _i )] (where i = 1, 2, ... , N × n) K = [E (a) E (b) E (c)] z (x _i ); input color component x at each pixel in the block
Reproduction value (where x = a, b, c) p _i ; value of first composite color component q _i at each pixel in the block; value of second composite color component L p at each pixel in the block; first Color component conversion vector Lq; Second color component conversion vector E (x); Average value of input color component x in the block (where x = a, b, c) In the reproduction value output process S15, in each pixel in the block The reproduction value z (x _i ) of the input color component x (where x =
a, b, c i = 1, 2, ..., N × n) is input to the block allocator 32 as the reproduction color component D31. In the block placement process S16, the block placement unit 32
The reproduced color component D31 input in block units is arranged at the same position as the input image, and the reproduced image D32 is output.

As described above, in this embodiment, the first composite color component p _i when the variance of the input color component D11 of the input image data IN becomes maximum is simpler than that in the prior art principal component analysis method. Since the input color components are calculated by the variance v _x , and the second composite color component q _i is calculated based on the first composite color component p _i , the calculation amount required for color coordinate conversion is Significantly reduced compared to. This embodiment is J
When combined with an encoding method such as PEG (Joint Photographic Coding Experts Group), it is possible to realize image data compression with higher efficiency and higher compression rate. The present invention is
The present invention is not limited to the above embodiment, and various modifications are possible. The following are examples of such modifications. (A) The storage device 20 in FIG. 2 may be transmission means such as a transmission line. (B) The n × n pixel block in the above embodiment is n × n.
It may be m (a natural number different from m; n) pixels. (C) In the embodiment, the compressed data D12 is composed of the first composite color component p _i , the second composite color component q _i , the first color component conversion vector Lp, the second color component conversion vector Lq, and the input color component x. Although the average value E (x) is used, the first composite color component p _i , the first color component conversion vector Lp, and the input color component x
The average value E (x) of In this case, the equation (12) becomes the following equation (13). Z _i = p _i L _i + K (13)

[0021]

As described above in detail, according to the present invention, block division processing, average value calculation processing, first dispersion calculation processing, dispersion vector construction processing, input color component difference vector construction processing, color components By performing a composite value calculation process, a second variance calculation process, a selection process, a residual vector configuration process, a maximum vector calculation process, a second color component conversion vector configuration process, a second composite color component calculation process, and an output process. , The first composite color component in the case where the variance of the input color components of the input image is maximum is obtained by an arithmetic operation based on the variance of the input color components, which is simpler than the principal component analysis method in the prior art. Since the second composite color component is obtained based on the component, the amount of calculation required for color coordinate conversion can be significantly reduced compared to the conventional case. That is, in the present invention, the amount of calculation required for color coordinate conversion is reduced by the amount that does not require the following calculation for determining the amounts (a) and (b) in the prior art. (A) Covariance of a plurality of color components expressing an input image. (B) Variance of input image data for a plurality of color components
Eigenvalues and eigenvectors of the covariance matrix.

[Brief description of drawings]

FIG. 1 is a flowchart of an image data compression method showing an embodiment of the present invention.

FIG. 2 is a functional block diagram of an image data compression / decompression device showing an embodiment of the present invention.

FIG. 3 is a flowchart of an image data restoration method showing an embodiment of the present invention.

[Explanation of symbols]

Reference Signs List 10 color coordinate converter 11 block divider 12 color component converter 12a first composite color component generator 12b second composite color component generator 20 storage device 30 color coordinate inverse converter 31 color component inverse converter 32 color component arranger S1 Block division process S2 Average value calculation process S3 First variance calculation process S4 Variance vector construction process S5 Input color component difference vector construction process S6 Color component composite value calculation process S7 Second variance calculation process S8 Selection process S9 Residual vector Configuration process S10 Maximum vector calculation process S11 Second color component conversion vector configuration process S12 Second composite color component calculation process S13 Output process

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Office reference number FI Technical display location H04N 1/46 H04N 1/46 Z

Claims (1)

[Claims] 1. A block division process for dividing an input image into a plurality of blocks, an average value calculation process for obtaining an average value for each input color component of pixel data of pixels forming the block, and forming the block. A first variance calculation process for obtaining a first variance, which is a variance for each input color component of pixel data of a pixel, and an input color component variance vector obtained by normalizing a vector whose elements are the first variance. Dispersion vector configuration processing configured for each block, the input color component difference is obtained by subtracting the average value corresponding to the input color component from the value of each input color component of each pixel data in the block, An input color component difference vector forming process for forming an input color component difference vector as an element for each pixel, the input color component dispersion vector and the input color component difference vector Color component composite value calculation processing for obtaining the composite value of the input color components of each pixel in the block by calculating the inner product with the second dispersion, and the second dispersion for obtaining the second dispersion which is the dispersion of each composite value. The maximum variance is calculated from the calculation process and the first variance and the second variance, and the value corresponding to the maximum variance is calculated from the values of the input color component differences and the composite value. A unit vector in which each pixel is selected to be the first composite color component for each pixel, has the same number of elements as the number of color components of the input color component, and only one element value is 1 and the remaining element values are 0. Alternatively, the first when converting from the input color component dispersion vector to the first composite color component
A selection process for determining a color component conversion vector; and subtracting the product of the value of the first composite color component and the first color component conversion vector from the input color component difference vector to obtain the input color component difference vector Residual vector construction processing for constructing, for each pixel, a residual vector that represents an error when approximated by one composite color component, and a norm of the residual vector is calculated for each pixel, and the norm is the maximum. Vector calculation processing for obtaining a maximum vector, a second color component conversion vector configuration processing for normalizing the maximum vector into a second color component conversion vector, the input color component difference vector and the second color component conversion vector A second composite color component calculation process for calculating a second composite color component for each of the pixels by taking an inner product of, and an average value, a value of the first composite color component, and a first color component conversion Vector and an output process for outputting a value of the second composite color component and second color component conversion vector as compressed data, the image data compression method comprising applying.