JPH06337926A - Method for optimizing display limited color and limited color display device - Google Patents

Abstract

PURPOSE:To prevent a false contour and dropping of a small area of color at the time of displaying a color picture with limited colors. CONSTITUTION:A histogram preparing means 3 prepares histograms H (m1, m2, m3) of picture data and a distance calculating means 5 finds out a distance (d) from each point in a color space up to the nearest point in a color pallet based on the histograms H (m1, m2, m3) and an initial representative point <C<0>n> prepared by an initial representative point preparing means 4. A minimum distance representative point selecting means 6 selects a distance from each point in the color space up to the nearest point of the color pallet, an adding means 7 adds the selected value to the preceding color pallet and an evaluating means 8 evaluates the added value to fined out a color pallet <C<k>n> for furthermore reducing the sum J (<Cn>) of products between the added value and the histograms H of the color picture.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for optimizing a display limited color and a limited color display device for displaying a color image in a limited color in a computer, a CD graphic or the like.

[0002]

2. Description of the Related Art Generally, for example, when a digital color image of 8 bits each for RGB is displayed in 2 ²⁴ full colors, a memory of 24 bits is required for each pixel. A method of displaying is known. Further, in order to display a color image on a computer having a display capability of about 256 colors, it is necessary to optimize the limited color to be displayed.

In this type of limited color display method, the color distribution area is simply divided into representative color numbers without considering the statistical quantity of the color distribution of the original image, and the median value of each area is defined as the representative color. Uniform quantization method, which is described later, or a subdivided quantization method (tapered quantization) that optimizes color packets based on statistics of the color distribution of the original image as described later.
method) or the nearest neighbor color search method that selects the closest color (in Euclidean distance in color space) of each pixel of the original image when the representative color is determined by some method,
Various proposals have been made as the subdivision quantization method.

As the subdivision quantization method, a frequency method in which colors having the highest appearance frequency are sequentially selected as representative colors, and R
Consider a color space linear division method that linearly divides only a partial area where the color of the original image is distributed in the GB color space to determine a representative color, and a rectangular parallelepiped that circumscribes the color distribution in the RGB space of the original image. , The RGB three axes are divided into two so that the total number of pixels of the original image is equally divided by the axis parallel to the longest side of the rectangular parallelepiped, and the rectangular parallelepiped is gradually subdivided by repeating this, and finally from the same number of pixels. The number of pixels included in each pixel in the process of generating N rectangular parallelepipeds and using the median value division method in which the average value of the color data of each rectangular parallelepiped is the representative color, and the process of gradually expanding the partitions that partition the RGB space. Population equalization that selects representative colors for each section so that the
Method, an adaptive subspace method in which the RGB space is partitioned into spherical subspaces according to the color distribution and a representative color is selected for each subspace, and a viewpoint in which the representative colors are arranged as far apart from each other as possible in the RGB space The cubic close-packing method, which arranges the representative colors in a cubic close-packed manner, and the Peano curve are used in a single stroke to perform a Peano scan to make the distribution of the three-dimensional space one-dimensional, and divide this into equal frequency divisions. There is known a method of determining a representative color by doing.

Further, as shown in, for example, Japanese Patent Laid-Open No. 4-190466, the upper 5 of each 8-bit RGB data
The color space is constructed using bits, the color space is divided into a limited number of areas based on the color distribution, and the representative color of each area is selected. , Calculating a three-dimensional Euclidean distance from a color similar to the representative color, assigning the representative color with the shortest distance to the original image, and
A method of making one-dimensional by using a Peano curve has been proposed.

Further, as another method, Japanese Patent Laid-Open No. 4-152
As shown in Japanese Patent No. 386, the representative color of each region is selected by dividing the color space into a limited number of regions based on the color distribution, and the color of each original image and the representative color of the region to which the original pixel belongs. When,
A method has been proposed in which a three-dimensional Euclidean distance is calculated between the representative colors of two areas adjacent to this area and the representative color having the shortest distance is assigned to the original image.

Another method is, for example, Japanese Patent Laid-Open No.
Sample points are distributed in a hexagonal crystal structure that includes the entire color space in order to reduce the maximum distance between any point in the color space and the closest color sample point as shown in US Pat. Is proposed.

[0008]

SUMMARY OF THE INVENTION Here, the above-mentioned Japanese Patent Laid-Open No.
In the method disclosed in Japanese Patent Publication No. 37696, since the sample points are arranged at the vertices of a regular hexagon, even if the same number of sample points is used, the space in the space is smaller than that in the case where the sample points are evenly arranged in a grid pattern. The distance to any point can be reduced. However, the method of taking the sample points, that is, the method of taking the color palette is invariant for any image data, so that in such a color palette, the color gradually changes like an empty picture, for example. There is a problem in that it is unavoidable that pseudo contours occur in an image.

The above "frequency method", "median division method", "population equalization method",
"Adaptive subspace method", "method by three-dimensional close packing",
The method using the Peano curve may be able to solve the problem of pseudo contours, but for example, in a picture in which red flowers are small while there is green on one side, the amount of data is If all the sample points are taken in the vicinity of "green", which is very common, it will not be possible to assign an appropriate color to "red", so it will be difficult to properly display the "red" of flowers. There is a problem.

In view of the above-mentioned conventional problems, an object of the present invention is to provide a method of optimizing a display-limited color and a limited-color display device which can prevent a false contour and a color loss of a small area.

[0011]

In order to achieve the above object, the present invention reduces the sum of the product of the distance from each point in the color space to the closest point in the color palette and the histogram of the color image. By doing so, I seek a color palette. That is, according to the present invention, the color palette is obtained by further reducing the sum of the product of the distance from each point in the color space to the closest point of the color palette and the histogram of the color image. A color optimization method is provided.

Further, according to the present invention, a histogram creating means for creating a histogram of a color image, and the color palette from each point in the color space based on the image data of the color image and the histogram created by the histogram creating means. Distance calculating means for calculating the distance to the closest point of, and means for obtaining a color palette that further reduces the sum of the products of the distance calculated by the distance calculating means and the histogram created by the histogram creating means. A limited color display device having the same is provided.

[0013]

According to the present invention, a color palette that reduces the sum of the product of the distance from each point in the color space to the closest point in the color palette and the histogram of the color image is calculated. Is optimized according to the image data, and the weight of the small area formed by the pixel values distant from the sample point is increased by the product operation, so that it is possible to prevent the false contour and the loss of the color of the small area. .

[0014]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a method for optimizing a display limited color and a limited color display device according to the present invention, FIG. 2 is an explanatory diagram showing arrangement positions of sample points by the method and device of FIG. 1, and FIG. FIG. 7 is a block diagram showing a modified example of the device of FIG. 1.

First, the terms in the present invention will be explained. First, the image in the present invention is a function from each coordinate in the coordinate space of a two-dimensional integer N to a pixel value. Further, each pixel value is usually a one-dimensional (monochrome image) or three-dimensional (color image) vector value, and the image P
Is

[0016]

[Equation 1] P: N ² → N (monochrome image) P: N ² → N ³ (color image)

It is represented by The pixel value at the point of A = (x, y) is represented by P _A or P (x, y), and P _A , P
(X, y) is also simply referred to as a pixel. Furthermore, in the case of a three-dimensional vector value, RGB (red, green, blue), YMC (yellow, magenta, cyan), HSI
Physical quantities such as (hue, saturation, lightness) can be considered,
A space composed of such three-dimensional values is called a color space.
Further, the value of the first coordinate axis of the point A in this three-dimensional color space is represented by P _A. X or P (x, y). X, and similarly, the second,
The value of the third coordinate axis is P _A. Y or P (x, y).
Represented by Y, _PA.Z or P (x, y) .Z. Therefore, in such a color space we can introduce a distance d, which is commonly expressed in mathematics, which for any point Q, R, S in space.

[0018]

## EQU2 ## d (Q, R) ≧ 0 d (Q, R) = d (R, Q) d (Q, R) = 0 ⇔ Q = R d (Q, R) ≦ d (Q, S) + D (S, R)

It is a positive real-valued function d that satisfies: In addition,
As the function d, the Euclidean distance and the city block distance are widely known, and it is also possible to introduce them for coordinates such as hue data that circulate.
Further, in this embodiment, it represents the frequency distribution in color space of the image P a (histogram) in H _P. Further, the frequency of the pixel value p in the image P is represented by _HP (p), and in particular p = (x,
y, expressed when the _{z) H P (x, y} , z).

Here, each coordinate axis of the three-dimensional color space is 8
In the case of bit precision, that is, 256 gradations, the image display device must have a display capacity of 2 ⁸ × 2 ⁸ × 2 ⁸ = about 16.7 million colors as described above. A computer capable of displaying is expensive, and 24-bit data is required for each pixel, resulting in a large amount of data. Therefore, at present, most of the pixels have a display capability of 8 bits in total, that is, about 256 colors, and in this case, pixel values numbered 0 to 255 are displayed. Therefore, in this embodiment, the pixel value columns C1, C2 ... CN are represented by <C _i | i = 0, 1 ... N>, and if it is not necessary to pay particular attention to the number, simply It is represented by C _i >.

In this embodiment using such a definition, a color palette suitable for each image is created (that is, image P
<C _i >, which is the most suitable for the color space), and one of <C _i > assigned to each pixel value in the color space is determined. Thus, each point of the color space can be regarded divided into a set number N of <C _i>, the <C _i> is a representative point of each set.

Next, a first embodiment of the present invention will be described. This
In this embodiment, the histogram of the image P is set to H (m1, m
2, m3) (however, m1 = 0,1 ... M1, m2 = 0,1
... M2, m3 = 0,1 ... M3) and display this image
<C to create a color palette n | i = 0, 1 ... N>
Then, J (<C given by the following equation (1)_n>)
Think of this, J (<C_n>) To minimize <C_n＞
Color palette <C_n> Is created.

[0023]

[Equation 3]

Equation (1) represents the amount of error in the entire image P when each pixel value is replaced by the closest distance value in <C _n >. From another viewpoint, this is Image P
Is the sum of the sampling noises when resampled.

In the case where the method of taking sample points, that is, the method of taking a color palette is invariant with respect to any image data as in Japanese Patent Laid-Open No. 3-37696, such a color palette is used to, for example, display an empty picture. It is unavoidable that a pseudo contour occurs in an image in which the color gradually changes as described above, but in the present embodiment, as shown in FIG.
Since many sample points are arranged in a place where the image data values are concentrated (green in the figure), the pseudo contour can be prevented.

In the equation (1), the distance d is multiplied by the histogram Hp so as to increase the specific gravity of the distant points (the area of the hatched portion shown in FIG. 2), and the number of distant points is also calculated. Since data will not be leaked, for example, a "red" flower can be displayed even in a picture in which a red flower is small while a green color is on one side,
Therefore, it is possible to prevent color loss in a small area. Further, “Σ” in Expression (1) means that many sample points are assigned to positions where much data exist.

[0027] Here, J shown in Equation _{(1) (<C n>} ) to minimize the simplest method of obtaining the <C _n>, all <C _n> J for the combination of (<C _n >) and select the smallest one, but if each coordinate axis of the color space can take 8 bits, that is, 256 values, <C _n
＞ combination is

[0028]

[Equation 4] _{256 * 256 * 256} C _N (2)

It becomes as follows. Therefore, this value is N = 2
Of (about 16.7 million) ^2/2 at the time, it is not practical to repeat the calculation of equation (1) this number of times. Here, when the initial representative point sequence <C ⁰ _n > is given in some form, the next representative point sequence <C ¹ _n > is obtained based on this, and similar operations are repeated to finally obtain J ( It is necessary to find <C _n > that minimizes <C _n >). The <C _n > obtained by this may not necessarily minimize the value of the equation (1), but it can be obtained as a better color palette, and this procedure will be described below. First,

[0030]

<C ⁰ _n >, <C ¹ _n >, ..., <C ^K _n > ... (3)

Sequential determination of means that for each C ^k _n

[0032]

ΔC ^k _n = C ^{k + 1} _n −C ^k _n (4)

It is nothing but the calculation of C ^k _n . This C ^k _n can be obtained by using the gradient from each state like the steepest descent method or the Newton method, and the gradient <d ^k _n is shown.
> Is

[0034]

[Equation 7]

Is given by Δ using this d ^k _n
While determining the C ^k _n, the d ^k _n αΔC ^k _n, especially the steepest descent method. The limited color display device of this embodiment will be described with reference to FIG. 1. First, a color image is read via an image input means 1 such as an image scanner and then stored in a buffer 2. Next, the histogram creation means 3 creates a histogram H (m1, m2, m3) of this image data, and the distance calculation means 5 creates this histogram H (m
1, m2, m3) and the initial representative point <C ⁰ _n > created by the initial representative point creating means 4, the distance d from each point in the color space to the closest point in the color palette is obtained.

Next, the distance minimum representative point selecting means 6 selects the distance from each point in the color space to the closest point in the color palette in the equation (1), and the adding means 7 adds the distance to the previous color palette and adds it. A color palette <C ^k whose value is evaluated by the evaluation means 8 to reduce the sum J (<C _n >) of the products of the color image and the histogram.
_n > (k = 1) is obtained.

Further, the gradient amount calculating means 9 obtains the gradient amount <d ^k _n > shown in the equation (5), and then the representative point updating means 10 converts the gradient amount <d ^k _n > into the equation (4). The next color palette C ^{k + 1} _{n shown} is determined and applied to the distance calculation means 5 for the next calculation. Therefore, by repeating this process, the sum of the product of the distance from each point in the color space to the closest point in the color palette and the histogram of the color image is made smaller, and the false contour and the color loss of the small area are lost. A color palette that can prevent

Here, in order to obtain the gradient amount d ^k _n , whether or not it is differentiable is a problem. However, each term (obviously differentiable) when the formula (1) is expanded is simply differentiated. For example, d ^k _n can be obtained, but it is not necessarily differentiable depending on mini in the equation (1), and therefore, there is a possibility of impairing the stability of convergence. To solve this problem, a positive integer δ and the set I = {− 1,0,1}, φ ^k _n ,
Using ζ ^k _n and ξ ^k _n εI, the variation ΔC ^k _n is

[0039]

[Equation 8]

Give

[0041]

## EQU7 ## δ (φ ^k _n , ζ ^k _n , ξ ^k _n ) (7)

Is determined as This method increases the calculation amount per time as compared with the case of using the equation (5),
Convergence stability can be guaranteed.

FIG. 3 is a block diagram showing the limited color display device in this case. Equation (6) is calculated by the next representative point candidate creating means 12, and the variation ΔC ^k _n is taken as the value shown in equation (7). The next representative point candidate is determined and applied to the distance calculating means 5 based on the initial representative point created by the initial representative point creating means 4. Further, as in the case shown in FIG. 1, the distance calculation means 5 obtains the histogram H (m1, m2, m3) and the distance d from each point in the next representative point candidate color space to the closest point in the color palette.

Then, the minimum distance representative point selecting means 6 is used.
In equation (1), the color palette is calculated from each point in the color space.
The distance to the closest point of the
Is added to the previous color palette and based on the added value
The next representative point is selected by the next representative point selecting means 11,
The next representative point is evaluated by the evaluation means 8 and the color image
Sum of products with histogram J (<C_n>) Smaller
Color palette <C^k n> is required. Also, allocation
This representative point sequence is assigned to the original image data by the clipping means 13.
Be kicked. The other configurations 1 to 3 are the same as those shown in FIG.
Since it is one, detailed description thereof will be omitted.

Next, a second embodiment of the present invention will be described. First, considering the calculation amount of the first embodiment, the calculation of the formula (1) is roughly calculated for each iteration.

[0046]

[Equation 10] N × 3 ³ times (8)

It will be done. Here, since the point of <C _n > closest to each point on the space must be obtained in the calculation of Expression (1), the calculation amount of Expression (1) is estimated as follows.

[0048]

[Equation 11] N × M1 × M2 × M3 times (9)

It is necessary to calculate the distance d of. Therefore, each coordinate axis of the color space has 4 bits, that is, M1 = M2.
= M3 = 16, equation (9) becomes

[0050]

[Equation 12] 4096N times (10)

It becomes Also, 8 bits, that is, M1 = M
When 2 = M3 = 256, the equation (9) is

[0052]

[Equation 13] Approximately 16.7 million times N times (11)

It becomes Therefore, apart from equation (10), it is difficult to actually calculate the number of times represented by equation (11) unless it is a very expensive computer or a high-performance dedicated device. This problem is due to the fact that the number of times the distance d is calculated is proportional to the cube of the number of gradations on each coordinate axis in the color space.

Therefore, in the second embodiment, the value of the equation (1) is reduced by the number of times the distance is calculated, which is not proportional to the third power but is proportional to the first power. First, N ≧ N1 × N2 ×
Given N1, N2, and N3 such that N3, N1 × N2
Consider creating × N3 color palettes. If the image display capability is 8 bits (256 colors), 7
X7x5 = 245, 7x6x6 = 252, 17x5x3
= 255 is suitable.

At this time, <C _n > is set to <0 ≦ D _n ≦ M1 | n
= 1, 2 ... N1>, and d1 is a distance function on the first axis of the color space

[0056]

[Equation 14]

Consider Here, H _P (m) is a histogram relating to the first coordinate in the color space. Then, as in the case of the expression (1), the expression (12) is made smaller.
Seeking D _n>, an initial value update amount for the <D ⁰ _n> of <D _n> to its <[Delta] D _n> is

[0058]

[Equation 15]

It is obtained as <δn> which gives Similarly, <D ⁰ _n >, <D ¹ _n >, ..., <D ^K1 _n > are sequentially obtained, and the process ends K1 times under the determined conditions. And
The image P is classified by the value of <D ^K1 _n > closest to each pixel value with respect to the obtained <D ^K1 _n >. When the set of two-dimensional coordinates of the pixel closest to the obtained <D ^K1 _n > is represented by Ψ _n , the set Ψ _n is given by the following equation.

[0060]

[Equation 16]

Since the original image can be divided into N1 pieces in this manner, each set Ψ _n is divided into N2 pieces and then into N3 pieces in a similar manner in the subsequent procedure. That is, HΨ _n1 (m) is a histogram of Ψ _n1 with respect to the second axis of the color space, and the representative column points on the second axis of the color space are <0 ≦ E _n ≦ M2 | n = 1,2 ... N1. >, And the distance function on the second axis of the color space is d2,

[0062]

[Equation 17]

As in the case of <D _n >, <E ^{K2 (n1)} _n > is calculated so that becomes smaller. Where K2 (n
1) is the number of iterations in the n1-th set Ψ _n1 . Then, each set Ψ _n is divided using <E ^{K2 (n1)} _n >. Further, the n2-th set Ψ _n1 , _{n2 obtained} by dividing each set Ψ _n1 into N2 pieces is given by the following equation.

[0064]

[Equation 18]

Further, on the third axis of the color space, exactly the same processing is performed.
And the representative column point <F^{K2 (n1, n2})_nFor each set Ψ
_n1,_n2Is divided into N3 pieces. In this case, the n3rd set
Ψ_n1, n₂,_n3Is the distance function on the third axis of the color space as d3
Then it is given as follows.

[0066]

[Formula 19]

By the above processing, all pixels are N1 × N2 ×
Pixel values classified into N3 sets and representing each set Ψ _n1 , _n2 , _n3 are (D ^K1 _n , E ^{K2 (n1)} _n , F ^{K2 (n1, N2)} _n ).
Alternatively, if an average of 3 pixels is used, a color palette is created.

FIG. 4 shows the apparatus of the second embodiment, and FIG. 5 shows the representative point determining means 2 for the first axis, the second axis and the third axis of FIG.
2a, 22b and 22c are shown in detail. The histogram creating means 21a, 21b, and 21c shown in FIG.
The buffer 2 and the division mask creating means 23a and 23b in the previous stage
A histogram of the first axis, the second axis, and the third axis of the color space is created on the basis of the data of 1.

Then, in the representative point determining means 22a, 22b and 22c for the first axis, the second axis and the third axis, as shown in detail in FIG. 5, as in the case shown in FIG. 226 and the equation (6) by the initial representative point creating means 227.
Is calculated, the variation amount ΔC ^k _n with respect to the previous color palette is determined as the value shown in Expression (7), and is applied to the distance calculation means 221. Equation (12) is calculated by the distance calculation means 221.
The distances d1, d2, and d3 from each point in the color space to the closest point in the color palette are obtained from the histogram of each axis in the above and the next representative point candidate created by the next representative point candidate creating unit 226.

Then, the minimum distance representative point selection means 222
Therefore, in equation (12), each point in the color space is
-The distance to the closest point on the pallet is calculated, and the
Step 223 adds to previous color palette
The next representative point is selected from the added values by the table point selection means 224.
And evaluated by the evaluation means 225 to obtain equation (12),
Minimize (15) etc. <D_n>, <E_n>, <F_n
> Is required. In addition, the next representative point candidate creating means 226
This <D_n>, <E_n>, <F_n> To next representative point
A candidate is created and applied to the distance calculation means 221, and the next
<D_n>, <E_n>, <F_n> Is required. Also, FIG.
The division mask creating means 23c shown in FIG.
Of the representative points of the 1st to 3rd axes finally obtained by the step
Pixel value (D K¹ _n, E^{K2 (n1)} _n, F^{K2 (n1, N2)} _n) Or 3 strokes
A raw average is assigned to the raw data.

Next, a third embodiment of the present invention will be described. In the first and second embodiments, since the update amount is δ, when the variation range of the color space can take a value of 256, that is, a value of 0 to 255, even in the worst case even if the value monotonously approaches the optimum value. The process must be repeated 255 / δ times. Note that if δ is increased, the number of times decreases, but the accuracy deteriorates. For example, if δ = 64, the possible values of D ^k _n (k = 1, 2 ...) With respect to the initial value D ⁰ _n = 0. , 0, 64, 1
There are four types, 28 and 192. Therefore, this is 3
The process is completed in about a few times, but it is no longer sufficient as a color palette.

Therefore, in the third embodiment, in order to solve the above problem, the optimum <D _n > (or <E _n >, <F _n >) with a small number of iterations and high accuracy is obtained. ing. Specifically, let r be a positive number less than 1.

[0073]

## EQU20 ## δk = r ^k (M1 + 1) (18)

<D ^k _n > is updated by using the above equation until δk <1. Therefore, especially when r = 1/2, M1 +
If 1 ≦ 2δ, an arbitrary point can be reached p times from any initial point. In addition, reachable proof is 0-2
^Since the integer of ^p −1 can be represented by a binary number of p digits, it can be easily derived.

At this time, the initial point sequence <D ⁰ _n > has to be arranged in an odd number or in random numbers so as not to be evenly spaced. This is because, if an even number of initial point strings <D ⁰ _n > are arranged at equal intervals, first of all, the first movable point always has other points <D ⁰ _n >. With this there is a point to duplicate in the _{<D n>, J (<} D n>) is increased always on the definition nature, therefore, update amount becomes "0" Further,, r = 1/2 Then, the remaining p-1 times cannot move to a point beyond (M1 + 1) / 2.

FIGS. 6A to 6C respectively show R, G, and B histograms of the standard image "skin color evaluation image (up-image of woman)" of the ITE (Institute of Television Engineers) as an example. The original data of R, G, and B of this standard image are all 8 bits, and the image size is horizontal 760 × vertical 4
83. With respect to such a standard image, the division numbers N1, N2, and N3 on each coordinate axis of the color space are set to 7, 7, and 5, respectively, and the initial point sequences <D ⁰ _n >, <E ⁰ _n >, and <F ⁰ _{n are set.}
A case where the symbols> are arranged at equal intervals will be described.
Further, a case will be described in which r = 1 and the absolute value of the difference between coordinate values is used as the distance function on the first, second, and third axes of the color space.

FIG. 7A shows a case where the initial point sequence <D ⁰ _n > is arranged at equal intervals as in the conventional example with respect to the R histogram shown in FIG. 6A. “↑” indicates each D ⁰ _n . 7 (b) is shown in FIG. 6 (a).
The case where <D ^K1 _n > obtained by the second embodiment is arranged with respect to the R histogram shown in FIG.
The boundary of the area divided by ^K1 _n > is shown.

FIG. 8A shows the set Ψ in the second embodiment.
₁Histogram and its initial point sequence <E⁰ _n> Is shown
Then, FIG. 8B shows the set Ψ.₁Histogram and its initial point
Column <E^K2^(K1) _n> Is shown. In addition, FIG.₁,₁ of
Histogram and its initial point sequence <F 0_n> Shows the layout
9 (b) is Ψ₁,₁ Histogram and its initial point sequence <F
^{K3 (K}1^{, K2)} _n> Shows the arrangement. And each set Ψ_{n1, n2, n3}
The average of RGB of
A color palette as shown in is obtained. In addition, () in the table
The numbers inside are RGB pixel values of each color palette.
The number on the right side of () is for each color palette.
Shows the number of pixels.

[0079]

[Table 1]

[0080]

[Table 2]

[0081]

As described above, according to the present invention, a color palette that reduces the sum of the product of the distance from each point in the color space to the closest point of the color palette and the histogram of the color image is obtained. Therefore, the arrangement position of the sample points is optimized according to the image data, and the product area increases the specific gravity of the small area.
It is possible to prevent the false contour and the loss of color in a small area.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of a display limited color optimization method and a limited color display device according to the present invention.

FIG. 2 is an explanatory diagram showing arrangement positions of sample points by the method and apparatus of FIG.

FIG. 3 is a block diagram showing a modified example of the apparatus of FIG.

FIG. 4 is a block diagram showing a second embodiment of the present invention.

5 is a block diagram showing in detail representative point determining means for the first, second, and third axes of FIG. 4. FIG.

FIG. 6 is an explanatory diagram showing R, G, and B histograms of a standard image.

FIG. 7 is an explanatory diagram comparing the arrangement positions of the sample points of the conventional example and the present example with respect to the histogram of FIG.

FIG. 8 is an explanatory diagram showing an arrangement of a histogram of a set Ψ ₁ and its initial point sequence in the second embodiment.

FIG. 9 is an explanatory diagram showing the arrangement of a histogram of a set Ψ _1,1 and its initial point sequence.

[Explanation of symbols]

 3 21a, 21b, 21c Histogram creating means 4 Initial representative point creating means 5 Distance calculating means 6 Distance minimum representative point selecting means 7 Adding means 8 Evaluation means 22a, 22b, 22c Representative point determining means

Claims (2)

[Claims] 1. A display-limited color optimization method for obtaining a color palette by further reducing the sum of the products of the distance from each point in the color space to the closest point in the color palette and the histogram of the color image. 2. A histogram creating means for creating a histogram of a color image, and image data of the color image and a histogram created by the histogram creating means, from each point in the color space to the closest point of the color palette. A limited color display device having a distance calculating means for calculating a distance, and means for obtaining a color palette that reduces the sum of products of the distance calculated by the distance calculating means and the histogram created by the histogram creating means.