JP2000333028A - Color image processing method and its device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a color image processing device that can execute color conversion by which no ripple is caused in an entire image in any color conversion or an image input and gradation is emphasized against a task for solving disturbed gradation having been caused in a conventional color management technology using three-dimensional color spatial interpolation. SOLUTION: A gradation direction decision section 104 detects a gradation direction on a color image for each partial area in an image pattern and a gradation direction coefficient interpolation section 105 obtains a gradation direction coefficient on an image while keeping continuity. Then an output mix section 106 applies weighting interpolation to an output result from a color space interpolation section 107 processing different pixels by using a gradation direction coefficient so as to attain color conversion while keeping continuity and smooth gradation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a color image processing technique in general, and particularly to a color management technique.

[0002]

2. Description of the Related Art In recent years, in the field of color management technology, a software mechanism has been incorporated at the OS level of a computer in order to achieve color matching between devices such as a scanner, a monitor, and a printer. The function of this CMS (color management system) is to prepare device color conversion characteristic information called a "profile" for each color device in advance, and execute a color conversion on a color image based on the profile by a color conversion engine. It is performed by The most recently used configuration of profiles and color conversion engines is multidimensional interpolation.

In this method, an input / output relationship in color conversion is represented by M
This is regarded as a mathematical transformation from a dimensional input to an N-dimensional output space, and color conversion is performed not by calculation but by lookup of a color conversion table. In many color images currently used, M and N are 3 or 4 because pixel values are represented by three variables and four variables such as (RGB) or (CIE-LAB) or (CMYK). . However, in the case of a color multispectral image expected to be used in the future, M and N may have ten or more dimensions. Here, for convenience of explanation,
The color conversion is considered as a three-input one-output conversion. In the table method, if each color signal has 8 bits in the three-dimensional case, it is necessary to construct an enormous size table for each of 16 million full-color colors as it is. Therefore, for example, when the input color is in the (RGB) space, first, the input color space is divided into grids, the output values on the grid points are stored in a three-dimensional table f (i, j, k), and this is stored in a “profile”. ".

At the time of actual color conversion, the color conversion engine searches the output values at the neighboring grid points in the three-dimensional space of the three-dimensional table for the color value (RGB) of each pixel of the image, and weights and adds them. Calculate output by interpolation method. This technique has been widely used because the conversion accuracy can be increased by increasing the table size, the conversion speed is high, and it is constant regardless of the color conversion content. FIG. 17 shows a typical configuration diagram of this color conversion engine. In this figure, a tetrahedron (triangular pyramid) is used for the three-dimensional interpolation of the color space, but as shown at the bottom of the figure, a cube (8 points), a triangular prism, a triangular prism (6 points), and a pyramid (5 points) ), Various three-dimensional interpolation methods have been developed.

[0005] In the three-dimensional interpolation method, occurrence of an interpolation error is inevitable. In particular, the turbulence (ripple) of the interpolation result resulting from color conversion of a smooth gradation of an image has a much greater effect on human vision than the absolute error between color values. Regarding this ripple, for example, the applicant's document 1: Kanamori: “Interpolation error analysis in color conversion using 3D table” (Color Forum JAPAN'9
It is explained in 8). Even in the case of three-dimensional interpolation using the same gamma conversion as a cube on the gray scale, the global (global) gamma conversion with the entire color space as a defined area shows an interpolation characteristic in which the converted tone curve is approximated by a broken line. When local gray-scale conversion is performed only on the gray gradation portion in the color space, the interpolation curve becomes more violent as the gradation is inverted in the interpolation result (ripple). Even if the table size is increased, the ripple amplitude is increased. An anomalous phenomenon has been found that does not change.

As can be seen from this example, in general, a linear multidimensional interpolation method needs to use a lot of lattice point information, which is a high-dimensional characteristic hidden on a smooth gradation in a two-dimensional plane of an image. As a result, ripples often appear and the gradation characteristics deteriorate. Also, the ripple phenomenon has already occurred in three-dimensional interpolation at present. However, when a higher-dimensional four-dimensional color space or a multi-spectral image is handled, the ripple phenomenon occurs.
It can easily be imagined that the problem will become more serious if it becomes a three-dimensional color space.

On the other hand, Reference 2: James M. Kasson: “Perfor
ming color space conversions with three-dimentiona
l linear interpolation ”, Journal of Electronic im
aging / Jul 1995 Vol.4 (3) pp.226-249) and the tetrahedron interpolation method using the body-centered structure described in JP-A-8-83330, in which the tetrahedron group is made to have the same shape. That the grid points are located at the center of the cube where the interpolation error is the largest, and that when the number of grid points is doubled for each axis, the total becomes eight times, it can be doubled. And has good interpolation characteristics. In particular, since tetrahedral ridge lines occur in many directions in the color space, the degree of matching of the ridge line direction with the input direction of the image increases, and as described above, there is little effect of generating large ripples. is there.

However, the body-centered structure has disadvantages such as the necessity of having unnecessary extrapolation points outside the RGB cube, the complicated spatial division and the long processing time. Also, no matter how much the direction of the ridge line increases, an input that deviates from it can always be made, and it is not a complete solution to the problem. In addition, it is necessary to use a complicated space division method even in three dimensions, and there is a problem that the amount of calculation tends to be enormous in high-dimensional color processing using multispectrum.

Conventionally, Japanese Patent Application Laid-Open No. 6-31355 discloses switching between tri-linear interpolation and tetrahedral interpolation depending on input colors, and Japanese Patent Application Laid-Open No. 7-95423 discloses tri-linear interpolation, triangular prism interpolation, and pyramid interpolation. There is also a method of reducing the overall color conversion error by switching or adopting different interpolation methods depending on the position of the input color in the color space.

However, these methods cannot be universally used in various non-linear color conversions because the interpolation method is determined by the position of the input color in the space.

[0011]

As described above, in the conventional color conversion using the multi-dimensional color space interpolation method, interpolation with good gradation is always performed in correspondence with various color conversions and various image inputs. Although the method is required to be implemented, there has been no solution that is sufficient for any image or multidimensional color processing expected in the future.

The present invention solves the above problems,
When a color image is subjected to color conversion by a multidimensional interpolation method in the field of color management, if the color conversion itself has non-linearity, the gradation (gradation) direction in the color space on the color image is determined in the image screen. Detect for each partial area and sequentially control the interpolation method for each partial area. More specifically, in interpolation in color space, not all grid points surrounding the input point are used but exist in the gradation direction of the image. An object of the present invention is to provide a color image processing method capable of reducing nonlinearity in multidimensional interpolation by selecting and interpolating a small number of grid points to be performed and performing color conversion with an emphasis on gradation without generating ripples. And

[0013]

According to the present invention, there is provided a pixel input section for scanning a color image and inputting each pixel value, and a direction in a gradation color space for each partial area on the image. A tone direction determining unit for determining the directionality and determining a direction weight, a tone direction coefficient interpolating unit for continuously interpolating the direction weights in the tone direction determining unit in an image plane and outputting a tone interpolation coefficient, A plurality of different color space interpolation units for performing color conversion on the pixel values using a conversion table, and an output control unit for mixing or selecting outputs from the plurality of color space interpolation units with the gradation direction coefficients. .

Thus, any color conversion and any image input can be performed with an emphasis on gradation, which does not cause ripples in the entire image.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS According to the first aspect of the present invention, a color image is scanned, each pixel value is inputted, and a color conversion table is referred to from position information of the pixel value in a color space by referring to a color conversion table. In a method of performing color conversion using an interpolation method, it is characterized in that gradation direction information of an image is obtained for each partial region on an image including the pixel, and the multidimensional interpolation method is controlled using the gradation direction information. It is a color image processing method that reduces the non-linearity that has occurred due to the use of unnecessarily many grid points because the interpolation method that matches in the gradation direction is output with particular emphasis from a plurality of interpolation methods. Has the effect of reducing artifacts such as ripples in the output gradation.

According to a second aspect of the present invention, there is provided a method of calculating gradation direction information in the color image processing method according to the first aspect, wherein the magnitude of the luminance gradient is determined for each color plate in a plurality of regions. It is characterized in that it outputs a direction weight for each direction axis discretized in the color space with respect to the direction of the gradation, and has the effect of expressing the direction of the gradation of the image as a direction in the color space.

According to a third aspect of the present invention, there is provided a method of calculating a gradation direction in the color image processing method according to the first aspect, wherein a color image is divided into discrete directions of a gradation near a large evaluation point in a divided area. The method is characterized in that directional weights for axes are continuously interpolated between the divided areas, and the continuity of the gradation of the image is obtained even when a plurality of different multidimensional interpolation methods are sequentially switched in the image plane according to the gradation direction of the image. Does not collapse.

According to a fourth aspect of the present invention, there is provided a pixel input section for scanning a color image and inputting each pixel value, and for determining a direction in a gradation color space for each partial area on the image including the pixel. A tone direction determining unit for determining and obtaining a direction weight, a tone direction coefficient interpolating unit for continuously interpolating the direction weight in the tone direction determining unit in the image plane and outputting a tone interpolation coefficient, A plurality of different color space interpolators for performing color conversion on the pixel values using a conversion table, and an output from the plurality of color space interpolators to a gradation output from the gradation direction coefficient interpolator. And an output control unit that mixes or selects according to the direction coefficient. Since the output is performed with particular emphasis on an interpolation method that matches the gradation direction from a plurality of interpolation methods, an effect that ripples are reduced in the output gradation is achieved. Have.

According to a fifth aspect of the present invention, there is provided a pixel input section for scanning a color image and inputting each pixel value, and determining a directionality in a gradation color space for each partial area on the image including the pixel. A tone direction determining unit for determining and obtaining a direction weight, a tone direction coefficient interpolating unit for continuously interpolating the direction weight in the tone direction determining unit in an image plane and outputting a tone interpolation coefficient, A plurality of different color space interpolation units for performing color conversion on the pixel values using a conversion table; an output comparison unit for comparing interpolation outputs from the plurality of color space interpolation units; and the plurality of color spaces An output control unit that mixes or selects the interpolation output from the interpolation unit based on the determination result in the output comparison unit based on the gradation direction coefficient, and performs a plurality of interpolations only when the color conversion itself has nonlinearity. Special emphasis on interpolation methods that match the gradation direction from methods For outputting Te, it has the effect of ripple is reduced to an output gradation without reducing the processing speed.

According to a sixth aspect of the present invention, there is provided a pixel input section for scanning a color image and inputting each pixel value, and determining a directionality in a gradation color space for each partial area on the image including the pixel. A tone direction determining unit for determining and obtaining a direction weight, a tone direction coefficient interpolating unit for continuously interpolating the direction weight in the direction determining unit in the image plane and outputting a tone interpolation coefficient, and a color conversion table. A plurality of different color space interpolators for performing color conversion on the pixel values using the table, a table non-linearity map describing an area having a strong non-linearity in the color conversion table, and a plurality of color space interpolators. And an output control unit for mixing or selecting the output according to the gradation direction coefficient based on the table non-linearity map. Only when the color conversion itself has non-linearity, a plurality of interpolation methods are used in the gradation direction. Matching interpolation method For outputting to focus on, has the effect of ripple is reduced to an output gradation without reducing the processing speed.

The invention described in claim 7 is the invention according to claims 4 to 6
A color image processing apparatus according to any one of the above, wherein the color image processing unit calculates the magnitude of the luminance gradient for each color plate in a plurality of regions, It comprises a part for calculating the weight for each direction axis discretized in the space, and has the effect of expressing the direction of the gradation of the image as the direction in the color space.

According to an eighth aspect of the present invention, a test signal prepared in advance for a multi-dimensional table is input, different multi-dimensional interpolation is performed, and a portion having a large nonlinear intensity in the color conversion table is determined based on a variation in the output. 7. A method for creating a table non-linear map according to claim 6, wherein a file having a strong non-linear intensity in the multidimensional color conversion table is determined in advance in the color image processing apparatus according to claim 6. It has the effect of being able to deal with it.

An embodiment of the present invention will be described below with reference to FIGS.

(Embodiment 1) FIG. 1 is a block diagram showing the configuration of a color image processing apparatus according to Embodiment 1 of the present invention. In the present embodiment, what is assumed as the color input image 101 is an image in which one pixel is defined by three variable signals of (R, G, B).
Version and B version.

In FIG. 1, reference numeral 102 denotes a color image 10
A pixel input unit 104 for scanning 1 and inputting each pixel value; a gradation direction 104 for determining a directionality in a color space of a gradation for each partial region on an image including a pixel and obtaining a direction weight u of each region A determination unit 105 is a gradation direction coefficient interpolation unit that continuously interpolates the direction weight u in the gradation direction determination unit 104 in the image plane and outputs a gradation direction coefficient Wi (m, n). A plurality of different types of color space interpolation units for performing color conversion on the pixel values using the color conversion table 103, and outputs from the plurality of color space interpolation units 107 to a gradation direction coefficient interpolation unit 105.
Is an output control unit that outputs an output color image 108 by mixing or selecting according to the gradation direction coefficient Wi (m, n) output from.

The operation of the color image processing apparatus configured as described above will be described below.

The xy coordinates on the screen of the target pixel in the color input image 101 are (m, n), and the color values are (R, G, B).
When the color value (R, G, B) of the target pixel is made up of 8 bits for each signal, the pixel input unit 102 sets a set of 3 bits of the upper signal (Rh, Gh, Bh) and a set of 5 bits of the lower signal (Rh, Gh, Bh) Rl, Gl, Bl) and supply them to the color conversion table 103 and the color space interpolation unit 107, respectively.

The color conversion table 103 is looked up with the upper signal (Rh, Gh, Bh) divided by the pixel input unit 102, and an interpolation operation is performed with the lower signal (Rl, Gl, Bl). The color conversion table 103 stores colors at 9 × 9 × 9 grid points, which are vertices obtained by dividing the entire RGB color space into 8 × 8 × 8 = 256 corresponding to the higher-order signals (Rh, Gh, Bh). Conversion output is stored. The bit division method of 3 bits and 5 bits used here determines the size of the color conversion table 103, that is, the number of lattice points (distance between lattice points: 2 ⁵ = 32 in this case), but is actually arbitrary. Color conversion table 103
Creates not only color conversions that can be expressed by mathematical expressions (model expressions) for the purpose of performing various color corrections that perform color reproduction, but also creates a table with a high degree of freedom that does not have mathematical models by measuring the color of color devices It is also possible.

There are various definitions of input / output of color conversion. Conventionally, color conversion for a printer from an integrated density space (Dr, DgDb) to an ink (toner) signal (C, M, Y, K) has been performed. Although it was common, in the world of color management, conversion from (R, G, B) to standard color space (L, A, B) (X, Y, Z) or inverse conversion etc. is targeted . For the present invention (R, G, B)
Let's consider a color conversion from to a certain output color f. That is, the three-variable function f (R, G, B) is stored in the color conversion table for one output color.

The upper signal (Rh, Gh, Bh) from the pixel input unit 102 designates one of the 8.times.8.times.8 divided cubes of the color conversion table 103. That is, the eight nearest neighbor points including the input color are found by the upper signal, and the output f there is obtained. The method of interpolating using these eight points is tri-linear interpolation, which is the most common three-dimensional interpolation method.

In the color space interpolation section 107, (0) is a part for executing the tri-linear interpolation method, and its function will be briefly described below with reference to FIG. The input signal is represented in the cube composed of the obtained eight points by lower signals (Rl, Gl, Bl), and these are used as weights in the three-dimensional interpolation processing. Here, as shown in FIG. 2, the input color is P, the cube in the color space including P is ABCD-EFGH,
If the position of P is (Rh + Rl, Gh + Bl, Bh + Bl),

[0032]

(Equation 1)

Therefore, the interpolation output f _I at P can be expressed by the following (Equation 2).

[0034]

(Equation 2)

However, the weighting factors at that time are as shown in (Table 1).

[0036]

[Table 1]

Tri-linear interpolation is conventionally used as a known technique. However, although it is generally said that the accuracy is high because the interpolation method uses eight points, it is known that nonlinear color conversion causes ripples in the interpolation characteristics. The cause is related to the fact that each weight coefficient in Table 1 is a cubic expression. A method conventionally used to solve this problem is tetrahedral interpolation. What is tetrahedral interpolation?
Although it is a generic term of linear interpolation using four points including the input color in the color space, the most common is to perform a tetrahedral division in which the above-mentioned cube ABCD-EFGH has six volumes equal to each other with the diagonal line as the border of the boundary, This is a diagonal-tetrahedron interpolation method of determining which tetrahedral region is included and interpolating using the four points. The diagonal / tetrahedron interpolation has a good property that no ripple occurs because linear interpolation is performed in the interpolation solid, and the amount of calculation is small because the number of points used in the three-dimensional interpolation is four, which is the minimum number. . However, unlike the eight-point interpolation that is isotropic in a three-dimensional space, the above-mentioned good properties differ in properties depending on the directionality.
Therefore, in the present invention, four types of diagonal / tetrahedron interpolation are used in consideration of the directionality. That is, from the color space interpolation unit 107 (1) to the color space interpolation unit (4) in FIG. 1, diagonal / tetrahedron interpolation is performed, but (D1) and (D2) in FIG. D2),
As shown in (D3) and (D4), the diagonal directions, which are the diagonal tetrahedron division methods, are different in four types.

[0038]

[Table 2]

Here, the (D1) type rule cube ABCD-EF
The interpolation method will be described by taking the division along the diagonal line of the GH as an example, but the description is the same for the other three types.

FIG. 4 is a diagram for explaining a six-division method of the diagonal / tetrahedron interpolation method (D1). According to (Table 3), there are six magnitude relations of the lower signals (Rl, Gl, Bl), which determine one of the six-divided tetrahedron. Rl
When = Gl = Bl, it may be included in any of Area0 to Area5, but here Area0 is used.

[0041]

[Table 3]

After the divided tetrahedron is determined, four-point interpolation is performed as follows.

[0043]

(Equation 3)

The weighting factor at this time is determined as shown in the following (Table 4).

[0045]

[Table 4]

Next, the gradation direction determination unit 104, gradation direction coefficient interpolation unit 105, and output mixing unit 106 will be described.

The gradation direction judging section 104 judges the direction of the gradation of the input image, and obtains the direction weight u. The gradation direction coefficient interpolation unit 105 continuously interpolates the direction weight u from the gradation direction determination unit in the image screen, and outputs a gradation interpolation coefficient Wi (m, n).

The output control unit 106 gives a large gradation direction coefficient to an interpolation method having a division line in a direction close to the direction, and gives a small gradation direction coefficient to an interpolation method having a division line in a different direction. Add the results. Output control unit 1
At 06, the outputs from the color space interpolation units (0) to (4) are expressed as f _I (i) (i = 0-4), and are mixed using the gradation direction coefficient Wi.

[0049]

(Equation 4)

The gradation direction coefficient Wi is given for each interpolation method.

[0051]

(Equation 5)

It is assumed that the above is satisfied.

Next, a method of calculating the gradation direction coefficient Wi will be described with reference to FIG. First, the color image is red,
Think of it as an image of a single color plate consisting of three color plates of Green and Blue,
A local image gradation level is determined for each plate.
First, as shown in FIG. 9, an image is divided into partial regions, and a representative point (m0, n0) is determined in a region including a target pixel position (m, n) for which color conversion is to be performed. Calculate the gradation direction. That is, the brightness gradient vector for the image
Filtering for calculating D may be performed. Here, partial differentiation is used, but in reality, this operation is a digital difference operation.

[0054]

(Equation 6)

It is desirable that the gradation direction vector D is normalized to length = 1 in order to make the quantity irrelevant to the sharpness of the gradation. Next, the direction of the gradation direction vector D is examined. This can be simply realized by taking an inner product with the seven direction vectors Vj. The seven direction vectors V are shown in FIG.
There are seven directions from 1 to V7, and the RGB coordinate components of each vector are as follows (Table 5).

[0056]

(Equation 7)

[0057]

[Table 5]

The seven weighting factors uj for each component obtained above
For, from u ₁ to u ₄ directly used as a weight of the color space interpolation unit (1) to (4), from u ₅ to u ₇ used as the weight of the color space interpolation section is added (0).

[0059]

(Equation 8)

Then, the gradation direction coefficients Wi of W ₀ , W ₁ , W ₂ , W ₃ , and W ₄ are determined for each area Bn.

This value may be output as it is and used for each divided area. However, if the gradation direction coefficient Wi greatly changes between the areas, there is a possibility that a color or gradation step occurs for each divided area. To prevent this, the gradation direction coefficient Wi (m, n) is interpolated by bi-linear interpolation within the screen so that the gradation direction coefficient Wi maintains continuity in the image. Output.

[0062]

(Equation 9)

[0063]

(Equation 10)

Here, the weighting coefficient ω _P of the gradation direction coefficient,
ω _Q , ω _R , ω _S are as shown in the following (Table 6).
Here, BX and BY indicate the width and height of the divided area. The above interpolation is performed for each Wi in the image.

[0065]

[Table 6]

As shown in FIGS. 9 and 10, in this method,
The weighting coefficient leaves an area that cannot be interpolated around the image. Here, the value of the representative point (m0, n0) in the extrapolation or the divided area may be used as it is. Representative point (m0, n0)
May be set at the upper left of each divided area, and extrapolation may be unnecessary by adopting a virtual fixed value at the right end of the image.

Next, the flow of actual image processing will be described with reference to FIG.
1 will be described. In this example, the color image is assumed to be an (RGB) dot-sequential image for both input and output, and the color conversion process is also performed in a dot-sequential manner.

After the start of the process, in step S1101, the color image is divided into regions as shown in FIG. S
In step 1102, the gradation direction at the representative point (m0, n0) in the region is determined, and the gradation direction coefficient for each divided region is determined.
Ask for Wi. In step S1103, a three-dimensional table used for color conversion is read. The processing up to this point is executed as preprocessing of image processing.

Next, while scanning each pixel of the color image, the processing from step S1106 to step S1110 is repeated for each pixel until the processing for all pixels is completed. First, in step S1106, the position of the pixel
(m, n) and a pixel value (RGB) are input. Step S1
At 107, the gradation direction coefficient W at the pixel position (m, n)
i (m, n) is obtained by interpolating the gradation direction coefficient Wi for each of the divided areas.

Next, in step S1108, different color space interpolation processes are executed in parallel, and the interpolation outputs f _I (0) to f
_{Find I} (4). In step S1109, S1107
The interpolation output is mixed using the gradation direction coefficient Wi (m, n) obtained in (1). In step 1110, the output pixel (R'G '
B ') is inserted into the output image.

Since the process is repeated as it is, the process returns to step S1104. When the process for all the pixels is completed, the process ends according to the determination in step S1105.

In the above description, the color space interpolation unit 10
7 has been described as four types of trilinear interpolation and four types of diagonal / tetrahedron interpolation. However, as long as it is a divided space having directivity, it can be similarly implemented by a combination of other interpolation methods, and the types are not limited to four types. Of course not.

The configuration and operation of the present invention have been described above. Here, the principle of the present invention that the content of interpolation is changed depending on the gradation direction will be described. First, why is trilinear interpolation and diagonal tetrahedron interpolation (and
It will be described whether the above-described method is used in combination with the above-described method and whether the weighting is performed in consideration of the gradation direction of the image.

Conventionally, there is no clear explanation as to which of the tri-linear interpolation and the diagonal / tetrahedron interpolation is a high-precision interpolation method when the color space used and the color conversion used are different. However, when the color conversion itself used is expressed by linear combination of one-dimensional conversion as described in the above-mentioned document 1, the interpolation result is the same in all linear interpolations, and the output result is different for each component. It is known that a broken line state is obtained. That is, in this case, there is no difference between the results of the tri-linear interpolation and the diagonal-tetrahedron interpolation, and there is no ripple.

This will be described graphically by taking two-dimensional interpolation as an example. First, in two dimensions, note that trilinear interpolation corresponds to bilinear interpolation using four points, and diagonal / tetrahedron interpolation corresponds to three-point interpolation of a triangle obtained by dividing a square into two. In FIG. 5, a square ABCD is one interpolation section set on the RG plane, and is represented by an arrow raised on this two-dimensional plane with one output component f. The color conversion is linear when f (A), f (B), f (C), f
(D) is equivalent to being on the same plane π. Therefore, the interpolated value by bilinear interpolation, which is linear interpolation from these four points, is constrained on the same plane π as shown at 501, and the output is a straight line for a color input in any locus within the square ABCD.

Next, reference numerals 502 and 503 in FIG. 5 show the results of interpolation in different types of triangular divisions such as AC division and BD division. In any of the division methods, interpolation is performed within each triangle after determining which of the two types of triangles the input point in the plane belongs to. The interpolation value within a triangle is equivalent to dividing this plane π into a triangle ABC or a triangle ADC and interpolating it, but since it originally divides the plane, it is already bilinear interpolation in either AC division or BD division Obviously, the same interpolation result is obtained. As described above, when the color conversion itself has no nonlinearity, a phenomenon (ripple) in which the result deviates from the ideal value or the output becomes a curve does not occur regardless of the interpolation method.

Next, in the case where the color conversion, which is said to have obtained better results than trilinear interpolation in diagonal / tetrahedron interpolation in Japanese Unexamined Patent Publication No. Hei 6-31355, is MIN operation, especially when diagonal / tetrahedron interpolation is used. Among them, the case of (D1) type, which adopts a dividing line passing through the origin, is taken up. Similarly, considering two dimensions, the MIN operation is
Since the smaller one of the pair of two variables (R, G) is output, a triangle corresponding to the region of R ≧ G as shown by 600 in FIG.
In ABC, MIN (R, G) = G, and in triangle ADC corresponding to the region of R <G, MIN (R, G) = R. Therefore, two triangular planes with straight line R = G as a ridgeline Form a plane that is bent and connected. This is the true calculated value in the case of the MIN operation.

Next, consider that this is calculated by interpolation. First, as shown by 601, in the ABCD four-point interpolation method, the interpolation surface forms a curved surface, which causes a ripple. Next, in the triangular division by the AC division indicated by 602 in FIG. 6, since the ridge line is formed in the AC direction, the interpolation result is completely the same as the correct result. However, 603 B-
In D division, the value is 0 for one triangle divided in the BD direction
And the interpolation output greatly differs between the two division methods.

In the above-mentioned Japanese Patent Application Laid-Open No. 6-31355,
Taking this case as an example, it is stated that triangular interpolation using AC division can obtain better results than four ABCD points. However, to be precise, in the calculation of the true value of the MIN operation, the dividing line of the AC is a ridge connecting two types of planes exhibiting completely different characteristics, and the ridge and the dividing boundary of the triangle coincide with each other. Therefore, it is correct that the interpolation error becomes 0 and good interpolation is performed. After all, it is important that the optimum interpolation method is determined by some index when the true calculated value is known as in the MIN operation even in the color conversion having a strong nonlinearity. Conversely, when the true calculated value is unknown when the nonlinearity is strong, there is no index for knowing what kind of division / interpolation should be performed, so that the user immediately faces difficulty.

A typical example of this difficulty is a non-linear conversion such as the local gamma conversion described in the aforementioned reference 1. In this conversion, since only a color conversion table having a strong nonlinearity is given, it is necessary to three-dimensionally interpolate this in color conversion. That is, in the previous example, f (A), f
Only the values on the lattice points (B), f (C) and f (D) are given in a state of strong nonlinearity, and no true calculated value connecting them is given. The strong non-linearity means that two-dimensional values take small values for A and C at diagonal positions and large values for B and D at each point of ABCD as indicated by 700 in FIG. When this is bilinearly interpolated at four points, a complicated quadric surface is formed in a space formed by the f-axis, the R-axis, and the G-axis as shown by 701, which causes ripples.

As shown by 702 and 703 in the three-point interpolation, in the AC division and the BD division, the former forms a deep valley in the AC direction, whereas the latter forms a ridgeline of the BD, resulting in completely different interpolation results. However, in each case, severe irregularities occur instead of ripples. In other words, the interpolation results are significantly different between the four-point interpolation and the two types of three-point interpolation, and none of them is a good interpolation.

The basic idea of the present invention is to determine the optimum interpolation by including information on the direction of the gradation of the input image. For example, in 700 of FIG. 7, chain lines (1) and (2)
Indicates the input gradation, when the gradation in the AC direction as shown in (1) is input, the three-point interpolation of the AC division is adopted, and the gradation in the BD direction as in (2) is obtained. B- if entered
It can be said that three-point interpolation of D division is adopted.

Next, after the input gradation direction is determined, what kind of interpolation method should be used will be described with reference to FIG. 12, taking a two-dimensional case as an example. In FIG. 12, four directions passing through the center point of the square ABCD are set as input gradations, the input point is indicated by a black circle, and the interpolation output value at that point is indicated by an arrow on the black circle. Reference numerals 1200 and 1201 denote cases where the input grayscale directions are the EF direction and the GH direction, respectively. In this case, the same bilinear interpolation may be used for both.

In FIG. 7, reference numeral 701 denotes a bilinear interpolation surface which looks as if it exhibits a complicated quadratic surface.
As shown by 200, when the input direction is the EF direction, the interpolation results are the values f (E) and f on the middle point E of AB and the middle point F of CD.
There is a property that it becomes a straight line connecting (F) and no ripple appears. Similarly, when the input tone direction 1201 is GH, the interpolation value is located on a straight line connecting f (G) and f (H) as a result of the bilinear interpolation.

Next, in 1203, since the input gradation is in the AC direction, triangular interpolation of AC division is used. In 1204, since the input gradation direction is BD, triangular interpolation of BD division is used. Is f
No ripple appears because it is located on the straight line connecting (A) and f (C) or f (B) and f (D). As described above, in the two-dimensional case, four types of gradation directions (1) EF,
(2) GH, (3) AC, (4) BD (1)
(2) may use bilinear interpolation, and (3) and (4) may use different triangle interpolation. Of course, changing the interpolation method differently for each pixel on the image one after another may cause noise on the image. For this reason, it is more desirable to control the interpolation method as a weight than to select one type of optimal interpolation. In other words, a large weight is given to the result of the optimal interpolation method in consideration of the gradation direction of the image, and a small weight is given to the interpolation result that is not so.

When the above discussion is extended to three dimensions, trilinear interpolation should be used for any input gradation parallel to the cubic axis such as V5, V6 and V7 in the direction vector of FIG. Other diagonal inputs V1, V2, V
3 and V4, it is concluded that it is better to use diagonal tetratedron interpolation for the division of the D1, D2, D3 and D4 methods in FIG. For this reason, the sum of the direction weights u ₅ , u ₆ , u ₇ , and u ₈ is used as the gradation direction coefficient W ₀ for trilinear interpolation, and the other W ₁ , W ₂ , W ₃ and W ₄ are u ₁ , u ₂ , u ₃ and u ₄
Is used as it is.

As described above, in the first embodiment of the present invention, in any nonlinear color conversion, the gradation of the input image is obtained, several types of interpolation methods are performed, and the results are weighted to obtain the output gradation. It suppresses ripple in the tone.

(Embodiment 2) FIG. 13 is a block diagram of an image processing apparatus according to Embodiment 2 of the present invention, and will be described.

In the first embodiment, different color space interpolation units 1
The output from 07 is always weighted and mixed for each pixel by the gradation direction coefficient calculated from the pixel position of the input image 101. However, since the calculation of this weighted mixing is performed for all the pixels of the image, the calculation processing load is increased. Is big. Actually, non-linear high color conversion and deterioration of the gradation of an image due to the color conversion do not always occur.

Therefore, in the present embodiment, the outputs of the five types of color space interpolation units (0) to (4) are compared by the non-linearity judgment unit 1301, and when these are greatly different, that is, when the non-linearity is high. Only by using the weighted mixing, the processing load is reduced. Non-linear judgment unit 13
In 01, the following operation is performed. Here, THRESH is a fixed threshold, and f _I (i) and f _I (j) are f _I (0) to f _I (4).
Indicates one of MAX indicates the maximum value operation.

That is, the maximum degree of fluctuation when a different division direction and a different type of interpolation are used for a certain input color is investigated, and if it exceeds a certain threshold value, it is determined that the nonlinearity is strong. The result is determined to be YES, the others are determined to be NO, and the result is sent to the output control unit 106 to determine whether to perform output mixing.

[0092]

[Equation 11]

In this method, the non-linearity judging section 1301
It seems that gradation jumps may occur due to output switching due to the boundary for determining YES / NO.However, if the threshold is selected as the limit value that human eyes can recognize the gradation skip, the difference due to the interpolation method Even if the output mixing is stopped before the value exceeds the threshold, and the output mixing is started at the time when the output mixing exceeds the threshold, the gap is guaranteed to be smaller than the threshold, so that a large recognizable gradation jump does not occur.

(Embodiment 3) FIG. 14 is a block diagram showing an image processing apparatus according to Embodiment 3 of the present invention, and will be described. In the third embodiment, the nonlinearity determination unit in the second embodiment is eliminated, and instead, the color signal value of the input pixel is converted into the nonlinearity in the color conversion table based on the information of the table nonlinearity map 1401 prepared in advance. The processing load is reduced by executing the weighted mixing only when the data enters the portion having a large value. In this case, a table non-linearity map is also created at the stage when the color conversion table is created, and this information can be stored as one information of a color profile such as an ICC profile.

Next, a method of creating a table nonlinearity map will be described with reference to FIG. FIG. 15 is a block diagram showing a table non-linearity map creating device. The non-linearity of the color conversion, mainly the magnitude of the difference of the interpolation result depending on the division direction, is determined at the position in the color space of the input grid point corresponding to the three-dimensional table for the color conversion, and the interpolation output of the color is mixed. Is to be determined. The creation method is similar to that of the second embodiment, and performs color conversion of the test color image using the target color conversion table. A non-linearity discriminating unit 1 compares interpolation outputs output from different color space interpolating units.
At 301, the same judgment as (Equation 11) is performed, and (Rh, Gh, Bh)
Is stored in the table non-linearity map 1501 as a result of the non-linearity when the is input as the address.

(Embodiment 4) FIG. 16 is a block diagram showing an image processing apparatus according to Embodiment 4 of the present invention and will be described. In the fourth embodiment, a plurality of different color space interpolation units 107 (0) to 107 (0) to
A dedicated color conversion table 1601 is provided for each (4). The first to third embodiments of the present invention have an advantage in that the color space interpolation unit 107 having directionality according to the gradation direction of an image is used for one three-dimensional color conversion table. By using the three-dimensional color conversion table itself properly, the gradation property of the gradation direction of the image can be further improved.

Although the present invention has been described with reference to a three-dimensional color image, the same concept can be applied to an input space of generally four or more dimensions. If the output space is N-dimensional, each color conversion unit of M inputs and 1 output is set to N
It can be realized simply by arranging them.

Therefore, the present invention is also effective in color conversion of a multispectral color image having M inputs and N outputs.

In the present invention, the unit of gradation determination is a block divided area on the screen, but another unit of segmentation in an image may be used.

[0100]

As described above, according to the present invention, in order to solve the problem of gradation disturbance caused by the prior art performing color conversion using only each pixel value, the gradation (gradation) ) By detecting the direction for each partial area in the screen and sequentially controlling the interpolation method for each partial area, a color that emphasizes the gradation that does not cause ripple in the entire image for any color conversion and any image input A color image processing device capable of performing conversion can be provided. In the method of interpolating a multidimensional color multispectral image, it is conceivable that higher order terms have a more serious effect on the interpolation result due to an increase in input dimension, and that an unknown ripple is likely to occur. By using the concept of considering the above, realistically good interpolation calculation can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram of a color image processing apparatus according to a first embodiment of the present invention;

FIG. 2 is a diagram showing a tri-linear interpolation method;

FIG. 3 is a diagram showing four types of tetrahedral division methods.

FIG. 4 is a diagram showing a diagonal / tetrahedron interpolation method in the D1 division method.

FIG. 5 is a diagram showing color conversion interpolation without nonlinearity in a two-dimensional case;

FIG. 6 is a diagram showing color conversion interpolation of a MIN operation in a two-dimensional case;

FIG. 7 is a diagram illustrating color conversion interpolation with strong nonlinearity in a two-dimensional case;

FIG. 8 is a diagram showing a gradation direction determination of a color image.

FIG. 9 is a diagram showing a color image block divided area and a representative point position for calculating a luminance gradient.

FIG. 10 is a diagram showing interpolation of a gradation direction coefficient.

FIG. 11 is a diagram showing a processing flow in the first embodiment of the present invention.

FIG. 12 is a diagram showing an interpolation method to be used in different gradation directions in a two-dimensional case.

FIG. 13 is a block diagram of a color image processing apparatus according to a second embodiment of the present invention.

FIG. 14 is a block diagram of a color image processing apparatus according to a third embodiment of the present invention.

FIG. 15 is a block configuration diagram for creating a table non-linearity map according to the third embodiment of the present invention;

FIG. 16 is a block diagram of a color image processing apparatus according to a fourth embodiment of the present invention.

FIG. 17 is a block diagram of a conventional color conversion engine.

[Explanation of symbols]

 Reference Signs List 101 input color image 102 pixel input unit 103 three-dimensional color conversion table 104 gradation direction determination unit 105 gradation direction coefficient interpolation unit 106 output mixing unit 107 color space interpolation unit 108 output color image 1301 nonlinearity determination unit 1401 table nonlinearity map 1601 Five kinds of color conversion tables

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5B057 AA11 BA02 BA11 CA01 CA12 CA16 CB01 CB12 CB16 CE18 CH07 CH18 DC08 DC30 5C077 LL04 MP08 NP01 PP21 PP31 PP32 PP43 PP47 PP58 PP68 PQ08 PQ18 PQ20 PQ23 RR19 5C079 LA01 H01LA LA39 MA05 MA10 MA11 NA05

Claims (8)

[Claims] 1. Scanning a color image and inputting each pixel value,
In a method of performing color conversion using a multidimensional interpolation method with reference to a color conversion table from position information of the pixel values in a color space, obtaining gradation direction information of an image for each partial region on an image including the pixel values And controlling the multidimensional interpolation method using the gradation direction information. 2. A method of calculating gradation direction information, comprising determining a magnitude of a luminance gradient for each color plate in a plurality of divided areas of a color image, and discriminating a direction axis of the gradation direction in a color space. 2. The color image processing method according to claim 1, wherein a direction weighting factor is calculated for each color. 3. The color image processing method according to claim 2, wherein the calculation of the gradation direction information is performed by continuously interpolating the direction weighting coefficients between regions in the image. 4. A pixel input unit for scanning a color image and inputting each pixel value, and determining a directionality in a gradation color space for each partial region on the image including the pixel to obtain a direction weight. A tone direction determining unit, a tone direction coefficient interpolating unit that continuously interpolates the direction weights in the tone direction determining unit in an image plane and outputs a tone interpolation coefficient, and the pixel using a color conversion table. A plurality of different color space interpolators for performing color conversion on the values, and outputs or outputs from the plurality of color space interpolators are mixed or selected by the gradation direction coefficients output from the gradation direction coefficient interpolation unit. A color image processing apparatus comprising: an output control unit. 5. A pixel input section for scanning a color image and inputting each pixel value, and a floor for determining a direction in a color space of a gradation for each partial area on the image including the pixel to obtain a direction weight. A tone direction determining unit, a tone direction coefficient interpolating unit that continuously interpolates the direction weights in the tone direction determining unit in an image plane, and outputs a tone interpolation coefficient, and the pixel using a color conversion table. A plurality of different color space interpolation units for performing color conversion on the values, an output comparison unit for comparing the interpolation outputs from the plurality of color space interpolation units, and an interpolation output from the plurality of color space interpolation units. A color image processing apparatus comprising: an output mixing unit that performs mixing or selection based on the gradation direction coefficient based on a determination result in an output comparison unit. 6. A pixel input section for scanning a color image and inputting each pixel value, and a floor for determining a direction weight in a color space of a gradation for each partial area on the image including the pixel to obtain a direction weight. Tone direction determining unit, a tone direction coefficient interpolating unit that continuously interpolates the direction weights in the direction determining unit in the image plane and outputs a tone interpolation coefficient, and a color conversion table for the pixel values. A plurality of different color space interpolators for performing color conversion, a table non-linearity map describing an area where the non-linearity is strong in the color conversion table, and an output from the plurality of color space interpolators. A color image processing apparatus, comprising: an output control unit that performs mixing or selection based on the gradation direction coefficient based on a map. 7. A gradation direction judging section judges a magnitude of a luminance gradient for each color plate in a plurality of regions of a color image, and discriminates the directionality of the gradation in a color space for each direction axis. The color image processing apparatus according to claim 4, wherein a direction weight is calculated. 8. A different multidimensional interpolation is performed by inputting a test signal prepared in advance to a multidimensional table,
7. The table non-linearity map creating method according to claim 6, wherein a portion having a large non-linear intensity in the color conversion table is determined based on the variation of the output and is filed.