JPH06124343A - Limited color determination method and limited color determination device - Google Patents

Abstract

(57) [Summary] [Purpose] When displaying image data in a small number of limited representative colors, an image that does not cause discomfort is created by a simple process. A color conversion means 5 converts color image data into a color distribution in a uniform color space. The division threshold determination means 14 determines a threshold for this color distribution using discriminant analysis. Then, at this threshold value, the color area dividing unit 16 divides the uniform color space into color areas, and the average color of this color area is used as the representative color. Then, when obtaining the frequency distribution for each coordinate axis of each pixel in the uniform color space, the class number of the coordinate axis indicating the luminance is made larger than the class number indicating the chromaticity. Then, by performing a discriminant analysis on the obtained frequency distribution, a representative color in which the luminance is emphasized is selected in correspondence with the human vision sensitive to the luminance.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention, when displaying a digitized natural image on a display device with a small number of limited colors, creates an image that does not feel uncomfortable using a limited number of representative colors. The present invention relates to a limited color determination method and a limited color determination device for displaying.

[0002]

2. Description of the Related Art Generally, a color image used in a color display device is represented by RGB color components, and each color component is digitized by allocating an 8-bit memory. Therefore, in the full-color display that displays the digitized color image, two pixels are used for each pixel.
It is possible to display 4-bit colors, that is, 2 ²⁴ colors at the same time, and it is possible to display a natural image without a feeling of strangeness.

On the other hand, CD-ROM, CD-I, CAD
(Computer Aided Design), CAI (Computer Assist)
ed Instruction) and the world of graphics, 2
^In many cases, a color display device that can display only a limited number of ²⁴ colors at the same time is used. In such a case, since the capacity per pixel of the image memory is usually set to 8 bits or less, simultaneous display is possible. 2 ^{8 possible} colors
= Up to 256. It is difficult to display a natural image without deterioration with a limited number of colors such as 256 colors or less, but there are many demands for enabling this. Also,
From the relationship of the data capacity of the display device, etc., the number of colors of the natural image is reduced so that the natural image does not deteriorate,
There are many demands for reducing the amount of image data.

For such a purpose, conventionally, for a color map display for selecting and displaying a small number of display colors from a large number of colors, a natural image is displayed with a limited number of colors while suppressing visual deterioration. A method of displaying is known.

As the above-mentioned method, for example, the colors R, G,
When represented by the respective values of the three primary colors of B, the space represented by the three-dimensional orthogonal coordinate system having the respective values of R, G, and B as coordinates is defined as a color space, and in this color space, color image data is Analyze the color distribution of each included pixel and select a display color from the color area in the color space in which the colors included in the color image data are mostly distributed and select a small display color from the color area with a small distribution. , These display colors are registered in the color map, the display color closest to the color of each pixel of the color image data is assigned to each pixel as a representative color, and the color code registered as the representative color in the color map is stored in each of the image memories. There is one in which approximate display is performed by writing at a pixel position. As one of the methods for selecting the representative color as described above, Otsu's discriminant analysis (Science theory (D) .J63-D, 4, pp.349-356.
Showa 55-4. ) Is used.

Further, the color space is not limited to one that expresses a color by a combination of the three primary colors of RGB, but also the L ^* u ^* v ^* color space and the L ^* a ^* b ^* color space (in consideration of human color difference). CIE
There is a uniform color space (such as that recommended by the International Commission on Illumination), and the discriminant analysis method can be performed using these color spaces.

Further, a method in which the human visual characteristic is utilized and the dither method is further applied (Journal of the Institute of Image Electronics Engineers, Vol. 18, No. 5)
No. (1989) p, 293-301; Tajima, Ikeda), but this method is not considered for limited color display with a small number of colors of 100 or less.

[0008]

By the way, since the discriminant analysis method simply determines the display color based on the color distribution, an image created by determining the display color by the discriminant analysis method in the RGB color space is For the human eye, it was not a natural reproduction of a natural image. For example, false contours may occur when a small number of representative colors are assigned to a portion where the color change of the image data is gradual, and in particular, because human visual characteristics are sensitive to luminance, There is a problem that the false contour becomes conspicuous when the change in the luminance gradation is expressed by a small number of representative colors.

Further, even if the discriminant analysis method is used in the L ^* u ^* v ^* color space or the L ^* a ^* b ^* color space in which human color difference is taken into consideration, it is not always a natural image when viewed from the human eyes. It was not an approximation. Further, as a conventional method, like the above-described discriminant analysis method, a limited display color is selected to create an image despite the fact that the state of the original image, for example, the color distribution is statistically different. There is a non-adaptive method in which the limited color is selected and an adaptive method in which the limited color is selected so as to suit the statistical properties such as the color distribution of the original image.

In the non-adaptive method, the image is not always an image approximated to the original image when seen by human eyes as in the case of the above-mentioned discriminant analysis method, and there is a problem in image quality. In the adaptive method, the color distribution of each image data is checked,
It is necessary to change the statistical processing method according to the state of the color distribution, and generally the processing becomes complicated and the processing time is long. Then, a non-adaptive method capable of maintaining a certain level of image quality with respect to general image data has been developed as compared with the adaptive method due to the advantage that the image data is not particularly selected and the processing time is fast. Some natural images have very different statistical properties such as color distribution depending on their contents, and the advantage of adaptive processing has not changed from the viewpoint of image quality.

The present invention has been made in view of the above problems, and paying attention to the fact that human vision is particularly sensitive to a change in luminance, enables more faithful reproduction of an original image, and An object of the present invention is to provide a limited color determination method capable of solving the problems of the non-adaptive method and the adaptive method.

[0012]

The limited color determination method of the present invention determines a representative color by performing a discriminant analysis on the frequency distribution of each pixel value of color image data represented by a coordinate system of a uniform color space. Obtained, which is to approximate the image data with a limited representative color, in the uniform color space, the class number in the coordinate axis showing the luminance component and the class number in the coordinate axis showing the chromaticity component are different, each of the above, The means for solving the above problems is to obtain a frequency distribution of each pixel value data on each coordinate axis, and perform a discriminant analysis on the frequency distribution to obtain a representative color.

Further, in the limited color determination method of the present invention, when obtaining the frequency distribution in the uniform color space, the class of the coordinate axis showing the luminance component is made larger than the class of the coordinate axis showing the chromaticity component, The selection of a representative color in which brightness is emphasized is the means for solving the above problems.

Further, in the limited color determination method of the present invention, when obtaining the frequency distribution in the uniform color space, the number of classes of the coordinate axis indicating the chromaticity component is made larger than the number of classes of the coordinate axis indicating the luminance component, The selection of a representative color in which the chromaticity is emphasized is the means for solving the above problems.

Further, in the limited color determination method of the present invention, a representative color is obtained by performing a discriminant analysis on the frequency distribution of each pixel value of the image data represented in the uniform color space, and the obtained representative color is obtained. , The frequency distribution in the uniform color space is obtained again, and a discriminant analysis is performed on this frequency distribution to obtain a smaller number of representative colors. By repeating this, the representative colors are reduced stepwise. To obtain the frequency distribution at each step or at any step, reduce the rank on the coordinate axis indicating the luminance component,
Changing the ratio with the class number on the coordinate axis indicating the chromaticity component is the means for solving the above problems.

Then, the limited color determining method of the present invention reads the lower bits of the digital data representing the values of the three primary colors of each pixel of the color image data, and obtains the periodicity in which the same value appears in the lower bits, On the other hand, for each pixel value that represents the color image data in the coordinate system of the uniform color space, a frequency distribution on the coordinate axis indicating the luminance component and the coordinate axis indicating the chromaticity component in the uniform color space is obtained, and the frequency distribution is calculated. Weighting correction is performed based on the periodicity data, and the frequency distribution is corrected by differentiating the weighting-corrected frequency distribution from the number of classes on the coordinate axis showing luminance and the number of classes on the coordinate axis showing chromaticity, and discriminant analysis Was carried out as a means for solving the above-mentioned problems.

Further, the limited color determining apparatus of the present invention obtains a representative color by performing a discriminant analysis on the frequency distribution of each pixel value of the color image data represented by the coordinate system of the uniform color space, and the color The image data is approximated by a limited representative color, and the class number of the coordinate axis indicating the luminance component in the uniform color space and the class number of the coordinate axis indicating the chromaticity component are different, and the color image data in each coordinate axis And a representative color determining means for determining a representative color by performing a discriminant analysis on the obtained frequency distribution, and a means for solving the above problems. did.

Further, the limited color determining apparatus of the present invention obtains a representative color by performing a discriminant analysis on the frequency distribution of each pixel value of the color image data represented by the coordinate system of the uniform color space, and the color The image data is approximated by a limited representative color, the periodicity calculating means for obtaining a periodicity coefficient in which the same value appears in the lower bits with respect to the digital data representing the three primary colors of the color image data, and the equality. The frequency distribution calculating means for obtaining the frequency part of each pixel value of the color image data with respect to the coordinate axis showing the luminance component and the coordinate axis showing the chromaticity component in the color space, and the frequency distribution obtained by the frequency distribution calculating means is the periodicity. A frequency distribution correction means for performing weighted correction with a coefficient, and a coordinate number class indicating a luminance component and a coordinate indicating a chromaticity component for the weighted frequency distribution. The frequency distribution by correcting the number of classes of the shaft as a different, that is formed by and a representative color determining means for determining a representative color by performing a discriminant analysis was solutions of the problems.

[0019]

According to the structure described in claim 1, when the representative color is obtained by performing the discriminant analysis on the frequency distribution of the pixels for each coordinate axis for the image data converted into the uniform color space, By making the class number of the coordinate axis showing the luminance different from the class number of the coordinate axis showing the chromaticity, it is possible to emphasize the luminance or the chromaticity when the representative color is selected. That is, considering the color tone of the image data and the fact that human vision is sensitive to luminance, etc., the selected representative color may have a large variation on the luminance side or a large variation on the chromaticity side. It becomes possible to do.

According to the second aspect of the invention, the representative color in which the brightness is emphasized is selected according to the human visual characteristics sensitive to the brightness, and the representative color corresponding to the human visual sense is selected. You can choose the color. For example, if only a small number of representative colors are selected in the part where the color of the image data changes gently, false contours will occur,
In particular, since human vision is sensitive to brightness, when a color having similar brightness is selected as the small number of representative colors, false contours are emphasized. That is, it is possible to prevent false contours from being emphasized by selecting a representative color that is highly variable in brightness.

An image in which false contours are easily emphasized as described above is one in which its color distribution is concentrated in the same hue and has various luminance components. In such image data, False contours can be prevented by selecting a representative color that emphasizes brightness in particular.

According to the third aspect of the invention, in the image data, when the color distribution is dispersed in various hues and concentrated on a relatively small number of luminance components, the human vision Since false contours are not emphasized even if they are sensitive to luminance, it is possible to display representative colors by selecting representative colors with enhanced chromaticity, that is, representative colors of various hues. The image to be displayed can be close to the original image data.

According to the structure of claim 4, when the representative color is selected by the discriminant analysis, the color is not reduced to the representative color required at one time, but for example, the original image data is equalized. Obtain the frequency distribution in the color space, make the representative color 10,000 colors by discriminant analysis, find the frequency distribution in the uniform color space of this representative color, and make the representative color 1000 colors by the discriminant analysis. To do so. Then, at each stage, by performing a discriminant analysis by emphasizing luminance or chromaticity corresponding to the color distribution of the original image data or the representative color of each stage, statistical analysis such as the color distribution of the original image data is performed. It is possible to select a representative color suitable for the property.

That is, it is possible to deal with the statistical properties such as the color distribution of image data by the process of repeating the non-adaptive method of selecting the representative color by the discriminant analysis as in the adaptive method which requires complicated processing. It is possible to select the selected representative color.

Further, according to the fifth aspect of the invention, in the portion where the color gradually changes in the image data, when the color data of the pixels arranged continuously are read in order, the lower order When the value of the bit is increased, it is moved up to the higher bit, and the series of cycles in which the value of the lower bit is decreased at once or the cycle in the opposite direction is repeated.

Therefore, the pixels successively arranged in the portion where the color change is gentle in the image data have a periodicity in the value of the lower bit, and the pixels arranged continuously. By examining the periodicity, it is possible to identify the portion where the color of the image data changes gently and the color of that portion, and it is possible to estimate the degree of the gentle color change from the strength of the periodicity. .

Therefore, when a histogram showing the frequency of appearance of colors in the image data is created and discriminant analysis is performed, as described above, in the image data, the portion where the color changes gently is specified, and the period of the lower bits is determined. For a color having a gentle change in color, a weighting coefficient is determined according to the strength of the periodicity, and the histogram is weighted using this weighting coefficient to perform a discriminant analysis. It is possible to increase the number of representative colors to be displayed, and suppress false contours that are generated by displaying a portion where the color change is gentle with a slight representative color.

Further, according to the limited color determining device of the sixth aspect, the frequency distribution calculating means obtains the frequency distribution by making the class number of the coordinate axis indicating the luminance component different from the class number indicating the chromaticity component, Since the representative color determination means determines the representative color based on the frequency distribution, the selected representative color is set to the luminance side in consideration of the color tone of the image data and the human visual sensitivity to luminance. It is possible to make many changes or make many changes on the chromaticity side.

Further, according to the limited color determining device of the seventh aspect, the periodicity calculating means obtains the periodicity coefficient indicating the gentleness of the color change, and the frequency distribution correcting means obtains the frequency obtained by the frequency distribution calculating means. Since the representative color determination means performs the discriminant analysis after the distribution is weighted and corrected by the periodicity coefficient, a large number of representative colors are selected from the portion where the color change is gentle. Further, the frequency distribution correcting means corrects the frequency distribution by setting the class number of the coordinate axis indicating the luminance component and the class number of the coordinate axis indicating the chromaticity component to be different, so that the tone of the image data and the gradual change in color are taken into consideration. The selected representative color is selected.

[0030]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a schematic view of a limited color determination device according to this embodiment, and FIGS. 2 to 8 are drawings for explaining a limited color determination method according to this embodiment. The limited color determination apparatus of this embodiment will be described below with reference to FIG.

The limited color determining apparatus described above includes an image storage means 1 for storing image data, a lower 3 bit storage means 2 for storing lower 3 bits of the image data, and each pixel in the lower 3 bit image data. From this, the value of the lower 3 bits of the surrounding pixels is read along the radiation direction, the periodicity of the lower 3 bits of the read pixel is obtained, and the periodicity calculating means 3 for normalizing the periodicity, and the periodicity. A periodicity coefficient storage means 4 for storing the periodicity normalized by the arithmetic means 3 as a coefficient, and a color conversion means 5 for converting the image data composed of RGB values into values in the L ^* a ^* b ^* space. And the L ^* a ^* b ^* space is subdivided into a large number of clusters according to the class number when the frequency distribution of each pixel is obtained on each coordinate axis of L ^* , a ^* , and b ^* , and these clusters are assigned cluster numbers, L ^The pixel value converted to the ^* a ^* b ^* space value is
A cluster image storage means 6 which replaces and stores the cluster number of the cluster including the pixel, a histogram calculation means 7 for obtaining the frequency distribution of pixels for each cluster for each coordinate axis of the L ^* a ^* b ^* space, and A histogram storage means 8 for storing the obtained histogram, an average / variance calculation means 9 for obtaining an average color / variance from the histogram, and an average / variance storage means 10 for storing the obtained average / variance.
A first spindle calculating means 11 for obtaining a first spindle, a first spindle storing means 12 for storing the obtained first spindle, a weighted variance computing means 13 for obtaining a weighted variance, and a division threshold by discriminant analysis. Dividing threshold value determining means 14 for obtaining the dividing threshold value, dividing threshold value storage means 15 for storing the obtained dividing threshold value,
The color region dividing unit 16 that divides the color region according to the division threshold according to the weighted dispersion and the first main axis, and until there is no color region that can be divided or the number of divided color regions reaches a predetermined number of display colors, A division control means 17 for controlling the processing by each means to be repeatedly performed, and a coordinate axis indicating a luminance component L ^* according to a ratio of dispersion to each coordinate axis of the L ^* a ^* b ^* space before dividing into each color area. An emphasis coefficient specifying means 18 for specifying how many times (1/8 times to 32 times) the class number is multiplied with respect to the class number on the coordinate axis indicating the chromaticity components a ^* b ^* , and the emphasis coefficient specifying means 18 Emphasis coefficient calculation means 19 for obtaining the class number of each coordinate axis by the magnification (emphasis coefficient) specified by, and emphasis coefficient calculation means 19
The weighting correction of the histogram read from the histogram storage means 8 is performed by the emphasis coefficient storage means 20 that stores the rank of each coordinate axis obtained by the above, and the periodicity coefficient read from the periodicity coefficient storage means 4, and , The histogram updating means 21 for changing the class number of the histogram to the class number stored in the emphasis coefficient storage means 20, rewriting the histogram, and storing the rewritten histogram in the histogram storage means, and the above-mentioned means 2 to 2
The inverse color conversion unit 22 that reads out the limited color obtained by performing the operation 1 from the average / dispersion storage unit and performs the inverse conversion of the color conversion unit 5, and the representative color allocation unit 23 that replaces the color of each pixel with the limited color. , The limited color color pallet means 24 for sorting the limited colors according to the distance in the color space and registering the representative colors in the color map 30a, and the image data replaced with the representative colors on the screen which is approximately displayed, Color correction means 25 that enables partial color correction, and each of the above means 2
It is mainly composed of a repetitive control means 26 for controlling the repetitive operation of each of the above means 2 to 21 for the limited color obtained by performing the operation of 21.

The image storage means 1 has an image memory 10a and stores (R, G, B) values (each color component is 8 bits) of each pixel of image data. The lower 3 bits storage unit 2 stores the lower 3 bits of the (R, G, B) value of each pixel of the image data stored in the image storage unit 1.

The periodicity calculation means 3 reads the lower 3 bits of data of each pixel from the lower 3 bits storage means 2 and radiates vertically, horizontally, diagonally left, diagonally right around each pixel on the image. The number of times that the same value appears in the lower 3 bits at a period of 4 pixels or less in the pixels arranged continuously in the direction is measured, and the period value described later is obtained for each pixel. Further, a ratio when the maximum value of the cycle value is set to 1 is calculated as a cyclic coefficient for each pixel.

The periodicity coefficient storage means 4 stores the periodicity coefficient for each pixel in the same array as the pixel. The color conversion means 5 converts the RGB value of each pixel of the image data into coordinates in the L ^* a ^* b ^* space. The cluster image storage unit 6 subdivides the L ^* a ^* b ^* space into clusters according to the class of each coordinate, and stores the values of pixels included in each cluster by replacing them with cluster numbers. The histogram calculation means 7 obtains the frequency distribution of each pixel belonging to each cluster, creates a histogram, and stores it in the histogram storage means 8.

The average / dispersion calculating means 9 finds the average and variance of the pixel distribution for each coordinate axis in the color area obtained by dividing the L ^* a ^* b ^* space or the L ^* a ^* b ^* space by a threshold value. The first main axis computing means 11 obtains the maximum coordinate axis of the variance read from the average / dispersion storage means 10 in the color region obtained by dividing the L ^* a ^* b ^* space or the L ^* a ^* b ^* space by a threshold value. The first spindle. The first spindle storage means 12 stores the obtained first spindle.

The weighted variance calculating means 13 obtains the weighted variance by multiplying the variance by the relative frequency of the number of pixels included in the color region with respect to the total number of pixels. The division threshold determination means 14 determines the division threshold by the discriminant analysis. The division threshold storage unit 15 stores the division threshold obtained by the division threshold determination unit 14.

The color area dividing means 16 sorts the weighting variance values of the respective areas obtained by the weighting variance calculating means 13 to obtain the color area having the largest weighting variance, and the color area (in the case of the first division) The first main axis of the color space) is read from the first main axis storage means 20, and the color region (color space) is divided into two by the division threshold of the first main axis read from the division threshold value storage means 15. The division control unit 17 repeatedly performs the above-described calculation and division for each color region until the total number of color regions divided by the color region division unit 16 reaches a prescribed number of colors or no more division is possible. Control of.

After the above division is completed, the representative color allocating means 23 averages the average colors of the respective color areas. The average color of the original image data is read from the distributed storage unit 10 and the average color thereof is adopted as the representative color of each color area, and the average color of the color area is assigned to the pixels included in each color area in the original image data. Replace the pixel color with the representative color. The image data in which the color of each pixel is replaced with the representative color is stored in the image memory 30d.

The emphasis coefficient designating means 18 obtains the ratio of each coordinate axis with respect to the variance of each coordinate axis of the L * a * b * space obtained by the average / dispersion computing means 9, and calculates the rank of each coordinate axis as described above. 1/8 with respect to other coordinate axes depending on the ratio
2 times, 1/4 times, 1/2 times, 2 times, 4 times, 8 times, 16 times,
Specify 32x or specify the magnification. The emphasis coefficient calculation means 19 obtains the class number of each coordinate axis according to the magnification.

The emphasis coefficient storage means 20 stores the rank of each coordinate axis obtained by the emphasis coefficient calculation means 19. The histogram updating means 21 is the periodicity coefficient storing means 4
The periodicity coefficient for each pixel read from is added for each cluster including the corresponding pixel, and the sum of the periodicity coefficients for each cluster and the number of pixels included in each cluster are added. Then, the value obtained for each cluster as described above is newly set as a histogram.

Further, the histogram updating means 21 subdivides the L ^* a ^* b ^* space into new clusters different from the clusters according to the ranks of the coordinate axes stored in the emphasis coefficient storage means 20, and The histogram is changed to the number of pixels for each new cluster that has been subdivided.

That is, the L ^* a ^* b ^* space is divided into new clusters according to the magnification designated by the emphasis coefficient designating means 18, and the histogram is updated based on the new clusters. Therefore, in the case where the weighting correction is performed by the periodicity coefficient and the luminance component is emphasized at the magnification specified by the emphasis coefficient specifying means 18, after the L ^* a ^* b ^* space is divided into new clusters, each new cluster is divided. Then, the histogram of the new clusters is weighted and corrected with the sum of the periodicity coefficients of each new cluster. Then, the histogram update means 21 rewrites the histogram stored in the histogram storage means 8 with the updated histogram as described above.

The limited color palette calculator 24 sorts the representative colors obtained by the operations of the above units according to the distance in the color space, and registers the representative colors as color codes in the color map 30a. The color correction means 25 changes the color of the display image consisting of the limited color obtained by the operation of each of the above means on the screen of the display 30b displaying this image,
It is used when you want to change it partially.

Then, the color correction means 25 encloses a portion of the display image whose color is to be changed with a rectangle according to an instruction of the mouse, selects the color by clicking the color to be changed within the rectangle with the focus of the mouse. By doing so, the selected color portion is highlighted in reverse on the color palette screen, and by selecting in order from the colors in the vicinity,
The color can be replaced in order from the closest color in the color space.

The limited color determination device having the above configuration displays the full-color image data in the limited display color on the color map display device 30 using the limited color determination method. In the present embodiment, the color map display device 30 is also capable of displaying the image data stored in the image storage means in full color, and comparing the full color image with the limited color image, The color can be corrected by the color correcting means 28. The reference numeral 30c indicates the image memory 10a,
It is a D / A converter for displaying the image data of 30d on the display 30b.

Next, a limited color determination method using the limited color determination device will be described. (A) Measurement of the degree of color change of image data to be displayed In the conventional method, the degree of color change of image data, that is, the color of each pixel in the image and the color of pixels in the vicinity of that pixel are large. When measuring whether they are different or have substantially the same color, there are many cases where the upper 5 bits of the RGB data of the image represented by 8 bits for each RGB are used for measurement.

However, the information showing the gradualness of color change is
Rather, it is included in the lower 3 bits lost by this. That is, in image data with little color change,
Even if the upper bits of each pixel were compared, the degree of color change could not be measured, and the part where the color change was extremely gentle,
It cannot be distinguished from the part where the color change is moderate to some extent. Therefore, in this embodiment, the lower 3 of each color component is
By examining the periodicity of the bit values, the gradualness of color change in the image data to be displayed is specified.

By the way, the values of the lower 3 bits are from 0 to 7
Is limited to 8 values up to. When the color change due to the difference in the position of each pixel of the image is gradual, the value of the data of each pixel continuously changes with a small variation amount between the pixels arranged continuously. Therefore, in the part where the color change is gradual, a fixed value range within the range of 0 to 7 of the lower 3 bits appears substantially regularly. In view of this, in the present embodiment, the periodicity calculation means 3 is, as shown in FIG.
Vertical (top / bottom), horizontal (left / right), diagonal left (upper left / lower right), diagonal right (upper right / lower left) with each pixel 41 as the center
The lower 3 bits (stored in the lower 3 bit storage means 2) of the pixels continuously arranged in the eight emission directions of are read, and the regularity (periodicity) is obtained for each of the RGB color components.
Find out.

A concrete method of calculating the periodicity by the periodicity calculating means 3 is as shown in FIG.
The value of the lower 3 bits of the digital data of the pixel 41 is read while sequentially moving pixel by pixel until the same value as the value of the lower bit of appears. Then, if the number of pixels moved by the time the first same value is found is 4 or less, the lower 3 bits are read while successively moving to the next pixel until the next same value appears, and then the same value is found. If the number of moving pixels at that time is equal to the initial number of moving pixels, it is considered that the pixel of interest 41a has periodicity.

Then, while the same value appears periodically with the same movement number, the lower 3 bits of the color data of the pixels 41 are read while moving along the direction. Then, the number of appearances in which the same value appears with the same movement number is counted.
And when the same value does not appear periodically,
The reading of the pixel data is completed. The above operation is performed for all the eight directions, and the total number of appearances of the same value is defined as the periodic value of the pixel of interest 41a.
Further, the above work is performed by using all the pixels 41 of the image data as the target pixel 41a. It should be noted that such measurement of the cycle value is performed for each of the RGB color components.

The above work is performed, for example, in the case shown in FIG.
When attention is paid to the pixel 41a on the image data, when looking at the values of the pixels 41 ... Continuing from the pixel 41a in the horizontal direction (to the right in FIG. 2), for example, every three pixels, a pixel 41b having the same value is obtained. May appear. In such a case, the number of appearances of the same value that appears in the cycle of 3 pixels is counted, the same operation is performed for the remaining radiation directions, the number of appearances of the same value is counted, and all are added. The period value of the pixel 41a is used.

Next, the period value of each color component is added for each pixel. Then, the sum of the periodic values for each pixel is calculated with a ratio with the peak being 1, and it is defined as a periodicity coefficient used for weighting described later. The periodicity coefficient b _i is _calculated by the following equation. b _i = a _i / a _max Here, i represents each pixel, a _i represents the sum of the periodic values in each pixel, and a _max represents the maximum periodic value among the sums a _i of the periodic values. It shows a _i (maximum value).

The periodicity coefficient b _i thus obtained is stored in the periodicity coefficient storage means 4. Then, in the histogram updating means 21, the periodicity coefficient b _i is assigned to the position of each pixel converted into the uniform color space, and the periodicity coefficient b _i is arranged at the position of each pixel in the uniform color space. Become.

When the histogram updating means 21 updates the histogram showing the number of pixels included in each cluster in the uniform color space divided into clusters, each of the pixels included in each cluster is updated. The periodicity coefficients b _i are added, and the sum of the periodicity coefficients b _i is used as the weighting coefficient for each cluster. Then, these weighting factors are respectively added to the number of pixels of each cluster to form a new histogram. Further, when performing the discriminant analysis on the first principal axis, the first
The number of pixels for each class of the main axis, that is, the frequency distribution with respect to the first main axis is obtained, the number of pixels included in each class of the first main axis is divided by the total number of pixels from this frequency distribution, and the occurrence probability of the pixel is obtained.
This probability of occurrence is used.

(B) Selection of representative color In order to select the representative color, first, the color distribution of the image data is arranged in the color space. At this time, considering the visual characteristics of the person,
Use uniform color space. As a uniform color space, the above L ^*
Use u ^* v ^* or L ^* a ^* b ^* . Therefore, the value of each pixel of the image data composed of RGB data is set to L ^* u ^* v ^* or L
Convert to ^* a ^* b ^* . In this embodiment, as an example, the conversion by the color conversion means 5 will be described in detail using L ^* a ^* b ^* . Note that the uniform color space of the present invention is L ^* a ^* b
^* Not limited to space.

Conversion from RGB data to L ^* a ^* b ^* is performed by the following equations via the XYZ system recommended by CIE 1931. X = 0.61R + 0.17G + 0.20B (1a) Y = 0.30R + 0.59G + 0.11B (1b) Z = 0.066G + 1.12B (1c)

When Y / Yn> 0.008856 L ^* = 116 (Y / Yn) ¹ / 3-16 (2a) When Y / Yn ≦ 0.008856 L ^* = 903.3 (Y / Yn ) …… (2b)

X / Xn> 0.008856, Y / Yn>
0.008856, Z / Zn> 0.008856 a ^* = 500 [(X / Xn) ¹ / 3- (Y / Yn) ^1/3 ] (2c) b ^* = 200 [(Y / Yn ) ¹ / 3- (Z / Zn) ^1/3 ] (2d)

0.008 for X / Xn, Y / Yn, Z / Zn
When there are 856 or less, the above formulas (2c) and (2
d) (X / Xn) ^1/3 , (Y / Yn) ^1/3 , (Z / Zn)
^1/3 term is 7.787 (X / Xn) +16/11 respectively
6,7.787 (Y / Yn) +16/116, 7.78
Replace with 7 (Z / Zn) +16/116. here,
The chromaticity of R, G, B and white complies with the NTSC television standard.

(C) Frequency distribution of pixels in L ^* a ^* b ^* space The value of each pixel deviates from an 8-bit integer value at the stage of converting RGB to L ^* a ^* b ^*. become.
Therefore, in order to obtain the frequency distribution in the L ^* a ^* b ^* space, the class number of each coordinate axis is determined, the L ^* a ^* b ^* space is divided into clusters according to the class of each coordinate axis, and the pixels included in the cluster are determined. You need to find the number.

To divide the L ^* a ^* b ^* space into the required number of colors and select the average color of each color region as the representative color, Watanabe's algorithm (Journal of the Institute of Electronics, Information and Communication Engineers, D Vol.
J70-D No. 4 pp. 720-726 198
(April 7) and improved. The main improvement is that instead of using the upper 5 bits of RGB as the original data, R
A value obtained by converting the GB 8-bit value into L ^* a ^* b ^* was used. As a result, the 8-bit precision of the data can be used as it is.

Further, the class number of each coordinate axis of the initial space for division is, for example, from RGB 32 × 32 × 32 to L ^* a ^* b ^*.
Like 256 × 64 × 64, the class of the coordinate axis showing the luminance component L ^* is made four times as large as the class of the coordinate axis showing the chromaticity component a ^* b ^* , and changed so as to emphasize the luminance component L ^*. It was
This is the report of Kira et al. (Science Techniques, IE83-92,
"Adaptive natural color image display with a limited number of representative colors"
(Kira, Inoue, Fukui) that human visual characteristics are sensitive to the luminance distribution about four times as much as the chromaticity component.

More specifically, the input image data usually has RGB values arranged in pixel order. Therefore,
When considering an RGB space whose coordinates are RGB values, RGB
The space is a three-dimensional space in which R, G, and B are coordinate axes, and each coordinate axis has an 8-bit value, that is, a value of 0 to 255. Therefore, by associating the number of pixels having the RGB values with the point on the space, the histogram of each color component can be obtained.

That is, histogram (r, g, b) = RGB (r,
It is the total number of pixels of g, b). Next, the uniform color space L ^* a ^* b ^*
In the case of the histogram of, the value (r, g, b) of each pixel represented by RGB is converted into (L ^* a ^* b ^* ) by the above equation. That is, histogram (L ^* a ^* b ^* ) = L ^* a ^* b ^*
It is the total number of pixels of (L ^* a ^* b ^* ). However, the RGB values are
Since each is 8-bit digital data, it is an integer value of 0 to 255, but the value of (L * a * b *) is 0 by the above conversion operation.
Since it deviates from the integer value of ˜255 and emphasizes the luminance component, it is necessary to provide an appropriate class number.

In this embodiment, in order to emphasize the luminance, that is, the L ^* component as described above, and to prevent the data volume from increasing due to the conversion into L ^* a ^* b ^* , each L
^It is necessary to determine the class spacing of ^* a ^* b ^* . As the data capacity, since each value of RGB is set to 8 bits, the maximum is set to 8 bits. Therefore, the luminance L ^* to be emphasized takes a value of 0 to 255. Therefore, the value of the L ^* component of the pixel obtained by the above equation is assigned to 0 to 255 corresponding to its size. Also, a
Regarding the ^* and b ^* components, a ^* and b ^* : L ^* are set to 1: 4 in order to emphasize the luminance component four times. That is, a ^* , b ^*
The component takes 6 bits and values of 0 to 63.

Here, the calculation of the actual class interval is performed for each L ^*.
When a ^* b ^{* is set} to data of 8 bits or less, it is calculated by the following formula. L ^* interval = (L ^* max-L ^* min) / (256 / m) a ^* interval = (a ^* max-a ^* min) / (256 / n) b ^* interval = (b ^* max-b ^* min) ) / (256 / n) L ^* interval, a ^* interval, b ^* interval is L ^* ,
The class spacing of a ^* and b ^* is shown. L ^* max, a ^* max, b ^* max
^Indicates the maximum values of L ^* , a ^* , and b ^* , respectively. L ^* min, a ^*
min and b ^* min represent the minimum values of L ^* , a ^* , and b ^* , respectively. Further, m and n are changed according to the ratio of the luminance component (L ^* ) and the chromaticity component (a ^* , b ^* ). In this embodiment, L ^* : a ^* , b ^{* is set} to 4 :. In order to obtain 1, the luminance component m is 1, and the chromaticity component n is 4.

Then, the L ^* a ^* b ^* space in which the class interval is determined as described above is three-dimensionally divided by the class of each coordinate axis, and the divided areas are set as color clusters. Then, each cluster is given a cluster number. And
The value of each pixel is replaced with the cluster number of the cluster including the pixel, and this is stored in the cluster image storage means 6. Then, the histogram calculation means 7 obtains the frequency distribution for each cluster of the image data.

A case will be briefly described with reference to FIGS. 4 and 5 in which the class number of the coordinate axis indicating the luminance is made larger than the class number of the coordinate axis indicating the chromaticity and the representative color by the discriminant analysis is selected. 4 and 5, a two-dimensional coordinate system will be described with one coordinate axis indicating chromaticity in the uniform color space.

First, when the y-axis is the luminance, the x-axis is the chromaticity, and each class is 8, the color space is divided by discriminant analysis as shown in FIG. And
When the number of classes on the x-axis is set to 4 so that the number of classes on the y-axis is twice as large as the number of classes on the x-axis, as shown in FIG. Cannot be divided, and the number of divisions of the luminance axis increases accordingly. That is, when the average color of the divided color space is used as the representative color, many colors having the same hue but different luminance are selected as the representative color, and colors of different hues are less often selected, and the representative color with enhanced luminance. Will be selected.

(D) Enhancement of Luminance or Chromaticity In the present embodiment, in the space division method, by emphasizing the sampling of the luminance component, the portion where the change in the luminance component is gentle is emphasized, that is, the luminance direction. Focusing on the fact that the division with the first main axis as the main axis is the main division, and by giving the sampling in which the luminance component is emphasized to the chromaticity component, conversely, the method of specifying the part and color where the luminance change is gentle Is.

Specifically, L ^* a ^* b ^* (or L ^* u
^* v ^* ) 16 in the luminance direction, such as 256 × 16 × 16
The number of limited colors is, for example, 10.
Designate with a small number such as 0 color, and obtain the representative color by the division method of (b). As a result, 100 colors, which are 16 times emphasized in this case, are selected for the luminance component. These are the top 100 colors that emphasize the colors of the part where the luminance change is gentle.

Then, as described above, the emphasis coefficient designating means 18 obtains the ratio between the coordinate axes with respect to the variance in the coordinate axis directions of the uniform color space obtained by the average / dispersion computing means 9, and corresponds to the ratio. Then, the magnification (enhancement factor) of the class of the coordinate axis indicating the luminance with respect to the class of the coordinate axis indicating the chromaticity is 1/8 times, 1/4 times, 1/2 times, 1: 1, 2
It was made possible to select any of 4 times, 4 times, 8 times, 16 times, 32 times, and 64 times.

Then, the emphasis coefficient calculation means 19 determines the class of each coordinate axis according to the magnification determined as described above. For example, when the magnification of the chromaticity component a ^* b ^* with respect to the luminance component L ^* is doubled, the class number of the coordinate axis of each L ^* a ^* b ^* is 256 × 128 × 128. An emphasis coefficient storage means 20 for class intervals determined by the class number.
Remembers.

(E) Change of histogram The weighting coefficient for each pixel stored in the periodicity coefficient storage means 4 and the class number stored in the emphasis coefficient storage means 19 are used to store in the histogram storage means 8. Update the histogram.

First, the update of the histogram using the periodicity coefficient stored in the periodicity coefficient storage means 4 will be described. First, the periodicity coefficient stored in the periodicity coefficient storage means 4 for each pixel is arranged at the position of each pixel converted into the uniform color space, and the periodicity coefficient of each pixel included in each cluster is clustered. Add every time. Then, a histogram is created with the sum of the periodicity coefficients added for each cluster and the number of pixels for each cluster as the number of pixels for each cluster, and this is used as a new histogram.

That is, the histogram 43 stored in the histogram storage means 8 as shown in FIGS.
(In fact, a histogram in a uniform color space, that is, a three-dimensional space, but here, a one-dimensional histogram is used for ease of explanation.) By adding the histograms representing the sums for each cluster, the weight-corrected histogram 45 shown in FIG. 8 can be obtained, and using this, the representative color is selected by the discriminant analysis described later.

Further, the histogram updating means 21 reconstructs the histogram using the class numbers stored in the emphasis coefficient storage means 20. In the histogram calculation means 7, the histogram is created with a class number in which the chromaticity component with respect to the luminance component is quadrupled regardless of the color tone of the image data. Decide

Then, the uniform color space is redivided into new clusters of a new class, and the number of pixels for each new cluster is used as a new histogram. Further, when weighting is performed with the periodicity coefficient, the sum of the periodicity coefficients for each new cluster is used as a weighting coefficient, and the weighting coefficient is added to the number of pixels for each new cluster to form a new histogram. Then, the histogram stored in the histogram storage means 8 is updated to the new histogram obtained as described above.

(E) Selection of representative color by discriminant analysis method The selection of representative color by the discriminant analysis method of the present embodiment is basically the same as that of the above-mentioned prior art, and a threshold value is obtained using a histogram for each color component. Is obtained, and the color space is sequentially divided into a number of color regions at this threshold value.In the present embodiment, the histogram weighted by the periodicity and reconstructed by the emphasized rank is used as a histogram. Perform discriminant analysis using.

Here, a method of determining the threshold T _i by discriminant analysis will be described. If the number of pixels in the region S is N and the number of pixels in an arbitrary class i on one coordinate axis is n _i , the occurrence probability q _i
Is expressed by the following equation (Equation 1). The number of pixels at this time is obtained from the histogram updated by the histogram updating means 21 as described above.

That is, the value obtained by adding the weighting coefficient to the number of pixels of the new cluster divided by the new class is set as the number of pixels of each cluster, and the number of pixels of each cluster included in an arbitrary class of one coordinate axis is set. The sum of the added pixels is the number of pixels n _i
Used as.

[Equation 1] Then, the section [a, b] is divided into two sections C ₁ [a, k] and C ₂ [k +] with a value k (value in units of classes) on the coordinate axis.
1, b], the pixel occurrence probabilities ω ₁ and ω ₂ in both sections, the average values μ ₁ and μ ₂ , and the variances σ ₁ ² and σ ² .
_{When 2} ² is set, the value k on the coordinate axis is determined so that the weighted dispersion sum σ _w ² given by the following equation (Equation 2) is minimized.

[Equation 2]

[0082] It is equivalent to the variance sigma _B ² between the two given a weighted variance sum sigma _w ² by minimizing the following expression (Expression 3) interval is maximized.

[Equation 3]

The k ^* (that is, the threshold T _{i to be obtained} ) that maximizes σ _B ² can be sequentially obtained as in the following equation (Equation 4).

[Equation 4] Here, μ _T is a total average, and is represented by the following equation (Equation 5).

[Equation 5]

As described above, the threshold value T _i of each color component can be determined. Then, the color space is first divided into two regions by the obtained threshold value T _i , but the coordinate axis of the color component having the largest variance σ ² among the color components is set as the first principal axis. , The threshold value T _{i of the} color component
At the position of the first main axis corresponding to, the color space is divided into two color regions on a division plane orthogonal to the first main axis.

Next, the weighted dispersion sum σ _w ² of each of the divided color areas is obtained, these are compared between the color areas, and the color area having the largest weighted dispersion sum σ _w ² is divided next. The area. Then, in the color area, the threshold value T _i and the first principal axis are obtained as if the color space was divided, and the color area is further divided into two color areas. Next, in all the divided color areas, a color area having the largest weighted variance sum σ _w ² is obtained, and the obtained color area is divided into two as described above. Then, when the number of divided color areas becomes equal to the number of representative colors to be displayed, the division is ended. Next, the average value of each color component in each color region is converted from L ^* a ^* b ^* to RGB by the inverse color conversion means 22 to be a representative color.

Then, this representative color is registered in the color map 30a of the color map display device 30 as a color code, and the color distribution in the color space of the color image data to be displayed is made to correspond to each of the color areas. A color similar to the original color of the pixel is assigned to the pixel as a color code from the color map 30a. In this way, the image data in which the color code is assigned to each pixel and the color is limited is displayed on the display 30b of the color map display device 30 having the color map 30a in which the color code is registered.

Further, the image data of the limited color can be stored in the storage device or the like, and the image once processed can be instantly displayed on the display 30b, and the limited image can be displayed. Since the memory capacity of the data is smaller than that of the original image data of the natural image, an apparatus having a small memory capacity can easily handle the data.

(G) Determination of a representative color by repeating the above operation. In the limited color determination method of the present embodiment, the selection of the representative color determined as described above is repeated for the selected representative color, thereby gradually reducing the representative color. That is, the repetitive control means 17 selects the representative color by the above-described operation again using the image represented by the representative color selected as described above as the original image data.
This is repeated a set number of times to reduce to the finally required representative color.

When the representative color is repeatedly selected, the histogram is corrected by the periodicity coefficient calculation means 3 and the emphasis coefficient calculation means 19 each time, and the histogram update means changes the color change gently. The parts are weighted and the brightness or chromaticity is emphasized. Therefore, in the image data, the representative colors are narrowed down while being adjusted with respect to the statistical properties such as the color distribution and the luminance distribution, and the representative color suitable for the statistical properties of the image data is adjusted similarly to the adaptive method. The color will be selected.

As described above, the limited color determination apparatus of the present embodiment identifies the portion where the color change is gentle by the periodicity of the lower 3 bits of the image data, and using this periodicity as a coefficient, the color distribution of the image is determined. Is corrected by weighting, and many colors are selected from the part where the color change is gradual, and the number of classes of the coordinate axis showing the luminance is made larger than the number of classes of the coordinate showing the chromaticity to correspond to the characteristics of human vision. By selecting many colors with different brightness,
False contours can be prevented.

Further, by comparing the variances of the luminance and the chromaticity on the respective coordinate axes, regarding image data with a large variation in the luminance,
Further, by emphasizing the brightness and emphasizing the chromaticity for an image having a large number of hues with a small change in the brightness, a representative color suitable for the property of the image data can be selected.
Further, by repeating the above operation and gradually reducing the representative colors, the measurement of the periodicity of the lower 3 bits and the comparison of the variances of the luminance and the chromaticity on each coordinate axis are repeated, and the histograms corresponding to these are repeated. By repeating the above correction, the representative colors can be narrowed down to those that are adapted to the statistical properties of the image data.

It is also possible to configure and execute the entire processing as a program on a general-purpose computer. Further, the memory capacity per pixel of the original image data does not necessarily have to be 24 bits, and further, the lower bits used for specifying the color change need not be the lower 3 bits. How many lower-order bits are taken may be changed depending on the memory capacity per pixel of the original image data. Further, although it takes a certain amount of processing time to reduce the number of colors according to the present invention, decoding processing is not required when displaying a compressed image obtained by this, so that it does not depend on a machine or a system and is instantaneously displayed. It is possible to display a compressed image with limited colors. Due to these facts, the limited color determination method can be expected to be applied to many industrial fields.

[0093]

With the method and apparatus for determining a limited color according to the present invention described above, a natural image which should originally be displayed on a full-color display device can be displayed on a color map display device which can simultaneously display only a small number of limited colors. Can be displayed without.

That is, when the representative color is determined by the discriminant analysis, by making the class number on the coordinate axis showing the luminance different from the class number on the coordinate axis showing the chromaticity, the distribution of the luminance and chromaticity of the image is obtained. Corresponding to, it is possible to select a representative color with emphasized luminance or a representative color with emphasized chromaticity. In particular, by selecting a representative color in which brightness is emphasized, it is possible to select a representative color corresponding to human visual characteristics sensitive to brightness, and it is possible to suppress false contours from being emphasized.

Further, the gradual color portion is specified by the periodicity of the lower bits, the histogram is corrected, and the representative color selected from the gradual color portion is increased, so that the false contour is surely obtained. Can be suppressed. Therefore, 1
Especially, it is necessary to reduce the number of colors, such as 00 or less, C
Even for a natural image such as a D-ROM, it is possible to effectively deal with it by specifying a portion where a feeling of strangeness is conspicuous and obtaining a limited number of specified colors from which there is the least feeling of strangeness. It is very effective in compressing.

[Brief description of drawings]

FIG. 1 is a schematic view showing a limited color determination device of the above embodiment.

FIG. 2 is a diagram for explaining a method of determining a limited color according to the above embodiment.

FIG. 3 is a diagram for explaining the limited color determination method.

FIG. 4 is a drawing showing a coordinate system for explaining the limited color determination method.

FIG. 5 is a diagram showing a coordinate system for explaining the limited color determining method.

FIG. 6 is a diagram showing a histogram for explaining the limited color determination method.

FIG. 7 is a diagram showing a graph for explaining the limited color determination method.

FIG. 8 is a diagram showing a histogram for explaining the limited color determination method.

[Explanation of symbols]

 2 Lower 3 bit storage means 3 Periodicity calculation means 5 Color conversion means 7 Histogram calculation means 9 Average / dispersion calculation means 11 First spindle calculation means 13 Weighted dispersion calculation means 14 Divided threshold value determination means 16 Color area division means 17 Divided control Means 18 Emphasis coefficient designating means 19 Emphasis coefficient calculating means 21 Histogram updating means 22 Inverse color converting means 23 Representative color assigning means 41, 41a, 41b Pixels 43 Histogram showing frequency distribution 44 Graph showing weighting coefficient 45 Weighted corrected histogram

Claims (7)

[Claims] 1. A representative color is obtained by performing a discriminant analysis on a frequency distribution of each pixel value of color image data represented by a coordinate system of a uniform color space, and the image data is approximated by a limited representative color. A method for determining a limited color, wherein in the uniform color space, the class number in the coordinate axis indicating the luminance component and the class number in the coordinate axis indicating the chromaticity component are different, and the frequency in each coordinate axis of the data of each pixel value. A limited color determination method characterized in that a representative color is obtained by obtaining a distribution and performing a discriminant analysis on the frequency distribution. 2. When obtaining the frequency distribution in the uniform color space, the number of classes of the coordinate axis showing the luminance component is made larger than the number of classes of the coordinate axis showing the chromaticity component, and a representative color with enhanced luminance is selected. The limited color determination method according to claim 1, wherein 3. When obtaining the frequency distribution in the uniform color space, the number of classes of the coordinate axis showing the chromaticity component is made larger than the number of classes of the coordinate axis showing the luminance component to select a representative color with enhanced chromaticity. The limited color determination method according to claim 1, wherein 4. A representative color is obtained by performing a discriminant analysis on a frequency distribution of each pixel value of image data represented in the uniform color space, and the obtained representative color is again distributed in the uniform color space. Then, a smaller number of representative colors are obtained by performing a discriminant analysis on this frequency distribution, and when the representative colors are gradually reduced by repeating this, each step or any step of reducing the representative colors 4. The ratio between the class number on the coordinate axis indicating the luminance component and the class number on the coordinate axis indicating the chromaticity component is changed in order to obtain the frequency distribution with. Limited color determination method. 5. The lower bit of the digital data representing the values of the three primary colors of each pixel of the color image data is read and the periodicity in which the same value appears in the lower bit is obtained, while the color image data is stored in the uniform color space. For each pixel value represented by the coordinate system of, the frequency distribution on the coordinate axis indicating the luminance component and the coordinate axis indicating the chromaticity component in the uniform color space is obtained, and the frequency distribution is weighted based on the periodicity data. The frequency distribution is corrected, and the frequency distribution is corrected by making the class number on the coordinate axis indicating the luminance different from the class number on the coordinate axis indicating the chromaticity, and the discriminant analysis is performed. 1 or claim 2 or claim 3
Alternatively, the limited color determination method according to claim 4. 6. A representative color is obtained by performing a discriminant analysis on a frequency distribution of each pixel value of color image data represented by a coordinate system of a uniform color space, and the color image data is represented by a limited representative color. In a limited color determination apparatus to approximate, the number of classes of the coordinate axis showing the luminance component in the uniform color space and the number of classes of the coordinate axis showing the chromaticity component are different, of each pixel value of the color image data in each coordinate axis A limited color determining apparatus comprising: a frequency distribution calculating means for determining a frequency distribution; and a representative color determining means for performing a discriminant analysis on the obtained frequency distribution to determine a representative color. 7. A representative color is obtained by performing a discriminant analysis on a frequency distribution of each pixel value of color image data represented by a coordinate system of a uniform color space, and the color image data is represented by a limited representative color. An approximate limited color determination device, comprising: a periodicity calculation unit that obtains a periodicity coefficient with which the same value appears in lower bits for digital data representing the three primary colors of the color image data; and a luminance component in the uniform color space. And a frequency distribution calculating means for calculating a frequency portion of each pixel value of the color image data with respect to a coordinate axis indicating a chromaticity component, and the frequency distribution obtained by the frequency distribution calculating means is weighted and corrected by the periodicity coefficient. Frequency distribution correcting means, and the weighting-corrected frequency distribution is different in the number of classes of the coordinate axis showing the luminance component and the number of classes of the coordinate axis showing the chromaticity component. The frequency distribution is corrected as shall, discriminant analysis limited color determination apparatus characterized by comprising anda representative color determining means for determining a representative color by performing.