JPH04261267A - Method and device for nonlinear conversion of color information by uniform color space - Google Patents

Abstract

PURPOSE:To attain proper conversion of color information with high accuracy between different devices as to color information of a wide range by utilizing nonlinear conversion capability of a neural network and its learning capability. CONSTITUTION:This device is provided with an input means 1, a conversion means 2, an output means 3 and a teacher means 4, and the conversion means 2 is formed to be a neural network having a conversion function and a learning function of the conversion. The input means 1 implements spectral analysis of an original color by which color information is to be converted and generates a coordinate on a uniform color space such as L*a*b*L*u*v corresponding to the color. When the coordinate on the uniform color space such as L*a*b is given to an input layer as an input pattern, and the neural network implements forward processing from the input layer to an output layer via an intermediate layer and outputs the coordinate on the uniform color space such as L*a*b to be converted to the output layer as an output pattern.

Description

[Detailed description of the invention]

0001

[Field of Industrial Application] The present invention relates to a method and apparatus for transferring color information between devices that perform color reproduction (CRT, color copier, color printer, etc.), and more specifically, to a method and apparatus for transferring color information between devices that reproduce colors. The present invention relates to a method and apparatus for nonlinear conversion of color information between color reproduction devices, which corrects or corrects discrepancies in reproduced colors due to differences in reproduction characteristics, and enables consistent color reproduction independent of devices.

0002

[Background Art] With the recent development of printing technology and computers, systems that read images, characters, etc. with a scanner and edit or process them on a computer display have become extremely important. There is a need for technology to accurately predict and correct color on the display. However, the colors reproduced on a display and the colors reproduced by printing or a color output device (color copy, etc.) using other color reproduction methods differ due to differences in the color reproduction method of each device and the color mixture system. They differ from each other due to differences in color reproduction characteristics, including differences in physical, chemical, or engineering factors, as well as differences in color reproduction effects and color reproduction ranges obtained by these factors. , attempts have been made to predict accurate colors.

[0003] Conventionally, in order to solve such problems, a lookup table is generally created to correct differences in color reproduction depending on the device, and the lookup table is referenced to correct the difference in color reproduction depending on the device. For points not given in the up-table, linear interpolation is used to compensate for them. In addition, the problem of correcting the color reproduction range and color differences in color space is treated as a geometric transformation, similar to rotating or moving a figure, and the problem of correcting color information by applying these conversion methods is a geometric transformation. Scientific linear transformation methods are also used.

0004

[Problem to be Solved by the Invention] As mentioned above, various attempts have been made to correct the differences in color reproduction between devices and predict accurate colors, but the processing operations are complicated. , and its effects are still insufficient. In other words, the table lookup method requires a huge amount of data, and even though the color space itself is a nonlinear space, linear interpolation is used to interpolate the data as described above, so the converted result is Colors may not match actual colors. Also, when using the above geometric linear transformation method for color space,
Similarly, since uniform conversion is performed across the entire color space, which is a non-linear space, the colors reproduced as a result of conversion may change unnaturally, but color information with high reproducibility is transferred between different devices. It was insufficient to do so.

SUMMARY OF THE INVENTION An object of the present invention is to provide a method and apparatus for converting color information that can transfer color information between different devices with a high degree of reproducibility, which has been difficult with conventional techniques. .

[0006]

[Means for Solving the Problems] According to the present invention, the above problem is solved by converting color information in one color reproduction device or color reproduction method to another color reproduction device which differs in color reproduction characteristics. A non-linear conversion method of color information that converts color information into color information for a device or color reproduction method, in which the former color information is expressed as a coordinate value on a uniform color space, and the coordinate value is converted on a uniform color space. The coordinate values of the latter are used as the coordinate values of the latter, and in this case, the above conversion is carried out into uniform colors so that the color reproduced by the one color reproduction device or color reproduction method is equivalent to the color reproduced by the other one color reproduction device or color reproduction method. This is achieved by a nonlinear conversion method for color information, which is characterized by using a neural network trained to convert spatial coordinate values.

In the above nonlinear conversion method of color information,
The neural network may be a multilayer feedforward neural network.

According to the present invention, the above-mentioned problem can be solved by transferring the color information in one color reproduction device or color reproduction method to another color reproduction device or color reproduction method which has different color reproduction characteristics. A non-linear conversion device for color information that converts the former color information into color information for , an input means for converting the former color information into coordinate values on a uniform color space, and a color information from the input means that converts the color information from the input means into coordinate values on a uniform color space. The converting means includes a converting means for converting and an output means for outputting the coordinate values after the conversion, and the converting means has an input layer having at least three units and one to three layers having at least one unit. middle class and at least 3
a 3-5 layer neural network having an output layer having units of This can also be achieved by a color information non-linear conversion device that is trained to convert coordinate values on a uniform color space so that they become equivalent.

[0009]

[Operation of the invention] As mentioned above, physical and
A single color reproduction device or color reproduction method that has different color reproduction characteristics, such as differing from each other in chemical or engineering factors and the color reproduction effects obtained by them, or differing from each other in the color reproduction range. Converting color information into color information for color reproduction by another color reproduction device or color reproduction method is performed on a uniform color space, and at this time, the coordinate values of the former are converted into color information for color reproduction by another color reproduction device or color reproduction method. To convert to coordinate values,
By using a neural network that has learned to convert the coordinate values of the one color reproduction device or color reproduction method into the coordinate values of the other one color reproduction device or color reproduction method, the neural network Appropriately trained to perform coordinate value conversion to make the color produced by the one color reproduction device or color reproduction method equivalent to the color produced by the other color reproduction device or color reproduction method for the color. When the learning example is exceeded, the color information in the one color reproduction device or color reproduction method can be non-linearly equivalently converted into the color information in the other color reproduction device or color reproduction method. can.

The color information conversion effect by the nonlinear conversion method or device of the present invention is as follows.

First, the input means converts the color information before conversion into coordinate values on a uniform color space such as L*a*b*, L*u*v*, etc. These coordinate values may of course be electrical signals that can be processed by a computer. Next, this coordinate value signal is converted into a color reproduction device that reproduces the original color of the input signal or a color reproduction method different from the color reproduction method using a conversion means consisting of a neural network such as a multilayer feedforward neural network. Due to the reproduction device or color reproduction method, it is nonlinearly converted into a coordinate value signal. The conversion function here is set in advance using the learning function of the neural network. In the above learning, in the input layer of the neural network in which neural circuit elements (units) are connected in layers, the coordinate values in the uniform color space after conversion are known, and the coordinate values in the uniform color space before conversion are The coordinate values are input, and the coordinate values on the uniform color space after conversion of the color are given as teacher signals to the output layer of the neural network, and the neural network is made to learn. A neural network that has completed such learning for some colors may be used as a conversion means as is.

The uniform color space coordinate values thus converted by the neural network are converted by the output means into electrical or optical signals suitable for processing the converted color information.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention will now be described with reference to the accompanying drawings.

FIG. 1 is a schematic diagram showing the basic configuration of a nonlinear conversion method or device for converting color information in one device into color information for another device according to the present invention. . As shown in the figure, this conversion method or conversion device includes an input means 1, a conversion means 2, an output means 3, and a teacher means 4. The conversion means 2 is a neural network that has a conversion function described below and also has a learning function for learning the conversion function.

The input means 1 performs spectral analysis of the original color for which color information is to be converted, and calculates L*a*b corresponding to the color.
*, L*, u* v*, etc., on a uniform color space is generated, and this conversion is proposed in the patent application No. A method or apparatus may be used. The uniform color space coordinate values determined by the input means 1 are introduced into the conversion means 2.

The neural network used in the conversion means 2 includes units Us1, Us2, . . . consisting of neural circuit elements as shown in FIG. ．． ．． It is preferable that the neural network is a multilayer feedforward combination type neural network in which a number of layers are combined. The neural network shown is
It has an input layer consisting of three units Us1, Us2, Us3, the number of which is L* a* b*, L* u* v
* Corresponds to the number of data of input coordinate values on the uniform color space. The illustrated neural network further includes:
Each of n1, n2, n3 units Ut11
, Ut12 ,. ．． , Ut21 , Ut22 , . ．． ,
It has 1 to 3 intermediate layers including Ut31 and Ut32, and further includes three units Ur1, Ur2, Ur
It has an output layer consisting of three. The number of output layers corresponds to the number of coordinate values on a uniform color space such as L* a* b* after conversion. In addition, bias units Ubt1, Ubt2, and Ubt are provided in each layer of the intermediate layer and each unit of the output layer.
3. A bias signal is given from Ubr.

When coordinate values on a uniform color space such as L* a* b* are given to the input layer as an input pattern, the neural network performs the following forward processing that proceeds from the input layer to the output layer via the intermediate layer. Then, the coordinate values on the uniform color space such as L* a* b* converted to the output layer are output as an output pattern.

The output value of the input layer unit Usi is Ii (
i=1,2,3) The output value of the intermediate layer unit Utjk is expressed as Hjk(j=1,2,3 k=1,2,...
n) The force value of the output layer unit Urm is Om (m=1, 2, 3), and from the input layer unit Usi to the intermediate layer unit U
The coupling coefficient to t1k is Wst1, the coupling coefficient between units in the middle layer is Wttjk, and the unit in the middle layer is Ut3k.
The coupling coefficient from to the output layer unit Urm is Wt3r
, and if the bias for the unit Utjk in the intermediate layer is θtj and the bias for the unit Urm in the output layer is θr, then H1k=f(SUM Ii
・Wst1 +θt1, i=1~3) (1)
H2k=f(SUM H1k・Wtt12+θt2
, k=1~n1) (2) H3k=f(SU
M H2k・Wtt23+θt3, k=1~n2
) (3) Om = f(SUM H3k・Wt
3r + θr, k=1 to n3) (4) Here, the function f is a non-decreasing sigmoid function whose output is monotonically increasing within the range (0, 1), and is generally defined by the following equation.

f(x)=1/{1+exp(-x)}
(5) The values of the output layer are the coupling coefficients Wst1, Wtt1
2. Wtt23, Wt3r and bias values θt1, θ
It is defined by conversion coefficients such as t2, θt3, and θr, and these values are formed by learning described below with reference to FIG.

FIG. 3 shows how the neural network shown in FIG. 2 is trained. One such learning method is backpropagation learning. This is done by converting the coordinate values on the uniform color space before conversion of a sample color whose coordinate values on the uniform color space are known as the desired output after conversion into electrical signals, and inputting it to the input layer as an input pattern. The coordinate values on the uniform color space after conversion of the sample color are converted into electrical signals, and the output layer is given as a teacher signal pattern to reduce the difference between the output signal pattern and the teacher signal pattern. This is a method of modifying the coupling coefficient and bias value in the opposite direction from the output layer to the intermediate layer and from the intermediate layer to the input layer.

[0021] After the neural network has completed learning about several sample colors as described above, the nonlinearity in its conversion function is due to the nonlinearity in the difference in color reproduction characteristics between the devices concerned. It has been confirmed through experiments that this adaptation provides converted values that are well-adapted to the input values, even for colors other than the colors targeted for learning.

Three units Ur1, which constitute the output layer,
The output signals from Ur2 and Ur3 are introduced into the output means 3 and converted into electrical or optical signals suitable for processing the converted color information. For this conversion, the method or apparatus proposed in Japanese Patent Application No. 5, filed concurrently by the same applicant as the present applicant, may be used.

[0023]

[Effect] Thus, according to the present invention, in the conversion of color information,
By using uniform color space coordinates and coordinate value transformation by a neural network therein, preferably a multilayer feedforward combination neural network, and by utilizing the nonlinear transformation ability of the neural network and its learning ability, conventional Compared to the look-up table method, it is possible to perform appropriate conversion of color information between different devices with high precision for a wide range of color information without relying on linear interpolation, which was insufficient in the look-up table method. Become.

[0024] Furthermore, L*a*b*, L*u*v*, which is characterized in that the rate of color change is at constant intervals, is useful for converting color information between devices with different color reproduction characteristics.
By using a homogeneous color space such as , the color values reproduced by the method and apparatus according to the present invention are well-suited to human senses and have no unnatural appearance.

[Brief explanation of the drawing]

FIG. 1 is a schematic diagram showing the basic configuration of a nonlinear conversion method or nonlinear conversion device for color information according to the present invention.

FIG. 2 is a schematic diagram showing one embodiment of a multilayer feedforward combination type neural network used in the conversion means in the configuration of FIG. 1;

FIG. 3 is a schematic diagram showing a learning method for learning the neural network of FIG. 2;

[Explanation of symbols]

1...Input means 2... Conversion means 3...Output means 4...Teacher means

Claims (3)

[Claims] Claim 1: Color information for converting color information in one color reproduction device or color reproduction method into color information for another color reproduction device or color reproduction method different in color reproduction characteristics. Using the non-linear transformation method, the former color information is expressed as a coordinate value on a uniform color space, and the coordinate value is transformed into the latter coordinate value on a uniform color space. A neural network trained to transform coordinate values on a uniform color space so that the color reproduced by the color reproduction device or color reproduction method is equivalent to the color reproduced by the other color reproduction device or color reproduction method. A non-linear conversion method of color information characterized by performing the following steps. 2. A nonlinear conversion method for color information according to claim 1, wherein said neural network is a multilayer feedforward neural network. 3. Color information for converting color information in one color reproduction device or color reproduction method into color information for another color reproduction device or color reproduction method different in color reproduction characteristics. a nonlinear conversion device, including an input means for converting the former color information into coordinate values on a uniform color space, a conversion means for converting the color information from the input means on a uniform color space, and a coordinate value after the conversion. an input layer having at least three units; an intermediate layer having one to three layers having at least one unit; and an output layer having three to five layers, wherein the color reproduced by the one color reproduction device or color reproduction method is equivalent to the color reproduced by the other one color reproduction device or color reproduction method. 1. A nonlinear conversion device for color information, characterized in that the device is trained to convert coordinate values on a uniform color space so that the color information becomes .