JPH05347703A - Method and device for nonlinear conversion of color information in even color space - Google Patents

Abstract

PURPOSE:To transfer color information between different color regenerating systems in a more proper way by making use the nonlinear conversion ability of a neural network. CONSTITUTION:A neural network used to a converter means contains the intermediate layers 1-3 including the units Ut11-Ut32 of n1, n2 and n3 respectively and also contains an output layer consisting of the units Ur1-Ur3. The number of output layers is changed by the coordinate value obtained in an even color space in response to the primary color elements like CMY or CMYK of a subtraction mixed color, a primary color element like RGB of an addition mixed color, or the color information value like three attributes of the color perception such as the hue, saturation and luminosity. Meanwhile the bias signals are applied to each of units of both intermediate and output layers from the bias units Ubt1-Ubt3 and Ubr. In such a constitution, the nonlinear conversion ability of the neural network can be effectively used.

Description

Detailed Description of the Invention

[0001]

The present invention relates to cyan (C), magenta (M), yellow (Y) or cyan (C), magenta (M), yellow (Y) and black (K) in subtractive color mixing. Such as primary color elements, primary color elements such as red (R), green (G) and blue (B) in additive color mixture, or hue (Hue, H), saturation (Croma, C), lightness (Value, V). ), Such as three attributes of color perception, and a method and apparatus for converting a combination of at least three information values related to color to corresponding coordinate values in the uniform color space.

[0002]

2. Description of the Related Art When artificially expressing a color, looking at a minute area expressing one color, a method of expressing the color by uniformly giving the same color to the entire minute area, To give individual colors to multiple sub-areas, which are completely independent of each other or overlap each other partially in a small area, usually 2 to 8, and these colors are mixed in the human visual sense. There is a method of recognizing the color according to. As the former example, that is, an example having a color that is also microscopically uniform, there is gravure printing by CMY in the subtractive color mixture and spotlight by RGB in the additive color mixture. In the latter example, that is, in the case where a plurality of colors are juxtaposed microscopically, there is process printing by CMY in the subtractive color mixture, and R in the additive color mixture.
There is a color display by GB.

With the development of computer technology, color images can be converted into various color output devices (eg, color copy,
Systems that reproduce on color displays, printing machines, etc.) are becoming very important. However, there are differences in the colors reproduced by each output device due to the differences in the physical and chemical characteristics of each output device, so that the colors reproduced by each output device must be accurately compared and corrected. Technology is required.

Conventionally, in order to solve these problems, in general,
A method of correcting a difference in reproduced color due to a difference in color reproduction system by using matrix conversion or a linear equation is adopted.

[0005]

However, in the conventional method using the matrix conversion and the linear equation, although the change of the color information is non-linear, the color that comes in the middle of the known data is not changed. Since the information processing is performed by linear interpolation, the correction result may not match the actual color perception. Also, when color information is converted using a linear compression method in a mathematical space, uniform conversion is performed over the entire color gamut, which is sufficient to deal with the non-linearity of color information. However, the same problem occurs as in the matrix conversion method.

In addition, the above-mentioned CMY, CMYK or RG
B is a color mixture system, not a color system created based on the human sense, so even if the same amount of correction is applied to different colors, the human sense does not perceive the same change. There was a problem, and the linear masking method allowed only limited color adjustment.

The present invention enables a non-linear conversion between color information and a uniform color space, which cannot be realized by the linear conversion, thereby making it possible to obtain a color which is well suited to human senses and which is difficult in the conventional method. It is an object of the present invention to provide a method and apparatus for non-linear conversion of color information, which makes it possible to process information and thereby accurately transfer color information between different color reproduction systems.

[0008]

According to the present invention, there is provided a method for converting a combination of at least three information values related to color from a coordinate value on a uniform color space, and the combination of the information is This is achieved by a non-linear conversion method of color information, characterized in that the conversion is performed non-linearly by a neural network trained to convert from coordinate values on a corresponding uniform color space.

In this case, it is preferable that the neural network is, in particular, a feedforward connection type neural network of another layer. Further, the three information values may be the three primary colors of the subtractive color mixture forming the color image, or the three primary colors of the additive color mixture, or the hue, the saturation and the lightness. . The subtractive color mixture or the additive color mixture may be given by a method in which each color is uniformly mixed in a microscopic manner, or in a microscopic manner, a plurality of colors are completely independent of each other or partially in each other. It may be provided by a method of coloring overlapping sub-area regions.

Further, according to the present invention, the above-mentioned problem is an apparatus for converting a combination of at least three values relating to color from coordinate values on a uniform color space, and a uniform color space corresponding to the three information values. Input means for converting the upper coordinate value into an electric signal corresponding thereto, conversion means for converting an electric signal from the input means into an electric signal corresponding to the three values, and an electric signal after conversion. An output layer for outputting an arbitrary output signal corresponding to the three values, wherein the conversion means has an input layer having at least three units;
A feedforward coupled neural network of 3 to 5 layers having an intermediate layer having 1 to 3 layers having at least 1 unit, and an output layer having at least 3 units, and the above-mentioned Both are achieved by a non-linear conversion device of color information, which is learned to convert a combination of three information values from corresponding coordinate values on a uniform color space.

[0011]

The concept of uniform color space itself is known as CIE uniform color space, and it includes L * a * b *, L * u *.
There is a uniform color space coordinate system such as v *. However, the concept of such a uniform color space is merely set mathematically, and primary color elements such as CMY or CMYK in subtractive color mixing, primary color elements such as RGB in additive color mixing, or hue and color. How to perform universal conversion between a combination of specific information values related to actual colors such as three attributes of color perception such as degree and lightness and coordinate values in the uniform color space Is another problem that has yet to be resolved.

According to the color information non-linear conversion method or apparatus of the present invention, by effectively utilizing the non-linear conversion function of the neural network, even when the color image is reproduced by subtractive color mixing, or again, the color image is reproduced. When is reproduced by additive color mixing, and when three attributes such as hue, saturation, and lightness are handled, the values of information about these colors are converted into coordinate values in the uniform color space and vice versa. In addition, it is possible to perform conversion while well retaining the non-linear characteristic of color information, which makes it possible to appropriately transfer color information between different color reproduction systems via a uniform color space coordinate system. Becomes Furthermore, in this case, with respect to the neural network, not only the non-linear conversion function but also the learning function can be effectively used, so that at least one, and preferably some color information For a combination, by letting a neural network learn its transformation, a wide range of color information can be obtained.
Appropriate non-linear conversion can be performed.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The preferred embodiments of the present invention will be described below with reference to the accompanying drawings.

FIG. 1 shows the primary color elements such as CMY or CMYK in the subtractive color mixture, the primary color elements such as RGB in the additive color mixture, or the three attributes of color perception such as hue, saturation and lightness according to the present invention. It is a schematic diagram showing a basic configuration of a color information conversion method or apparatus for converting a combination of color information values from coordinate values in a uniform color space. As shown in the figure, this conversion method or conversion device has an input means 1, a conversion means 2, an output means 3 and a teacher means 4. The conversion means 2 has not only a conversion function described below, but also a learning function for learning it.

The input means 1 is L * a * b *, L * u * v.
Coordinates in uniform color space such as *, each value of primary color element such as CMY or CMYK in subtractive color mixture, each value of primary color element such as RGB in additive color mixture, or hue, saturation , The value of each of the three attributes of color perception, such as lightness, to obtain each value of the combination of information values relating to the color, of the value to be extracted by the transformation from the corresponding coordinate value in the uniform color space. The coordinate values on the uniform color space are generated as electrical signals. A commercially available spectrophotometer may be used for this.

As shown in FIG. 2, the neural network used in the converting means 2 includes units Us1, Us2 ,. ． ． It is a multi-layer feed-forward coupled neural network in which a large number of etc. are combined. The illustrated neural network has an input layer consisting of three units Us1, Us2, Us3, which number is a primary color element such as CMY or CMYK in subtractive color mixing,
Primary color elements such as RGB in additive color mixture, or hue,
Corresponds to the number of data of coordinate values in the uniform color space such as L * a * b *, L * u * v *, etc. for a combination of information values related to color such as three attributes of color perception such as saturation and lightness. .. The illustrated neural network further includes n1 and n2, respectively.
, N3 units Ut11, Ut12 ,. ． , Ut21,
Ut22 ,. ． , Ut31, Ut32 and one to three intermediate layers, and further three units Ur1, Ur2, U
It has an output layer consisting of r3. The number of output layers is converted from the coordinate values in the uniform color space, and CM in subtractive color mixing
Primary color elements such as Y or CMYK, R in additive color mixture
It corresponds to the number of primary color elements such as GB or the number of color information values such as three attributes of color perception such as hue, saturation and lightness. Bias signals are applied from the bias units Ubt1, Ubt2, Ubt3 and Ubr to each unit of the intermediate layer and each unit of the output layer.

When the coordinate values on the uniform color space as described above are given to the input layer as an input pattern, the neural network performs the following forward processing of proceeding from the input layer to the intermediate layer to the output layer, and then to the output layer. Corresponding primary color elements such as CMY or CMYK in subtractive color mixing, primary color elements such as RGB in additive color mixing, or hue, saturation,
An electric signal of a color information value such as three attributes of color perception such as lightness is output as an output pattern.

The output value of the input layer unit Usi is Ii (i = 1, 2, 3) and the output value of the intermediate layer unit Utjk is 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 the unit Usi of the input layer to the unit Ut1 of the intermediate layer
The coupling coefficient to k is Wst1, the coupling coefficient between units in the hidden layer is Wttjk, the coupling coefficient from the unit Ut3k in the hidden layer to the unit Urm in the output layer is Wt3r, and the bias to the unit Utjk in the hidden layer is θtj, and the output is Layer unit U
Assuming that the bias for rm is θr, H1k = f (SUM Ii · Wst1 + θt1, i = 1 to 3) (1) H2k = f (SUM H1k · Wtt12 + θt2, k = 1 to n1) (2) H3k = f (SUM H2k · Wtt23 + θt3, k = 1 to n2) (3) Om = f (SUM H3k · Wt3r + θr, k = 1 to n3) (4) Here, the output of the function f is within the range of (0, 1). Is a monotonically increasing non-decreasing sigmoid function, 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, Wtt12, Wtt23, Wt3.
It is defined by transformation coefficients such as r and bias values θt1, θt2, θt3, θr, which are formed by learning described below with reference to FIG.

FIG. 3 shows a procedure for training the neural network shown in FIG. As a learning method, there is back propagation learning. This is a desired output after conversion, which is a primary color element such as CMY or CMYK in the subtractive color mixture, a primary color element such as RGB in the additive color mixture, or three attributes of color perception such as hue, saturation and lightness. A coordinate value in a uniform color space for a sample color whose value is known is converted into an electric signal and is given to an input layer as an input pattern, and the output layer is provided with CMY or a subtractive color mixture of the sample color. Primary color elements such as CMYK, primary color elements such as RGB in additive color mixture, or hue, saturation,
A value obtained by converting the values of the three attributes of color perception such as lightness into an electric signal is given as a teacher signal pattern, and the coupling coefficient and the bias value are set from the output layer so as to reduce the difference between the output signal pattern and the teacher signal pattern. It is a method of correcting in the opposite direction from the intermediate layer to the input layer.

In the neural network in the state where learning has been completed for some sample colors as described above, the non-linearity in its conversion function is CMY in subtractive color mixing.
Or primary color elements such as CMYK, RG in additive color mixture
By being adapted to the non-linearity between the primary color elements such as B, or the three attributes of color perception such as hue, saturation, and lightness and the corresponding coordinate values in the uniform color space It was confirmed by the experiment that, for the colors other than the colors described above, the conversion values that are well fitted to the input values are obtained.

FIG. 4 shows an example in which the color information in the L * a * b * uniform color space coordinate system is converted into the color information in the CMY coordinate system by using one intermediate layer unit of 10 intermediate layers. In the center of the figure, the neural network of FIG. 2 is j = 1, k =
10 and the input layer of the neural network as an input, Japanese Patent Application No. 3-4302
L * a * b as shown in the upper part of the figure obtained by a neural network using 10 intermediate layers in the same manner as in Japanese Patent Application No. 3- with domestic priority claim based on No. 9.
* Indicates a color information conversion device that inputs a coordinate space of a uniform color space coordinate system and outputs a coordinate space of a corresponding CMY coordinate system from an output layer. With such a color information conversion device, the coordinate space of the L * a * b * uniform color space coordinate system shown in the upper part of the figure is converted into the CMY rectangular coordinate space as shown in the lower part of the figure.

FIG. 5 shows the progress of conversion error with respect to the number of times of learning, that is, the learning speed, when the color information conversion apparatus of FIG. 4 is learned using eight types of color information in the same manner as the learning procedure shown in FIG. It is a graph which shows. As you can see from the figure, 1 stage ×
A neural network with 10 intermediate layers completes the learning of the above color information with a relatively small number of learnings, and due to its generalization ability, also performs good conversion of color information other than the 8 types of color information used as teaching materials. It was confirmed to do.

[0024]

As described above, according to the present invention, by using a neural network, particularly preferably a multi-layered feed-forward connection type neural network, primary color elements such as CMY or CMYK in subtractive color mixture and RGB in additive color mixture are used. By converting the values of the primary color elements or the three attributes of color perception such as hue, saturation, and lightness from the uniform color space coordinates, the nonlinear conversion ability of the neural network, especially the nonlinear conversion ability based on its learning function. By using the primary color elements such as CMY or CMYK in the subtractive color mixture, the primary color elements such as RGB in the additive color mixture, or the non-linearity such as the three attributes of color perception such as hue, saturation and lightness. Color information is converted from linear coordinate values, and color information for more appropriate transfer of color information between different color reproduction systems is created. Management is possible.

[Brief description of drawings]

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

FIG. 2 is a schematic diagram showing one embodiment of a multilayer feedforward coupling type neural network used in the converting means in the configuration of FIG.

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

FIG. 4 is a diagram showing a multi-layer feedforward connection type neural network having 10 intermediate layers in one stage, and L * a * b * uniform color space coordinates as shown in the upper part of the figure as shown in the lower part of the figure. FIG. 6 is a schematic diagram showing an example of conversion into a corresponding coordinate space of a CMY coordinate system.

5 is a graph showing the learning speed of the neural network of FIG.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Input means 2 ... Conversion means 3 ... Output means 4 ... Teacher means

Front page continuation (51) Int.Cl. ⁵ Identification code Office reference number FI Technical display location G06F 15/66 310 8420-5L G06G 7/60 H04N 1/46 9068-5C

Claims (6)

[Claims] 1. A method for converting a combination of at least three information values relating to color from coordinate values on a uniform color space, and learning to convert a combination of the information sets from corresponding coordinate values on a uniform color space. A non-linear conversion method of color information, wherein the conversion is performed non-linearly by the neural network. 2. The non-linear conversion method according to claim 1, wherein the neural network is a feedforward coupled neural network of another layer. 3. The non-linear conversion method according to claim 1 or 2, wherein the three information values are three primary colors of subtractive color mixture forming a color image. 4. The non-linear conversion method according to claim 1 or 2, wherein the three information values are three primary colors of additive color mixture forming a color image. 5. The non-linear conversion method according to claim 1 or 2, wherein the three information values are hue, saturation and lightness. 6. An apparatus for converting a combination of at least three information values relating to color from coordinate values on a uniform color space, and converting the coordinate values on a uniform color space corresponding to the three information values to corresponding electrical signals. Input means for converting into an electric signal from the input means, conversion means for converting an electric signal from the input means into an electric signal corresponding to the three information values, and an electric signal after conversion corresponding to the three information values Output means for outputting as an output signal of, the converting means includes an input layer having at least three units, an intermediate layer having 1 to 3 layers having at least one unit, and at least three. An output layer having a unit, and a feedforward connected neural network of 3 to 5 layers having an output layer having a unit, and a combination of the at least three information values relating to color corresponding to a uniform color space. Non-linear conversion device color information, characterized in that it is allowed to learn to convert from the above coordinate values.