JPH05188899A - Method for selecting limited color for color image - Google Patents

Abstract

(57) [Abstract] [Purpose] A method of selecting a color corresponding to an LUT in a limited color selection method using a lookup table (LUT) is based on a fixed relationship between the arrangement relationship of the input pixel group and the block size. Provide a method to dynamically select a selection color. A pixel arranged in an RGB space is appropriately divided into block sizes, and a block to be divided is divided based on a fixed relationship between the number of pixels included in a selected block used in the LUT of the divided blocks and the number of blocks. The unit is changed corresponding to the input pixel group to select the limited color.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a limited color selection method by a look-up table method when a color image inputted is processed by a computer and outputted.

[0002]

2. Description of the Related Art In recent years, due to the rapid progress of technology, the integration density of memory capacity has been improved and the price has been reduced, as shown in FIG. There is a display device using a.
Use 8 bits for each RGB component,
At the time of reproduction, data for each RGB is D / A converted (80R, 8
0G, 80B) and sent to the display area 83 of the display. In this case, for example, the memory area used for the display area 83 of 400 lines at 640 pixels / line is 8 bits × 3 (RGB) × 640 (the number of pixels / line) ×
400 (lines) = 6, 144,000 bits (about 6 Mbits) are required.

On the other hand, the data area assigned to one pixel is limited to about 8 bits, and a limited color type of about 256 colors is arbitrarily selected from the color image to input the color image. Is also widely used (hereinafter referred to as a limited color selection method). With this limited color selection method, the memory area used for the display area 83 is 8 bits × 640 (number of pixels / line) × 400 (lines) = 2,048,000 (about 2 Mbits).

The limited color selection method will be described with reference to FIGS. 6, 7 and 8. FIG. 6 shows a palette number 44 of a look-up table (hereinafter, referred to as LUT) 42 separately created based on a data area 40 consisting of 8 bits assigned to one pixel existing in the input color image. Refer to the color number of each RGB data 46
Is extracted, D / A converted 48, and reproduced in the display area 83 of the display. The memory area occupied by the LUT 42 requires 8 bits × 3 (RGB) × 256 = 6, 144 bits.

Further, the display area 83 represents, for example, the total number of pixels 52 of the input color image of 640 pixels (H) per line and 400 lines (V), and FIG. , The display area 83 shown in FIG.
8 represents the total number of pixels of each of the RGB data in hexadecimal, and arranges each in a three-dimensional RGB space 60 with each of the values ranging from 00 to FF (Hex). , For each pixel 52 arranged in the RGB space 60 shown in FIG.
This shows a state in which the block is divided into a total of 256 block sizes up to the FF.

The conventional limited color selection method is shown in FIG.
As shown in FIG. 8, the total pixels of the input color image are set to RG.
Since the pixels arranged in the B space 60 and having a short distance in the RGB space 60 have similar colors, the RGB space 60
Each RGB data is equally divided into blocks of a certain size, the color of the center point of each divided block is determined as a representative color, and the pixel values existing in the block are replaced with the representative color values. However, it is possible to reproduce an image that does not differ much in mood compared to the input color image.

Therefore, it is possible to assign a color number having the same value as the representative color to pixels existing in the same block in the RGB space, and select 256 types of the representative color to extract the limited color. The types of limited colors are not limited to 256 types.

Then, a palette number 46 of 0 to 255 is given to the extracted limited color, and the palette number 46 and RG are assigned.
When the LUT 42 is created in association with each B data and the input color image is displayed in the display area 83, the color number 44 of each RGB data is selected from the replaced palette number 46 and the D / A conversion is performed. 48 and display (see FIG. 6).

FIG. 7 shows the R of each pixel 52 of the color image.
It shows an example of the distribution of each pixel 52 arranged in the RGB space 60 based on GB data. Therefore, RG
By setting the RGB space 60, which is a three-dimensional space having each primary color of B as the coordinate axis, each pixel 5 in the color image
2 is represented by one coordinate in the RGB space 60.

The color coordinate RGB (00,00,00) of each pixel 52 arranged in the RGB space 60 is "black" and is the farthest from this color coordinate RGB (00,00,00). Color coordinates RGB (F
F, FF, FF) can be defined as "white". For example, the center on the line drawn from the color coordinate RGB (FF, FF, FF) farthest from the color coordinate RGB (00,00,00). The dot color becomes "gray".

FIG. 8 shows a block name B suitable for a block obtained by dividing the RGB space 60 shown in FIG. 7 into equal sizes.
(X, X, X) is attached.

In the division of the RGB space of FIG.
For each block size, the gradation of R is Gr = (1 / Δr),
The G gradation is Gg = (1 / Δg), and the B gradation is Gb = (1
/ Δb), and the total number of blocks is Bn = Gr × Gg
Represented by × Gb. For example, the block name B (0, 0, 0) means a block including black pixels, and the number of pixels arranged in this block represents the distribution density of pixels. One pixel (representative color) having a central color is arbitrarily selected from each block. Even if a limited number of limited colors consisting of a group of representative colors is extracted from the representative colors of each block and each pixel existing in each block is replaced with this limited color, the number of limited colors is increased. By setting the above, it is possible to reproduce an image that does not differ much in mood compared to the input color image.

However, the types of selected colors used in the LUT 42 shown in FIG. 6 are limited to, for example, 256 types. Even if the RGB space is finely divided into a large number of blocks, the division is performed. There is a circumstance in which pixels included in all blocks cannot be selected as limited colors.

Therefore, the division of the RGB space 60 corresponding to the pixels 52 of the input color image 50 is constant as long as the LUT 42 has a limited color type, no matter how finely the block size is divided. Limited colors must be mechanically selected based on selection criteria. As described above, when the limited color determined by the mechanical selection is used, a well-balanced image cannot be reproduced as compared with the input color image.

Pixels which are not included in the extracted limited color are given the same color number 46 as the block having the extracted limited color at the closest distance, and the RGB space 60
Limited color number 46 for all pixels 52 present in
Can be given. When reproducing, the color number 46 of each RGB data is selected from the palette number 44 given based on the replaced color number 46, the color number 46 is D / A converted 48, and the display area 83 is displayed. Is displayed (see FIG. 6).

The image reproduced and produced by the limited color selection method in this way is inferior to the input color image 50, but the data is compressed and very convenient for data storage. Therefore, a relatively natural color image can be displayed even with a simple display method for reproducing the input color image 50 with limited colors, and therefore it is widely used at present.

The problem here is that the RGB space 60 is set to the LUT for the input color image 50.
Dividing into appropriate blocks corresponding to 42 limited color types (for example, 256 types) has a great influence on the quality of the displayed image.

[0018]

However, the limited color selection method using the LUT 42 described above also has drawbacks, and satisfactory results have not yet been obtained. That is,
When the pixels existing in the RGB space 60 are reproduced by dividing each RGB into equal block sizes, the limited color is extracted based on the block size evenly divided even though the density distribution of the pixels existing in the RGB space is not constant. Therefore, depending on the reproduced image, it is rough with respect to the input color image, or does not match the distribution density of pixels in the natural world, resulting in uneven display color. As a method for coping with such a problem, there is a method in which the reproduced image is visually observed and the block size of each RGB is changed appropriately, but the processing time is slow and it is not practical. Therefore, in the conventional example, there is a problem that the limited color selection that can reproduce a more balanced image compared to the input color image must be solved, while the color types adopted in the LUT 42 are limited. is doing.

The present invention has been made in view of these problems, and in order to reproduce an image that is extremely balanced as compared with the input color image 50, the pixels of the color image existing in the RGB space 60. It is an object of the present invention to provide a method for selecting a limited color by appropriately changing the block size to be divided from the relationship between the distribution density of the above and the total number of pixels included in all blocks selected as the limited color.

[0020]

In order to achieve the above object, the present invention provides a color image composed of a group of pixels with RGB
R consisting of a three-dimensional space in which each color has orthogonal coordinate axes
Arranging the pixels in the GB space based on RGB data,
In the method of dividing the RGB space into a plurality of blocks allocated at equal intervals in the RGB directions and selecting a plurality of limited colors from the representative colors of the pixels present in each block, the pixels present in each block When the number of blocks is on the Y-axis and the frequency range of pixels is on the X-axis based on the distribution density of, the gradient change point is determined from the relationship between the frequency range of pixels and the number of blocks, and all the limited colors having the selected limited color are obtained. A boundary line that is substantially parallel to the Y-axis is obtained based on the total number of pixels included in the block, and the RGB space is divided into blocks of appropriate sizes so that the boundary line is near the gradient change point including the gradient change point. A limited color selection method for a color image is characterized in that the limited color is selected.

[0021]

From the relationship between the gradient change point and the total number of pixels included in all blocks having limited colors, it is possible to automatically and appropriately divide the block size corresponding to the input color image. As for the image reproduced thereby, the most balanced image can be reproduced as compared with the input color image.

[0022]

Embodiments of the present invention will be described in detail below with reference to the drawings and tables. The limited color selection method in the present invention is
Arranging all the pixels 52 of the color image 50 in the RGB space 60 is the same as in the conventional example (see FIGS. 7 and 8).

Since the pixels 52 that are close to each other in the RGB space 60 have similar colors, the RGB space 6
It is divided into blocks of a size corresponding to the distribution density of the pixels 52 of the color image 50 arranged at 0. For example, the color of the center point of each block is set as the representative color, and the value of the pixels 52 included in the block is set to the block. By replacing the value with the representative color value of the block inside or closest to the pixel, an image having a different atmosphere can be reproduced compared to the input color image 50.

The following method is used for dividing into an appropriate block size according to the distribution density of the pixels 52 of the input color image 50 existing in the RGB space 60. (1) The RGB space 60 is divided with an arbitrary block size, and the distribution density of pixels included in the divided block is obtained.

(2) Next, the number of blocks included in the pixel frequency range is obtained from the obtained relationship between the distribution density of pixels and the number of blocks (see Table 2 described later).

(3) From the relationship between the pixel frequency range and the number of blocks included in the pixel frequency range (Table 2), when the pixel frequency range increases, the relationship between the pixel frequency range and the block number is almost the same. The number of blocks gradually decreases as the frequency range of pixels further increases, although the state changes to a certain state. At this time, a break point (hereinafter referred to as a gradient change point P) at the time of gradually decreasing from the constant transition state is obtained (see FIG. 1 described later).

(4) A representative color (256 kinds of colors in this embodiment) is selected as a limited color in the descending order of pixel distribution density of each block obtained in (1) above.

(5) Then, a boundary line between the total number of pixels included in all the blocks having the selected limited color and the total number of pixels included in the unselected block is determined.

(6) The gradient change point P obtained in (3) above and
The above-mentioned (1) to (5) are repeated until the boundary line obtained in (5) is at the same position or in the vicinity thereof, and the block size is dynamically changed to perform division. This dynamic change increases the block size when the boundary line is on the right side of the gradient change point P, and decreases the block size when the boundary line is on the left side of the gradient change point P.

(7) A palette number 44 of 0 to 255 is given to the limited colors obtained by the methods (1) to (6), and the LUT 42 is created by associating the palette number 44 with each RGB data. (See Figure 6).

(8) A palette number having the same value as the block is given to the pixels 52 included in all the blocks having the limited color (see FIG. 6).

(9) Pixels that are not included in the block having the limited color selected in (8) above have the same palette number 44 as the representative color of the block having the selected limited color located at the closest distance. Is given. Therefore, the palette number 44 can be given to all the pixels 52 existing in the RGB space 60.

(10) When reproducing, RGB data is called from the color number 46 based on the palette number 44 replaced as in the conventional example, and the RGB data is D / A converted 48 and displayed in the display area 83. Display (see FIG. 6).

Table 1 shows the RGB values obtained by the above method.
It shows a part of the relationship between a block divided into an appropriate block size for obtaining the limited color of the color image 50 existing in the space 60 and the distribution density of pixels existing in the block.

[0035]

[Table 1]

For example, the block name B (0,0,0) is "
The block name includes pixels of "black", and similarly, the block name B (0, 0, Gb-1) is "blue" and the block name B
(0, Gg-1, Gb-1) is a block name including "cyan", and the block name B (Gr-1, Gg-1, Gb-1) is a block name including "white". Hereinafter, since the same applies to the remaining other blocks (for all Bn total blocks shown in FIG. 8), description thereof will be omitted.

Here, the total number of pixel distribution densities is the total number of pixels on one screen. For example, the total number of pixels represented by the display area 83 shown in FIG. 6 is 640 (pixels / line).
× 400 lines = 256,000.

Table 2 shows the results of the limited color selection method described above.
The RGB space 60 is divided into arbitrary fixed sizes, the distribution density of pixels existing in the divided blocks is calculated, and the frequency range of the pixels and the pixel density range are calculated from the relationship between the calculated distribution density of the pixels and the number of blocks. The relationship with the number of blocks included in the frequency range is exponentially represented in ascending order of the pixel frequency range.

[0039]

[Table 2]

That is, the meaning of Table 2 is that, when the pixel frequency range is below a certain value (below 16 to 31 in Table 2), the number of blocks belonging to each frequency range is substantially constant. The number of blocks gradually decreases in the frequency range of pixels having a certain value or more, which can be clearly seen from an experimental example described later. That is, the gradient change point where the number of blocks starts to decrease is determined more easily from the logarithmic function relation between the logarithmic function relation of the two, with the horizontal axis representing the logarithm of the pixel frequency range and the vertical axis representing the logarithm of the block number. be able to.

FIG. 1 shows the frequency range of pixels and the number of blocks obtained on the basis of the data obtained in Table 2 in a logarithmic relationship, and shows the limited color selection method of the present invention.

The function f (x) is represented by the number of blocks (log
Y), the frequency range (logX) of pixels is plotted on the horizontal axis, and is expressed in a logarithmic relationship. That is, f (x) changes with the value of the same number of blocks (logX = logC) up to a certain frequency range (logX = (0 to logA)) with respect to the horizontal axis. The frequency range is a certain value or more (logX = (logA ~ log
B)), it gradually decreases with a constant gradient (logC / (logA-logB)). In this way, the relationship between the pixel frequency range and the block number is that the pixel frequency range has a constant value (logX
= LogA), the slope change point P at which it starts to decrease can be easily obtained.

Then, on this graph, the horizontal axis, logX =
A portion 5 (hatched portion) surrounded by logT and the function f (x) is a pixel included in all blocks having the selected limited color among the pixels 52 of the color image 50 existing in the RGB space 60. It represents the total number of 52. That is, by changing the block size (Δr, Δg, Δb), the graph of the function f (x) and the boundary line (logX = lo
gT) will change. The boundary line is located at a substantially boundary position between the selected block and the unselected block, and is in a state of being substantially parallel to the Y axis.

This graph of f (x) is an ideal relationship obtained by dividing the RGB space 60 so that the position of the boundary line (logX = logT) and the gradient change point P (logX = logA) are almost the same position. And that the image reproduced by dividing by the block size in this way replaces the pixels existing in the natural world with the limited color most efficiently, and can be reproduced with the best balance and beautifully as compared with the input color image 50. Was found from various experiments. It should be noted that the limit line indicates that the line is not crossed even when all images are assumed.

FIG. 2 shows the RGB space 6 as shown in FIG.
When 0 is divided evenly, the block size (△
FIG. 1 shows the relationship between the functions f1 (x) and f2 (x) when r × Δg × Δb) is small and when it is large, and the total number of pixels 52 included in all blocks having limited colors. Similarly, it is represented by a graph and substantially shows a comparative example with the present invention.

Function f1 when the block size is small
In the graph of (x), the number of blocks increases in the pixel frequency range, the slope change point P1 arrives at a value (logX = logA1) in which the pixel frequency range is relatively small, and the number of blocks decreases thereafter. Start (logA1 <logX ≦ logB1).

In this case, the number of blocks selected due to the memory capacity (256 palettes in LUT 42)
Is limited, and when the block size is small, the number of selected blocks is smaller than the total number of blocks (Bn), so the boundary line (logX = logT1) is on the right side of the gradient change point P1 ( (See part 10 of the area indicated by the right diagonal line)

When a color image divided by the block size represented by the function f1 (x) in the case where the block size is small is reproduced, a part with light and shade is expressed beautifully, but many colors are present there. Is used too much,
The areas where the colors are greatly different from the color image become noticeable.

The graph of the function f2 (x) when the block size is large has a smaller number of blocks corresponding to the frequency range of pixels than the graph of the function f1 (x) when the block size is small. The slope change point P2 arrives at a large value (logX = logA2) in the pixel frequency range, and the number of blocks starts to decrease (logA2 <logX ≦ log
B2).

In this case, since the number of selected blocks is larger than the total number of blocks (Bn), the boundary line (logX
= LogT2) is on the left side of the gradient change point P2 (see the area 20 shown by the left diagonal line).

Since the number of selected blocks is larger than the total number of blocks (Bn), the total number of pixels in the selected blocks is also large for all pixels, and this means seems good at first glance. However, since the size of each block becomes large, even the pixels in the selected block have a long distance from the limited color (the amount of correction needs to be increased), which is different from the input image in the end. It becomes a thing.

Therefore, when the input color image is reproduced based on the data divided by the block size represented by the function f2 (x) when the block size is large, there is little difference from the input color image 50. However, a striped pattern like the contour lines of a map is formed in the part where the shading is gentle.

Next, three types of experimental examples for color images 50 of different subjects will be described. Tables 3 and 4 and Table 5
Table 6 and Table 7 and Table 8 are divided into three block sizes (corresponding to FIG. 1 and FIG. 2), and each represents experimental data of three types of color images 50 of subjects. is there.

Then, based on the experimental data shown in this table, by the method shown in FIGS. 1 and 2, the frequency range of pixels and the number of blocks are expressed in a logarithmic relationship to obtain the gradient change point, and the limited color is determined. Boundary lines are obtained from the total number of pixels included in the included blocks in FIGS. 3, 4, and 5.

Table 3 and Table 4 and FIG. 3, Table 5 and Table 6 and FIG. 4, Table 7 and Table 8 and FIG. 5 correspond to each other, and 480 (pixels / line) × 400 (line) per screen. ) = 19
The experiment is conducted with the total number of pixels of 2,000.

[0056]

[Table 3]

Table 3 is an experimental example 1 in which the input color image 50 is a still image containing a person, and is divided into three block sizes corresponding to FIGS. 1 and 2. is there. That is, if it is a conventional example, Experimental example 1 (original image)
As shown in, the R block size is divided into 64 gradations, the G block size is divided into 64 gradations, and the B block size is divided into 64 gradations, and the number of colors is required to be about 260,000.

On the other hand, Experimental Example 1-A (f1 in FIG. 2)
(Corresponding to (x)) divides the RGB space into 16 gradations of R block size, 32 gradations of G block size, and 16 gradations of B block size, and limits the number of limited colors to 256. It is a thing.

Experimental Example 1-B (corresponding to f (x) in FIG. 1)
Shows that the RGB space is divided into R block size of 16 gradations, G block size of 16 gradations, and B block size of 16 gradations so that the number of limited colors is 256.

Experimental Example 1-C (corresponding to f2 in FIG. 2) is R
The number of limited colors is limited to 256 by dividing the RGB space with a block size of 16 gradations, a G block size of 8 gradations, and a B block size of 16 gradations.
It should be noted that only the G block size of the RGB block size is changed because the change of green (G) is other red (R) or blue (B).
This is because it has the greatest effect on human vision compared to.

Table 4 shows data obtained by experiments on the frequency range of pixels included in each block and the relationship between the frequency range of the pixels and the number of blocks when divided by the block size shown in Table 3. It is a thing.

[0062]

[Table 4]

That is, in the experimental example 1-A, the number of blocks is substantially the same in the pixel frequency range from "0 to (21-1)" to "27 to (28-1)", and "28 to (29-
It is clear from 1) that the number of blocks gradually decreases.

In Experimental Example 1-B, the pixel frequency range is "0".
From ~ (21-1) "to" 26 ~ (27-1) ", the number of blocks is almost the same, and it is clear that the number of blocks gradually decreases from" 27 ~ (28-1) ". ..

In Experimental Example 1-C, the pixel frequency range is "0".
From ~ (21-1) "to" 28 ~ (29-1) ", the number of blocks is almost the same, and it is clear that the number of blocks gradually decreases from" 29 ~ (210-1) ". ..

FIG. 3 shows the input color image in FIG. 1 and FIG. 2 based on the data (see Tables 3 and 4) obtained in Experimental Example 1 in which a still image containing a person is the subject. This is a graph showing the relationship between the frequency range of pixels and the number of blocks by the above method.

Experimental example 1-A is marked with a circle, and experimental example 1-B is marked with a triangle.
The symbol □ indicates experimental example 1-C, the logarithm of the pixel frequency range is on the horizontal axis, the logarithm of the number of blocks is on the vertical axis, and the approximate values of the functional relationship between the two are connected by a line to form a graph. Is.

The broken line represents a boundary line (logX = logA, logX = logB, logX = logC) which is one side of the area of the total number of pixels included in all the blocks having the selected limited color. There is. Therefore, this boundary line (logX = logA, logX =
The area surrounded by the right side of (logB, logX = logC) is the total number of pixels included in the block having the selected limited color.

Here, the X mark indicates the slope change point P (PA, PB,
PC). That is, in Experimental Example 1-A, the boundary line (logX
= LogA), and in the experimental example 1-B, the gradient change point P-B
The boundary line (logX = logB) is located at substantially the same position as the above, and in Experimental Example 1-C, the boundary line (logX = logC) is located on the left side of the gradient change point P-C.

Therefore, when reproduced with the selected colors obtained in these three different experimental examples 1, the experimental examples 1-A and 1-C have a striped pattern or a large difference in color from the original image. To do. However, in Experimental Example 1-B, it is possible to reproduce a well-balanced image as compared with the input color image even if the limited color is reproduced.

Table 5 shows an experiment example 2 in which the target of the input color image is not a person but a still image of fruit or the like placed in front of the person as an object. The table is obtained by dividing into three block sizes. That is, in the conventional example, as shown in Experimental Example 2 (original image), the R block size is divided into 64 gradations, the G block size is divided into 64 gradations, and the B block size is divided into 64 gradations. About 260,000 colors are needed.

[0072]

[Table 5]

On the other hand, Experimental Example 2-A (f1 in FIG. 2)
(Corresponding to (x)), the R block size is divided into 16 gradations, the G block size is divided into 32 gradations, and the B block size is divided into 16 gradations, and the number of limited colors is limited to 256. It is a thing.

Experimental Example 2-B (corresponding to f (x) in FIG. 1)
Is an R block size divided into 16 gradations, a G block size divided into 16 gradations, and a B block size divided into 16 gradations so that the RGB space is divided into 256 limited colors.

Experimental Example 2-C (corresponding to f2 (x) in FIG. 2)
The R block size is 16 gradations, the G block size is 8 patterns, and the B block size is 16 gradations, so that the RGB space is divided and the number of limited colors is limited to 256 colors. It should be noted that changing only the G block size out of the RGB block sizes has the effect that the change in green (G) has an effect on human vision as compared with other red (R) and blue (B) as in Experimental Example 1. Because it is the largest.

Table 6 shows data obtained by experiments on the frequency range of the pixels included in each block and the relationship between the frequency range of the pixels and the number of blocks when divided by the block size shown in Table 5. It is a representation.

[0077]

[Table 6]

In Experimental Example 2-A, the pixel frequency range is "0".
From ~ (21-1) "to" 26 ~ (27-1) ", the number of blocks is almost the same, and it is clear that the number of blocks gradually decreases from" 27 ~ (28-1) ". ..

In Experimental Example 2-B, the pixel frequency range is "0".
From ~ (21-1) "to" 29 ~ (210-1) ", the number of blocks is almost the same, and it is clear that the number of blocks gradually decreases from" 210 ~ (211-1) ". ..

In Experimental Example 2-C, the pixel frequency range is "0".
From ~ (21-1) "to" 29 ~ (210-1) ", the number of blocks is almost the same, and it is clear that the number of blocks gradually decreases from" 210 ~ (211-1) ". ..

FIG. 4 shows a color image to be input as the first experimental example.
Based on the data (see Tables 5 and 6) obtained in Experimental Example 2 in which the target of the color image is not centered on the person like this, but a still image of a fruit or the like placed in front of the person is the subject, as shown in FIG. 3 is a graph showing the relationship between the frequency range of pixels and the number of blocks by the method shown in FIG.

Experimental example 2 is the same as experimental example 1 shown in FIG.
-A is indicated by a circle, Experiment 2-B is indicated by a triangle, and Experiment 2-C is indicated by a square. The horizontal axis represents the logarithm of the pixel frequency range, and the vertical axis represents the logarithm of the number of blocks. It is a graph in which the approximate values of the relationship are connected by a line.

The broken line represents a boundary line (logX = logA, logX = logB, logX = logC) which is one side of the area of the total number of pixels included in the block having the selected limited color. is there. Therefore, this boundary line (logX = logA, logX = log
The area surrounded by the right side of (B, logX = logC) is the area of the total number of pixels included in the block having the selected limited color.

Here, the X mark indicates the gradient change point P (PA, PB,
PC). That is, in Experimental Example 2-A, the boundary line (logX
= logA), and in Experimental Example 2-B, the slope change point P-B
The boundary line (logX = logB) is located at almost the same position as the above, and in Experimental Example 2-C, the boundary line (logX = logC) is located on the left side of the gradient change point P-C. Therefore, these 3
When reproduced with the selected color obtained in the above experimental example, Experimental Examples 2-A and 2-C have a striped pattern or a large difference in color from the original image. However, in Experimental Example 2-B, it is possible to reproduce a well-balanced image compared to the input color image even if the limited color is reproduced.

Table 7 shows Experimental Example 3 in which the target of the input color image is a subject centering on three stationary dolls, which are arbitrarily divided into three block sizes. It is a representation. That is, in the conventional example, as shown in Experimental Example 3 (original image), the R block size is divided into 64 gradations, the G block size is divided into 64 gradations, and the B block size is divided into 64 gradations. About 26000 colors are needed.

[0086]

[Table 7]

On the other hand, Experimental Example 3-A (f1 in FIG. 2)
(Corresponding to (x)) is one in which the RGB space is divided into 16 gradations of R block size, 32 gradations of G block size, and 16 gradations of B block size, and the number of limited colors is limited to 256 colors. Is.

Experimental Example 3-B (corresponding to f (x) in FIG. 1)
Is an R block size divided into 16 gradations, a G block size divided into 16 gradations, and a B block size divided into 16 gradations so that the RGB space is divided into 256 limited colors.

Experimental Example 3-C (corresponding to f2 (x) in FIG. 2)
The R block size is 16 gradations, the G block size is 8 patterns, and the B block size is 16 gradations, so that the RGB space is divided and the number of limited colors is limited to 256 colors. It should be noted that only the G block size of the RGB block size is changed because the change of green (G) is similar to that of the other red (R) and blue (B) as in Experiments 1 and 2 and is more human vision. This is because it has the greatest effect.

Table 8 is obtained by an experiment of the frequency range of pixels included in the block having the limited color and the number of blocks corresponding to the frequency range of the pixel when divided by the block size shown in Table 7. This is a table showing the obtained data.

[0091]

[Table 8]

In Experimental Example 3-A, the pixel frequency range is from "0".
From (21-1) "to" 27- (28-1) ", the number of blocks is almost the same, and it is clear that the number of blocks gradually decreases from" 28- (29-1) ".

In Experimental Example 3-B, the pixel frequency range is from "0".
From (21-1) "to" 28 to (29-1) ", the number of blocks is almost the same, and it is clear that the number of blocks gradually decreases from" 29 to (210-1) ".

In Experimental Example 3-C, the pixel frequency range is from "0".
From (21-1) "to" 28 to (29-1) ", the number of blocks is almost the same, and it is clear that the number of blocks gradually decreases from" 29 to (210-1) ".

FIG. 5 shows data obtained in Experimental Example 3 in which the input color image is a subject whose main object is not a person but a three dolls (see Tables 7 and 8). 3 is a graph showing the relationship between the frequency range of pixels and the number of blocks by the method shown in FIGS.

Similar to Experimental Example 1 shown in FIG. 3, Experimental Example 3
-A is indicated by a circle, Experimental Example 3-B is indicated by a triangle, and Experimental Example 3-C is indicated by a square, and the logarithm of the pixel frequency range is plotted on the horizontal axis and the logarithm of the block number is plotted on the vertical axis. It is a graph in which the approximate values of the relationship are connected by a line.

Further, the broken line is a boundary line (logX = logA, l) of the total number of pixels included in the block having the selected limited color.
ogX = logB, logX = logC). Therefore, this boundary line (l
The area enclosed by the right side of (ogX = logA, logX = logB, logX = logC) is the area of the total number of pixels included in the block having the selected limited color.

Here, the X mark indicates the gradient change point P (PA, PB,
PC). That is, in Experimental Example 3-A, the boundary line (logX
= LogA), and in Experimental Example 3-B, the slope change point P-B
The boundary line (logX = logB) is located at substantially the same position as the above, and in Experimental Example 3-C, the boundary line (logX = logC) is located on the left side of the gradient change point P-C. Therefore, these 3
When reproduced with the selected color obtained in the same experimental example, Experimental Examples 3-A and 3-C have a striped pattern or a large color difference from the original image. However, in Experimental Example 3-B, even if a limited color is reproduced, a well-balanced image can be reproduced as compared with the input color image.

As can be seen from the above experimental example, three kinds of target color images are selected, and from each of the three types of experimental data for each color image, a gradient change point P and a block having the selected limited color are selected. If the block size (ΔrΔgΔb) is divided so that the boundary line (logX = logT) obtained from the total number of included pixels is almost at the same position, the input color image 50 is reproduced with limited colors. Even with this, a well-balanced image can be obtained.

As described above, R, G, B of each input pixel
Value f and the block size (ΔrΔgΔb)
(X) is calculated and the gradient change point P point (logX = P)
And T point (logX = logT) are obtained. Therefore, by changing the block size (ΔrΔgΔb), the optimum condition (P = lo
The block size of gT) can be obtained by arithmetic processing with a computer.

[0101]

According to the limited color selection method of the present invention,
The pixels of the input color image are converted into RGB that consists of a three-dimensional space.
Arranged in space, select a fixed number of limited colors from the order of the distribution density of pixels contained in the block divided by an appropriate size, and determine the gradient change point from the block number and the frequency range of pixels, The boundary line is obtained from the total number of pixels included in the block having the limited color, and the block size is appropriately divided so that the boundary line is almost at the same position as the gradient change point, and the representative color of the selected block is limited. An image having a good balance with respect to the input color image can be obtained by using the color and reproducing with the limited color. Moreover, the method of the present invention can be automatically performed by software when it is processed by a computer.

[Brief description of drawings]

FIG. 1 shows a theoretical value of limited color selection corresponding to the method of the present invention in a logarithmic relationship between a pixel frequency range and the number of blocks.

FIG. 2 shows a frequency range of pixels and the number of blocks in a logarithmic relationship according to the size of a block size for comparison with the method of the present invention.

FIG. 3 is an experimental example 1 in which the frequency range of pixels and the number of blocks are represented by a logarithmic relationship in the method according to the present invention and in which the block size is changed for comparison.

FIG. 4 is an experimental example 2 in which the frequency range of pixels and the number of blocks are represented by a logarithmic relationship in the method according to the present invention and in which the block sizes are changed for comparison.

FIG. 5 is an experimental example 3 in which the frequency range of pixels and the number of blocks are represented by a logarithmic relationship in the method according to the present invention and in which the block sizes are changed for comparison.

FIG. 6 is a schematic diagram of a limited color reproduction method using a generally known look-up table (LUT).

FIG. 7 is a schematic diagram showing an example of the total number of pixels arranged in an RGB space obtained from a generally known display area.

FIG. 8 is a schematic diagram in which pixels arranged in the RGB space shown in FIG. 7 are equally divided into appropriate block sizes.

FIG. 9 is a schematic view showing a theory for displaying a color image in a display area by a conventional method.

[Explanation of symbols]

 5 Total number of pixels included in the selected block 10 Total number of pixels included in the selected block 20 Total number of pixels included in the selected block 40 Data area 42 Look-up table (LUT) 44 Palette Number 46 Color number 48 D / A conversion 52 Pixels arranged in RGB space 60 RGB space 83 Display area

[Table 2]

[Table 2]

[Table 4]

[Table 4]

[Table 6]

[Table 6]

[Table 8]

[Table 8]

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Internal reference number FI technical display location G09G 5/36 9177-5G

Claims (1)

[Claims] 1. A color image composed of a group of pixels
B The pixels are arranged on the basis of RGB data in an RGB space formed of a three-dimensional space in which each color is a coordinate axis orthogonal to each other, and the RGB space is divided into a plurality of blocks allocated at equal intervals in the RGB direction. In the method of selecting a plurality of limited colors from the representative color of the pixels present in the block, based on the distribution density of the pixels present in each block,
When the number of blocks is on the Y-axis and the frequency range of pixels is on the X-axis, the gradient change point is obtained from the relationship between the frequency range of pixels and the number of blocks, and the pixels included in all the blocks having the selected limited color are included. A boundary line that is substantially parallel to the Y-axis is obtained by the total number of the, and the RGB space is divided into blocks of appropriate sizes so that the boundary line is in the vicinity of the gradient change point including the gradient change point, and the limited color is selected. A limited color selection method for a color image, comprising: