JPH1051648A - Color image processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To execute plural kinds of color conversion and color correction at a neural network by switching their weight. SOLUTION: A reading part 1 reads a color image and a communication part 2 receives the color image. A neural network 3 executes color change and color-correction of read data or received data, the part 2 transmits this data and a recording part 4 prints this data. A data input arithmetic part 5 calculates data inputted to the neural network 3 to calculates high-order data RG, GB, RB, etc., from first data R, G and B corresponding to the constitution of the neural network 3. A weight switching part 6 sets weight corresponding to the kind of color change to the neural network.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a color image processing apparatus for performing color correction and color coordinate conversion using a neural network.

[0002]

2. Description of the Related Art In recent years, color image processing apparatuses such as color facsimile machines and color copying machines have been used. A color image is displayed on a display or printed by a printer. However, there is a large difference between the color of data displayed on the display and the color when the data is printed, which has been a problem. In order to solve this problem, there is a method in which correction is performed using a neural network. That is, an additive color mixture RGB signal of the original image is input to each input layer of the neural network, and learning is performed so that a subtractive CMY signal having no color shift from this input is output from each output layer unit. Also, some color image processing apparatuses use a neural network to correct nonlinearity in ink color mixture, input CMY signals to each input layer unit, and have no color shift after correction from each output layer unit. Learning is performed so that CMY signals are output.

[0003]

However, in such a color image processing apparatus using a conventional neural network, the conversion from the RGB color system to the CMY color system or C
The conversions and corrections such as the correction of the color shift of the MY color system have been performed independently. In addition, a color conversion circuit has been used to obtain a developed color signal such as L ^* a ^* b ^* which is indispensable for communication means and does not depend on a device to be used. Also, as inputs of the neural network, R, G,
Only B or L ^* , a ^* , b ^* was input, and these higher-order data, RG, R ² , RGB, etc. were not input, and the output accuracy was often poor.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned problems, and has as its object to provide a device capable of performing a plurality of types of color conversion with a neural network by switching weights. Also, the input to the neural network is not limited to the primary of R, G, B, etc., but also RG, R ²
It is an object of the present invention to improve conversion and correction accuracy by adding higher-order data such as data.

[0005]

In the color image processing apparatus of the present invention, input image data of each color system is converted into a color system suitable for output by converting weights. According to the present invention, it is possible to easily convert each color system into another color system by using a neural network.

[0006]

According to the first aspect of the present invention, a reading unit for reading a color image, a receiving unit for receiving a color image, a color image received by the receiving unit, and a color of the color image read by the reading unit. A neural network for performing image processing including conversion and color correction, input conversion means for converting a color image input to the neural network into data corresponding to the configuration of the neural network, and weights of the neural network for image processing. Weight switching means for switching according to, a transmission unit for transmitting an output of the neural network,
A recording unit for recording the output of the neural network.

The data read by the reading means and the data received by the receiving means are input to the neural network. The input conversion means converts the input data into data suitable for the configuration of the neural network and outputs the data to the neural network. The weight switching means outputs a weight corresponding to the content processed by the neural network to the neural network. The data processed by the neural network is output from the transmission unit to the line according to an instruction from the operator, and is printed by the recording unit.

[0008] In the invention according to claim 2, the reading means is RG.
The color image read in the B color system is input by the neural network, color-converted and corrected in the CMY color system, and output to the recording unit.

[0009] To reduce the color difference between the read data and the print data by performing color conversion and color correction of the data read by the RGB color system by the reading means to the CMY color system by a neural network, and printing out by a recording unit. Can be.

[0010] In the invention according to claim 3, the receiving unit is L ^* a.
^The neural network inputs a color image received in the ^* b ^* color system, performs color conversion and color correction in the CMY color system, and outputs the color image to the recording unit.

Since the data input from the receiving unit in the L ^* a ^* b ^* color system is converted to the CMY color system by the neural network, color-corrected, and output to the recording unit, the color difference between the input data and the output data is determined. Can be reduced.

In the invention according to claim 4, the reading means is RG.
The neural network inputs a color image read in the B color system and outputs it to the transmitting unit in the L ^* a ^* b ^* color system.

The reading means converts the data read in the RGB color system into the L ^* a ^* b ^* color system by a neural network, corrects the color, and outputs the data from the transmission unit. Can be reduced.

According to the fifth aspect of the present invention, the contents of the image processing are from the RGB color system to the CMY color system, from the L ^* a ^* b ^* color system to the CMY color system, and from the RGB color system to the L ^* color system ^. a ^* b ^*
This is at least one of color conversion to a color system and color correction processing.

In the neural network, by switching weights, the RGB color system is changed to the CMY color system, the L ^* a ^* b ^* color system is changed to the CMY color system, or R is changed.
Conversion from the GB color system to the L ^* a ^* b ^* color system is possible.

According to a sixth aspect of the present invention, when the data inputted to the neural network is of an RGB color system,
R, G, B and higher order RG, RB, GB, R ² ,
G ² , B ² , RGB, and L ^* a ^* b ^* color system,
L ^* , a ^* , b ^* and the higher-order L ^* a ^* , L
^* b ^* , a ^* b ^* , L ^{* 2} , a ^{* 2} , b ^{* 2} , L ^* a ^* b ^*
The neural network is configured according to the input data.

The input to the neural network is, for example,
For example, in the RGB color system, not only primary R, G, and B, but also
, RB, GB, R of higher order data of^Two, G^Two, B^Two,
Neural network that can input this data as RGB
Work configuration. L^*a ^*b^*The same applies to the color system
You. This improves the accuracy of conversion and color correction. In addition,
For high-order data, up to the above third order is sufficient for practical use
Accuracy is obtained.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram illustrating a color image processing apparatus according to an embodiment. The reading unit 1 is a scanner that reads an original image by separating the original image into R, G, and B in an RGB color system. The communication unit 2 exchanges data between the color image processing apparatus and another apparatus. The neural network 3 converts the data of the RGB color system read by the reading unit 1 to the CMY color system at the time of copying, and L ^* a ^* b at the time of transmission.
^* Color conversion and color correction to data of the color system and color conversion of the data of the L ^* a ^* b ^* color system received from the communication unit 2 to the CMY color system for recording and color correction. The recording unit 4 is a printer that prints data subjected to color conversion and color correction to the CMY color system by the neural network 3.

If the input image data is of the RGB color system, the data input operation unit 5 outputs the primary constituent color data;
These higher-order data, RG, GB, R
B, R ² , G ² , B ² , RGB, etc. are calculated, and primary and higher order data are used as input data to the neural network 3. The same applies when the input data is in the L ^* a ^* b ^* color system. The weight switching unit 6 converts the RGB color system to CM.
To the Y color system, from the RGB color system to the L ^* a ^* b ^* color system,
The weights obtained by learning are switched and output according to the conversion from the L ^* a ^* b ^* color system to the CMY color system and color correction.

A line a represents a case where the data of the RGB color system read by the reading unit 1 is subjected to color conversion and color correction to the CMY color system by the neural network 3 and output from the recording unit 4. A line b represents L ^* a ^* b data of the RGB color system read by the reading unit 1 by the neural network 3.
^* Shows the case where color conversion and color correction are performed for the color system, and transmission is performed from the communication unit 2. A line c indicates a case where the data of the L ^* a ^* b ^* color system received by the communication unit 2 is color-converted into the CMY color system by the neural network 3, color-corrected, and recorded by the recording unit 4.

FIG. 2 is a diagram showing the configuration of the neural network. The output circuit is composed of an input layer, an intermediate layer, and an output layer. The output is compared with a teacher signal by a comparison circuit, and the weights of the output layer and the intermediate layer are converted according to the error. The number of input layers is determined according to the input data because primary and higher-order data of the RGB color system and the L ^* a ^* b ^* color system are also input. The output layer is C, M, Y, or L. ^* , A ^* , and b ^* .

Next, learning of the neural network will be described. First, in the learning of the first weight, a color chart (standard sample of color) is read from the reading unit 1 (scanner), and RGB values on the display are used as input values of the neural network, and CMY values of the color chart are used as teacher data.
A color facsimile, a color shift between a scanner input and a printer output which may occur in a color copier, is corrected by a neural network having this weight. This allows RGB
Color conversion and color correction (color shift correction) from the color system to the CMY color system are performed. In the next weight learning, RGB on the display of the color chart read from the scanner 1 is used as an input value of the neural network, and L ^* a ^* b ^{* of the} color chart is used ^.
The value is used as teacher data. This corresponds to the case where a document is read and transmitted. In the third weight learning, L ^* a ^* b ^* obtained by measuring the color of a printed color chart with a spectrophotometer is used as an input value of a neural network, and the CMY values of the color chart are used as teacher data. This corresponds to the case where the received data is printed by the recording unit 4.

The input data to the input layer of the neural network will be described using an RGB color system as an example.
It is better to use higher-order data than to use only primary data such as R, G, and B. As secondary data, R
G, GB, BR, R ² , G ² , and B ^{2. If} a total of 10 input data of RGB is used as the third order, color conversion and color correction can be performed with accuracy that does not cause any practical problem. Even if higher-order data is added, the accuracy does not improve even if the calculation becomes complicated. Therefore, the number of input data to the input layer of the neural network is set to 10, and a neural network corresponding to this is configured. Arithmetic operation of seven higher-order input data than three of R, G, and B is performed by the data input operation unit 5. The use of up to higher order data is L
^The same applies to the input of the ^* a ^* b ^* color system.

[0024]

As is apparent from the above description, the present invention can perform color conversion and color correction of a plurality of color systems by switching the weight of the neural network. Therefore, a device with a small amount of hardware can be obtained. If the input to the neural network is an RGB color system as an example, not only the primary data of R, G, and B but also RG, GB, BR, R ² , G ² , B ² ,
By using higher-order data such as RGB, accurate color conversion and color correction can be performed.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a color image processing apparatus according to an embodiment of the present invention.

FIG. 2 is a configuration diagram of a neural network.

[Explanation of symbols]

 1 reading unit 2 communication unit 3 neural network 4 recording unit 5 data input calculation unit 6 weight switching unit

Claims (6)

[Claims] A reading unit that reads a color image; a receiving unit that receives a color image; and an image processing unit that performs color conversion and color correction of the color image received by the receiving unit and the color image read by the reading unit. A neural network to be performed, input conversion means for converting a color image input to the neural network into data according to the configuration of the neural network, weight switching means for switching a weight of the neural network according to the content of image processing, A color image processing apparatus, comprising: a transmission unit that transmits an output of a neural network; and a recording unit that records an output of the neural network. 2. The neural network inputs a color image read by the reading means in an RGB color system,
2. The color image processing apparatus according to claim 1, wherein color conversion and color correction are performed for a CMY color system and output to the recording unit. 3. The color image received by the receiving unit in the L ^* a ^* b ^* color system is input by the neural network, color-converted and corrected in the CMY color system, and output to the recording unit. The color image processing apparatus according to claim 1, wherein: 4. The neural network inputs a color image read by the reading means in an RGB color system,
2. The color image processing apparatus according to claim 1, wherein the image data is output to the transmission unit in an L ^* a ^* b ^* color system. 5. The contents of the image processing are from an RGB color system to a CMY color system, from an L ^* a ^* b ^* color system to a CMY color system, and from an RGB color system to an L ^* a ^* b ^* table. 2. The color image processing apparatus according to claim 1, wherein at least one of color conversion processing and color correction processing for a color system is performed. 6. In the case of an RGB color system, data input to the neural network is R, G, B and higher-order RG, RB, GB, R ² , G ² , B ² , RGB.
In the case of the L ^* a ^* b ^* color system, L ^* , a ^* , b ^* and higher-order L ^* a ^* , L ^* b ^* , a ^* b ^* ,
^2. The color image processing apparatus according to claim 1, wherein L ^{* 2} , a ^{* 2} , b ^{* 2} , and L ^* a ^* b ^*, and the neural network is configured according to the input data.