JPH01264092A - Compression coding system for color picture - Google Patents

Abstract

PURPOSE:To simplify a device by treating RGB signals representing the color of a picture element as a three-dimension vector and representing the information of the RGB signal by a 1st major component reflecting the information at a maximum. CONSTITUTION:A color picture is split into small areas and the major component is analyzed with respect to the RGB signal of the picture element included in each small area and each picture element is divided into two classes based on the score of the 1st major component of each picture element obtained as a result and each small area is approximated in 2 colors. Thus, in case of approximating the small area by a mean color optimum in the means of square error, a small area whose mean square error is smaller than a permissible value between a color vector and a mean color vector of each picture element included in the small area can be approximated by one color (mean color) only. That is, the small area able to be approximated by one color is approximated by one color and other small areas are approximated by 2 colors. Thus, considerable information compression is attained and the device is simplified.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a compression encoding method for color images, and particularly to an information compression encoding method for transmitting still color image data.

[Prior Art] Conventionally, this type of method includes a method in which RGB signals are encoded independently, or a method in which RGB signals are encoded independently, as shown in FIG.
A new coordinate system is created by converting the signal into a new coordinate system Coding a
nd Its Application to Co1
or Images,”IEEE Trans,COM
,,vol,com-32,No,10,0ctorb
er1984) is known.

[Problems to be Solved by the Invention] However, the conventional compression encoding method requires three sets of encoders and decoders, and has the disadvantage that it is difficult to reduce the size and cost of the device. was seen.

Therefore, in view of the above-mentioned points, an object of the present invention is to treat the RGB signal representing the color of a pixel as a three-dimensional vector, and to represent it with the first principal component that reflects the information of RGB to the maximum, thereby solving the above problem. It is an object of the present invention to provide a color image compression encoding system that can significantly simplify the apparatus by eliminating three sets of encoders and decoders that were required in the conventional system.

[Means for Solving the Problems] In the encoding method according to the present invention, a color image is divided into small regions, principal component analysis is performed on the RGB signals of pixels included in each small region, and each pixel obtained as a result of the principal component analysis is By dividing each pixel into two classes based on the score of the first principal component, each small region is approximated with two colors (see Figure 1 (a)).
.

Jusaku = Pack → With the encoding method according to the present invention, each small area is a bitmap (0/1) indicating which class each pixel belongs to.
and is encoded into the representative color of each class.

Furthermore, decoding can be realized by assigning a representative color to each pixel according to the bitmap (Figure 1(b)).
reference).

The above method is basic and approximates all small areas with two colors, but in reality it is not necessary to approximate all small areas with two colors, and most small areas , a one-color approximation is sufficient.

A small area smaller than a certain tolerance can be approximated with only one color (average color).

That is, small areas that can be approximated with one color are approximated with one color, and other small areas are approximated with two colors.

This allows considerable information compression.

Further information compression is possible by data compressing the bitmap and representative colors using conventional data compression techniques.

[Example] Hereinafter, the present invention will be described in detail based on Examples.

First, the general principle of the present invention will be explained.

In order to divide a color image into small areas, for example, divide the image into 4 x 4 small blocks.
2 = (R2, G2゜==J B2), ..., X', = (R,, G□BJ, 3'l-(J'('L-4) (i-1,2
s=, m) (1) (see Figure 4). However, the moose is the solution of the eigenvalue problem Σ law-λuL (2), and χ and Σ are the mean vector and covariance matrix, respectively, and can be calculated as follows. Here, ′ represents the transposition of the matrix (vector).

The eigenvalue problem in equation (2) is a third-order eigenvalue problem, and the solution can be found directly using the following method.
High-speed calculation is possible. Since the covariance matrix Σ is a 3×3 symmetric matrix, it becomes as follows. Here, bo-011-022033 C"σ■σ22+σ22σ33+σ33σI1-CI+
22-023"' (71:12 d:σ■σ232+σ22σ!32+033(7122
−σ11σ22033−2σ12σ23σI3.

The root of equation (3) is It can be calculated by here, o == 4p3 q2 is the maximum eigenvalue λ of the eigenvalue problem of equation (2).

That is, λ=max (input 1.λ2.λ3). Eigenvector group for maximum eigenvalue λ = (u,
u2u3) can be calculated by the following procedure.

C11= ((722-λ) (a,,-in)
0232C22= (σII−λ) (a 33−λ)
(7132c3s” (σ eyes-in) (σ22-λ)
−σ122C,2= (σ33−λ)σ12+σ1
3σ23013”-(σ22-λ)σ13+σ1202
゜C23 knee (σ■-in) 023 〇12σI3 If e11≠0, if c22≠0, if c33≠O, if C11+C22+C33 are all 0, Cl3-0
+C23-0+CI2≠0 then u1 abuse u2■0. u, = 1 If C11, C22, C33 are all 0, c, 2-0
．． c23-0゜If CI3≠0 then u, -u, -0, u2ml If CIl +C22+C33 are all 0 and Cl2-
If 0+Cl3-0+C23≠0, then u2-us-0, ul-1. Otherwise, the solution is indeterminate.

Normally, the solution is rarely indefinite, but if the solution is indefinite, for example, the Y coordinate of YIQ transformation can be used instead of the principal axis. In other words, 1.11'= (0
, 299, 0, 587, 0, 114).

The following method can be used to divide each pixel into two classes.

Method 1 (Division based on the average value of principal component scores) Since the average value of principal component scores is 0 according to equation (1), pixels are divided into two classes depending on the sign of the principal component score of each pixel (5th class). (see figure).

In practice, the above method is sufficient, but it is also possible to use the following method (Nobuyuki Otsu, "Automatic value calculation method based on discriminant and least squares criterion", Journal of the Institute of Electronics and Communication Engineers) J63-4, 349-356 (1980)).

yI≦y2≦...≦y.

shall be.

Find k (1<k<+n-1) that maximizes.

Pixels with principal component scores greater than yk are assigned to class 1. y, pixels with smaller principal component scores to class 2 Divide (see Figure 6).

Method 2 improves the reproduced image, albeit slightly.

Since it can be shown that the average color of each class is optimal in the sense of least square error, the average color of each class is set as the representative color of each class.

Whether a small area can be approximated with only one color is determined by the average 2 between the color vector of the pixels included in the small area and the average color vector.
When a multiplicative error small region can be approximated with only one color, all pixels in the small region are approximated to the average color X.

In one embodiment of the present invention, this method was successfully applied to a color image of 256 x 256 pixels.

Each pixel has 24 bits/8 bits each for R, G, and B.
It is expressed in pixels. If the small area is a 4 x 4 small block and all the small areas are approximated with two colors, each block (4
Assuming that the representative colors of 2 classes are represented by 24 bits (R, G, B, 8 bits each) x 2 classes = 48 bits,
Color images are compressed to 4 bits/pixel, or 1/6. Information can be further compressed by allowing small blocks to be approximated using only one color.

According to the present invention, compression of color image data can be achieved with very little memory and calculation time. Furthermore, it is easy to implement this method in hardware.

Further, the image reproduced by decoding is preferably robust against communication channel noise and has good reproducibility of image edges.

[Brief explanation of the drawing]

Figure 1 is a flowchart showing an overview of the present invention, Figure 2 is an explanatory diagram of a conventional color image compression encoding method, Figure 3 is an explanatory diagram showing the encoding method of the present invention, and Figure 4 is a diagram showing principal components. Flowchart showing a method for calculating scores. FIGS. 5 and 6 are flowcharts showing a method for dividing each pixel into two classes. Color Jl kLf) Touki Ichitake Go I and the 1st ＼ month ＼su b
] Fig. 1 - Number of numbers +'t't - Asahi j Sarame's 1 Raei Elephant's three-phase call and whistle to show t Fig. 2 To Jiro's waiting number Ib Te style r Figure 3 1. χ21...-2χm (Pler vector of 8 hoops) Main ft, Keaka's banquet'JF is not D
-Chart Figure 4 What is the main Peke score?
The flowchart shown in Figure 5 shows a minimum score of 2 points and 1 ball! [■
Flow diagram 6

Claims (1)

[Claims] (1) Divide the color image into small regions, perform principal component analysis on the RGB signals of pixels included in each of the small regions, and based on the score of the first principal component of each pixel obtained as a result, By dividing each pixel into two classes,
A color image compression encoding method characterized by approximating each small area with two colors.