JPH02215277A - Color compressing and restoring device - Google Patents

Abstract

PURPOSE:To attain the optimum compression of a color image by repeatedly executing a scanning over the whole picture with the use of a segmenting means before encoding and simultaneously, giving a real chrominance signal at such a time to a neutral circuit network type means at the same time and optimize the internal parameter through case-by-case study and restore color information with the use thereof. CONSTITUTION:For example, a color component 20 among three-color components 20, 21 and 22 of the color image is used for a reference color, it is inputted as a block signal B of nXm to a neutral circuit network type means 11, simultaneously, color synthesizing components G for executing a predictive estimation are inputted to the means 11, and a laboratory such as a back propagation algorithm is executed. Inside, the circuit network type composed of the combinations of neutral dummy elements 111 is embodies by a hardware or software, and such an internal condition parameter C is sent to be outside as data through a parameter reading means 11a. Namely, the block of nXm picture elements is segmented as to the featured reference color to express the color image, the scanning is executed one by one picture element over the whole picture, and the internal parameter is made optimum.

Description

[Detailed Description of the Invention] Industrial Application Field The present invention is particularly applicable to the field of deriving high-quality codes from a small amount of information, such as a color image storage device or a color compression/restoration device that can be realized as a color facsimile. It is related to.

Conventional technology Conventionally, the three primary color signals are considered as independent variables,
There has been little research on compression and encoding of color images using the correlation of three primary colors. However, in recent years, research has been conducted on the reproduction of color images from monochromatic images by utilizing local color image correlation. (Reference: Koharu et al., Predictive reproduction of full-color images from monochromatic images using correlation between three-color signals, 1986 Institute of Electronics, Information and Communication Engineers General National Conference) This method is explained based on Fig. 5. Reference numerals 51, 52, and 53 in FIG. 5 indicate three primary color originals (red, green, and blue, or cyan, magenta, and yellow) of a color image. 52 as the reference color data, an n×m block (
FIG. 5 shows a 3×3 example. ), calculation units 54 and 55 calculate prediction coefficients regarding the color data of predetermined polynomials 51 and 53 based on the reference color data.

The reference color data and the above prediction coefficients for each block are sent to the next stage, and a prediction function section 56.57 predicts the input data block by block using a prediction function, and the data section 58.59 reproduces the data.

Problems to be Solved by the Invention However, in the prior art described above, an adaptive prediction function is obtained for each image block, but the shape of the function is a predetermined polynomial, and approximation by nonlinear combinations of these functions cannot be handled. Therefore, there was a problem that it was not possible to optimize the approximation based on the image.

In view of the above problems, the present invention provides a color compression/restoration device that is adaptively optimal for color images.

Means for Solving the Problem Therefore, the present invention provides an image cutting means for cutting out a block of n×m pixels from a target color image with respect to a characteristic reference color representing the image, and each pixel in the cut out block. Neural network model means estimates and outputs the color of a pixel at a certain position in the block by inputting representative n×m color values representing the block; The neural circuit includes a scanning means for scanning one pixel at a time over the area, a parameter reading means for reading out internal state parameters of the neural circuit, and a parameter means for resetting the already read internal state parameters.

Effect of the Invention With the above configuration, the present invention repeatedly scans the extraction means over the entire screen before encoding, and simultaneously gives the true color signal at that time to the neural network model means to adaptively adjust its internal parameters. Learning optimization is performed, and then the values of the internal parameters of the learned neural network and the data of the reference color are output data, and the color information based on the data is restored.

EXAMPLE An example of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of a color compression/restoration device according to an embodiment of the present invention. In FIG. 1, 20.
21 and 22 are the three components of the color image to be compressed. Hereinafter, 20 color components will be used as reference colors. The signal A of 20 is processed by the image cutting means 10 into n×m (
1 over the entire screen for each block (n, m are positive integers)
The signal is scanned pixel by pixel and cut out as an n×m signal, resulting in signal B.

The signal B and the color component signals 21 and 22 are input to the neural network model means 11, and learning is repeated until the optimal internal state parameters of the neural network are determined over the entire screen of the image. There are various learning algorithms, such as the backpropagation algorithm (references:
[Rummelhart, Day, E. and Matsukriland, J.E., L.

1, 1,
2 MIT Press Cambridge (198
6) ); Runmelhart, D., E.
and McClelland, J.L., ”
Parallel Distributed Proc
e-ssing, Exploration in t
he Microstructure of Cogn
iti-on, Vol, 1, 2, MIT
Press, Cambridge (1986))
Find the optimal solution using the lowest descent method. The determined internal state parameter C is sent to the transfer device 12.13 along with the reference color.
The reference colors are transmitted to the receivers 15 and 16 via the communication line 14, and the reference colors are cut out by a means 10a similar to the image cutting means 10 described above for each n×m block, and input to the neural network model means 17. be done.

At this time, the state parameter E is the neural network schematic means 17
is set as a parameter of the internal state of

As a result, the neural network model means 17 outputs the three color components F estimated from the reference colors. In addition, the standard color is
In the configuration shown in FIG. 1, one of the three color components is used, but the following signal Y generated from the three color components may also be used.

[Here, F(') is the reference color generation function, R (red).

G (green), B (blue), C (cyan), 2M (magenta),
It is a color component of Y (yellow). ] The signal Y includes, for example, the average value of three color components and the brightness. Furthermore, one of the three color components and the signal Y described above may be used as the reference color.

FIG. 2 is a detailed diagram of the neural network schematic means 11 of FIG. 1. 20.21.22 are the three color components of a color image and are 20
When the reference color is set as the reference color, an n×m block signal B is input to the neural network model means 11. Further, the neural network model means 11 is inputted with a color component G for predictive estimation,
The above-mentioned backgropage algorithm, etc. will be learned. Inside, a neural network model constituted by a combination of neural simulation elements 111 is realized by hardware or software, and the internal state parameter C of the circuit network is sent to the outside as data via the parameter reading means 11a. issued. FIG. 3 shows the structure of the neural pseudo-element in the neural network model means 11.

"Neural pseudo-element" has input x1. x2.・・・・・・・・・
・, Xi and internal state parameters w1 j, w2 j
, . . . , the results of multiplying with WiJ are accumulated, and a nonlinear function F() is applied to output the result y.

FIG. 4 is a detailed explanation of the neural network model means 17 shown in FIG. After setting, the input standard color n×m
The color components are estimated based on the signal of the block, and the three color components Y
It is intended to reproduce.

Effects of the Invention With the above-described configuration, the present invention can adaptively and optimally compress a color image that is a target color image.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a color compression/reconstruction device according to an embodiment of the present invention, FIGS. 2 to 4 are block diagrams of main parts of the same device, and FIG. It is a main part block wiring diagram. 10, 10a... image cutting means, 11.17...
- Neural network schematic means, 20.21.22...color component. Name of agent: Patent attorney Shige Takashi Awano and one other person

Claims (4)

[Claims] (1) Image cutting means on the sending and receiving sides that cuts out a block of n×m pixels for a characteristic reference color representing the image from the target color image, and a representative block that represents each pixel in the cut out block. Neural network model means on the transmitting side and the receiving side that receives n×m color values as input and estimates and outputs the color of a pixel at a certain position within the block; and the block of n×m pixels. A transmitting side scanning means for scanning the entire screen pixel by pixel, a transmitting side parameter reading means for reading out internal state parameters of the transmitting side neural network model means, and resetting the internal state parameters that have already been read out. receiving side parameter setting means, and before compression on the transmitting side, the transmitting side image cutting means is repeatedly scanned over the entire screen by the scanning means, and the transmitting side neural network model means is configured to set the true color at that time. The signals are applied simultaneously to adaptively learn and optimize the internal parameters thereof, and then the values of the internal parameters of the neural network schematic means on the transmission side that have already been learned and the data of the reference color are output data, and the output is performed. A color compression/restoration device characterized by restoring color information on the receiving side using data. (2) The color compression system according to claim 1, characterized in that a signal of a certain color among the color components is always used as the reference color.
Restoration device. (3) The color compression/restoration device according to claim 1, wherein a signal representing a function of three color components is used as the reference color. (4) The color compression/restoration device according to claim 1, wherein both a signal of one constant color among the color components and a signal representing a function of three color components are always used as the reference color.