JPH05284346A - Color conversion device - Google Patents

Abstract

PURPOSE:To attain high speed color conversion and to use memories effectively by dividing a 3-dimension space formed by three color separation input color signals into rectangular prisms, dividing one rectangular prism into two triangular prism areas and applying inter-line interpolation to the areas so as to attain color conversion through parallel read from 6 memories. CONSTITUTION:A triangular prism discrimination section 107 discriminates into which of two triangular prism areas being divisions of a 3-dimension space a separated input value is included. On the other hand, outputs and differences with respect to input points of the triangular prism are stored in color conversion table memories 110-115 and read by using a high-order signal of an input signal as an address. Outputs of each color conversion table memory are subject to weighting by sing low-order signals 104-106 of the input color signal at multiplier groups 119, 120, 126, 127 and adder groups 121, 122, 128, 129, and the inter- line interpolation is implemented by using the output at each of 6 apexes of the triangular prism. Thus, high speed color conversion is implemented and the memories are used effectively through the memory structure of 2's power.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an application for inputting a color image signal or a color video signal and performing arbitrary color conversion in real time, for example, a color hard copy device, a color display device, a color television camera device, The present invention relates to a color conversion device used in a color recognition device, a video editing device, or the like.

[0002]

2. Description of the Related Art Conventionally, in image processing of a monochrome image, information contained in one pixel of an image is one-dimensional information called lightness (density), and lightness conversion is so-called gamma curve conversion.
LUT (look-up table) for various non-linear curves
It was possible to convert the color in real time by writing in.
Even if the image to be handled is a color image, it is treated as three monochrome images of R (red) plane, G (green) plane, and B (blue) plane for the purpose of performing color conversion in real time, and each is independent. It was often converted by the LUT. However, in this kind of processing, the color conversion that can be handled is essentially one-dimensional processing, and R '= hR (R), G' = hG (G), B '= hB (B). Only color conversion is possible.

In color image processing, the information contained in one pixel is three-dimensional information (R, G, B), and color conversion in the original sense means three-dimensional conversion R '= fR. (R, G, B) G '= fG (R, G, B) B' = fB (R, G, B).

For example, in hard copy type color image processing, complicated color conversion such as "color belonging to a specific hue increases saturation" is required, and even in the case of color image editing processing such as video, "blue color Sometimes, there is a request for color conversion of only a complicated specific color such as "I want to convert only the background to black". These color conversions mathematically belong to the above-mentioned three-dimensional conversion because one output is a function of three inputs. However, if you try to convert these with a general-purpose table, 1
Assuming full-color image processing with 8-bit color signals 1
16 (Mbyte) memory capacity is required for conversion per color. Therefore, there is a need for hardware that can perform three-dimensional color conversion universally and in real time for arbitrary color conversion. On the other hand, color hard copy,
An example of a color signal interpolation method in which the input color space is divided into a plurality of color spaces, and a plurality of pieces of color correction information located at the vertices are selected, weighted and interpolated for the main purpose of color correction of a color scanner. (Japanese Patent Publication No. 58-16180 or U.S. Pat. No. 4,275,413). In this example, a unit cube, which is the basic solid in the three-dimensional color signal space, is set for interpolation processing, this unit cube is divided into multiple tetrahedra, and interpolation calculation is simplified from the output signals at each vertex of the tetrahedron. The idea to change is disclosed. FIG. 9 shows a configuration example of a color scanner device using this conventional example. The input RGB signal 901 is divided into an upper signal 902 and a lower signal 903 by 4 bits each, and the lower signal is inputted to the four kinds of weighting coefficient generators 905 at the same time as the magnitude relation is judged by the comparator 904. It On the other hand, the higher-order signal is subjected to address modification by the selector 907 and the adder 906, and the color conversion table memory 908 is read at different addresses four times in total, and the output data is paralleled by the weighting coefficient and the multiplier 909. The calculated value is added by the adder 910 to obtain an interpolation output 911. FIG. 10 shows a method of dividing a cube formed by a coarse representative grid point in the color space stored in the color conversion table memory into six tetrahedra, and FIG. 11 shows a tetrahedron which is a divided unit interpolation area. .. If this conventional example is used, it can be used as a general-purpose color conversion device, and non-linear free color conversion such as color change of a specific color in a color space can be performed while maintaining the gradation of an image.

[0005]

However, firstly, in the conventional technique, although the number of multiplications involved in interpolation is four, which is very small, the color conversion table memory is read sequentially. Therefore, it takes too much time to perform high-speed color conversion such as color conversion of a color moving image. Therefore, there is a problem that it is preferable to have a plurality of color conversion table memories, read them in parallel, and multiply them in parallel for higher speed.

Secondly, in the conventional technique, since the input signal is divided into upper and lower signals by 4 bits each, RGB
Since the three-dimensional color space of the entire 8-bit color signal is divided into 16 areas each having a length of 16 on each color signal axis, the color conversion table memory 9 as shown in FIG.
The modified address line input to 08 is the upper signal 902.
At first glance it looks like it's 4 bits. However, when it is actually divided into 16 areas
According to the principle of the above, there are a total of 17 division points including both ends, and in order to perform mathematically correct three-dimensional interpolation, the input address line of the color conversion table memory is 5 bits for each color to represent 17 types. is necessary. However, in this case, only 1 out of 32 kinds of numbers expressed by 5 bits.
Since only 7 are used, the waste of memory usage reaches about half, and the economy is too bad. If the memory parallel read-out type configuration as described in the first problem is to be taken, this would cause a large-scale memory waste and is impossible. So, for the purpose of parallel type configuration,
The way of thinking about the role of the memory is changed, and the address of the color conversion table memory represents not the grid points themselves but the number of the area between the grid points. In this case, since there are only 16 areas on each color signal axis, the upper signal 4 bits may be directly input to the color conversion table. However, this time, it is necessary for the color conversion table to output in parallel the four grid points that are different from each other, and it is determined which of the tetrahedrons of the six color cubes the input color signal is divided into is determined. Therefore, it is necessary to determine the four vertices to be used for the interpolation, and prepare and store the values to be used for the interpolation in the color conversion table memory for each of the six planes in advance according to the six types of cases. However, since the number 6 is not a power of 2, there is a problem in that the structure of the color conversion table memory is again wasted. Actually, the color conversion table memory is prepared with 8 planes that are the cube of 2 and only 6 planes of them are used, so that a quarter is wasted this time.

Therefore, the present invention solves the above first and second problems and parallelizes the color conversion table memories,
Moreover, while realizing mathematically correct high-speed color conversion, by further dividing one cube or rectangular parallelepiped in the three-dimensional color space into regions of powers of 2, there is no waste in the structure of the color conversion table memory. The purpose is to lose.

[0008]

In order to achieve this object, the present invention divides a three-dimensional space formed by three-color separated input color signals into a plurality of rectangular parallelepipeds, and divides the rectangular parallelepiped having an input color into two triangular prisms. In the case of a triangular prism that determines which is included in the triangular prism and the two triangular bases of the triangular prism area, the output value corresponding to the reference vertex position and the output values at two different vertex positions are compared. A color conversion table memory that stores an output first difference value and an output second difference value, which are the difference values, and a multiplier that multiplies and adds the output first difference value and the output second difference value, and addition. And a container.

[0009]

In the color conversion apparatus of the present invention, since color conversion is performed by parallel reading and multiplication of six memories, high-speed color conversion can be executed, and each color conversion table memory address is a rectangular parallelepiped whose color space is divided. Since the address of the area is expressed, mathematically correct three-dimensional interpolation is performed, and since each rectangular parallelepiped of the three-dimensional color space is divided into two triangular prisms,
In terms of the hardware configuration, the color conversion table may be prepared in two planes depending on which of the two triangular prisms the input color belongs to, and the memory configuration is a power of two, and it is possible to suppress the generation of wasted memory that is not used. ..

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The principle of an embodiment of the color conversion device of the present invention will be described below with reference to the drawings and the formulas. The input color signals are three-color separated color signals Red (hereinafter referred to as R), Green
n (hereinafter referred to as G) and Blue (hereinafter referred to as B), but these density signals may be used. The following description will proceed with R, G, and B. Next, the three-dimensional color space represented by R, G, and B is divided into respective axes as shown in FIGS. 2 and 3, and the color space is roughly divided into unit rectangular parallelepiped regions. FIG. 2 shows that when each axis is equally divided, FIG. 3 shows that the number of divisions of each axis is fine in the G direction and is coarse in the R and B directions, which is effective for human visual characteristics and has a limited memory capacity. Examples that can be used
Each unit rectangular parallelepiped has a rectangular shape in the R and B planes, but is further divided into two triangular prisms along the diagonal line of the rectangle. That is, R and B form the base of a triangle, and form a triangular prism having a main axis in the G direction. The two signals at the bottom are arbitrary, but here, the estimation of the G signal from the two signals of R and B obtains a better result than the other combinations based on the nature of the color image and the human characteristics. Report ("Study on Color Image Reproduction from Two Colors", The Institute of Image Electronics Engineers, Japan 19
91, VOL20, NO. Based on 5), R and B were treated equally as ridges of the same triangle.

FIG. 4 shows the unit rectangular parallelepiped abcd-efgh.
And a triangular prism that is a division of it. FIG. 5 shows one of the triangular prisms abc-efg divided into two. When a color point is input in this triangular prism, the corresponding output value can be interpolated from the output value at each vertex of the triangular prism. Now, in FIG. 5, the vertices a, b, c,
The output values at d, e, f, and g are (a), (b), (c),
When (d), (e), (f), and (g) are set, the output value (p) corresponding to the input color point p is from the point a to the input point p.
Interpolation can be performed in three steps as follows using ΔG, ΔR, and ΔB, which are the components of the vector ap toward G. For the sake of simplicity, it is assumed that the weighting factors such as ΔG are normalized as the maximum value = 1 here.
This can be done by dividing the actual ΔG or the like by the length of each ridge of the unit rectangular parallelepiped. In the first stage, a straight line is drawn from p in parallel with the G axis, and a triangle abc and a triangle efg are drawn.
Let the intersections with and be p ₁ and p ₂ , respectively. And in the triangle abc, the output value at the p ₁ a (p _1),

[0012]

[Equation 1]

Interpolation is performed as follows. This has the following graphical meaning. FIG. 6 is a view of the triangular prism of FIG. 5 observed from the G axis direction. From this direction, the two bottom faces of the triangular prism completely overlap, and p and p ₁ and p ₂ , a and e, b and f, c.
And g overlap. Therefore, it may be considered that this triangle is the triangle abc. The Δp which is the component of the input first difference vector ab (601) and the ΔB which is the component of the input second difference vector bc (602) of the vector ap in FIG. 5 are obtained, and in the output space, each input difference vector The output first difference value {(b)-(a)} and the output second difference value {(c)-(b)} corresponding to are weighted by ΔR and ΔB, and the first and second outputs are weighted. It is an operation of obtaining an increment and adding it to the output value (a) at a. In the present embodiment, the linear interpolation operation always uses the difference, but if (a), (b), and (c) are weighted and added without using the difference, the multiplication only increases once. , It is clear that it gives the same result.

[0014] In the second stage, the output value (p ₂₎ a first difference value output the same weighting in the p ₂ to FIG. 6 is regarded as a triangle efg {(f) - (e )}, outputs the second difference value {(G)-
(F)},

[0015]

[Equation 2]

## EQU1 ## In the third stage, (p ₁ ) and (p ₂ ) are linearly interpolated as follows on the line segment p ₁ -p _{2 in} FIG.

[0017]

[Equation 3]

Substituting (Equation 1) and (Equation 2) into (Equation 3) and rearranging

[0019]

[Equation 4]

[0020] As described above, the input point P is a triangle
The output value at p is determined when it is in the column abc-efg
It The input color point is the other triangular prism ac that divides the rectangular parallelepiped.
If it exists in d-egh, it is shown in Fig. 7 and Fig. 8.
Growls As before, from p to triangle acd and triangle
Draw a straight line parallel to the G axis at egh, and then intersect p₁, P₂Seeking
Therefore, G, R, and B axial direction components of the vector ag
Calculate ΔR and ΔB, p in the first step ₁The output value at

[0021]

[Equation 5]

In the second stage, the output value at p ₂ is

[0023]

[Equation 6]

In the third step, the output interpolation value at p is

[0025]

[Equation 7]

It is calculated as follows. (Equation 7) to (Equation 5),
Substituting (Equation 6) and rearranging,

[0027]

[Equation 8]

It becomes like this. The determination of whether the input color point p is included in the above-mentioned two types of triangular prism abc-efg or triangular prism acd-egh is performed by determining the components ΔR and Δ of the vector ap.
The determination is made based on the magnitude of B. When ΔR ≧ ΔB, triangular prism abc-efg ΔR <ΔB, triangular prism acd-egh (Equation 9).

Next, the structure of the color conversion apparatus in one embodiment of the present invention will be described with reference to FIG.

FIG. 1 is a block diagram showing the essential parts of a color conversion apparatus according to an embodiment of the present invention. In FIG. 1, the input signal (R,
Since only the part where the output signal R ′ is generated from (G, B) is shown in the figure, (R, G, B) to (R ′, G ′, B ′)
To generate, three sets of components shown in FIG. 1 are required.

In FIG. 1, the G, R, and B input spaces are divided into rectangular parallelepipeds, and further divided into triangular prisms to perform interpolation. This division into rectangular parallelepipeds expresses the G, R, and B signals in FIG. It is performed by dividing the digital signal to be processed into an upper signal and a lower signal. For example, the G, R, and B signals are 8-bit signals, and the respective higher-order signals are 3 from the top of the signal,
3 and 3 bits are used, and the lower order signals are the remaining 5, 5, and 5 bits. In this case, the three-dimensional space represented by G, R, and B is G
The 3D space is divided into 2 ³ = 8 interpolation sections in the axial direction, 2 ³ = 8 interpolation sections in the R-axis direction, and 2 ³ = 8 interpolation sections in the B-axis direction. (8) x (8) x (8)
= 512 rectangular parallelepipeds, and the lengths of the sides of each rectangular parallelepiped are 32, 32, and 32 in the G, R, and B directions, respectively.

In FIG. 1, reference numeral 107 denotes a triangular prism determining unit for outputting a triangular prism determining signal 108 from the R lower 5 bit signal 105 and the B lower 5 bit signal 106, and the input color point is determined by the magnitude determination according to (Equation 9). The determination result is output by outputting 1 or 0 as to which one of the two triangular prisms obtained by dividing the rectangular parallelepiped is included. 133 is a weight complement calculation means,
Calculate (1-ΔG). Reference numerals 110 to 115 denote color conversion table memories, which are high-order 3-bit signals of G, R, and B (total 9 bits).
Bit) and the triangular prism determination signal 108 of the triangular prism determination unit 107.
From the output value (a), the output first difference value {(b)-(a)}, and the output second difference value {(c)-
(A)}, (e), output first difference value {(f)-
(E)} and the output second difference value {(g)-(f))} are output. Reference numeral 119 denotes a multiplier, which is a color conversion table memory 119.
Difference value {(b)-(a)} and R lower 5
Multiply with the bit signal 105. 120, 126, and 127 are also multipliers similar to the multiplier 119, and output second difference value {(c)-(a)}, output first difference that is the output of each color conversion table memory 112, 114, 115. Value {(f)-(e)}, output second difference value {(g)-
(F))} is multiplied by the R lower 5 bit signal 105 or the B lower 5 bit signal 106. 121 is a multiplier 1
An adder for adding the outputs of 19 and the multiplier 120, 128
Is an adder that adds the outputs of the multiplier 126 and the multiplier 127. 122 is an adder for adding (a) which is the output of the color conversion table memory 110 and the output of the adder 121, and 129 is (e) which is the output of the color conversion table memory 113 and is added with the output of the adder 128. It is an adder that does. Reference numeral 135 denotes the output (1-Δ of the weight complement calculation means 133.
G) a multiplier for multiplying the output of the adder 122, 130
Is a multiplier for multiplying the G lower 5 bit signal by the output of the adder 129. 131 is a multiplier 135 and a multiplier 130
It is an adder that adds the outputs of and to obtain the final result R′132.

The operation of the above arrangement will be described below. In FIG. 1, as described above, the higher-order 3-bit signals of G, R, and B, which are inputs, are 101, 1 respectively.
02 and 103, each of the lower 5 bit signals is 10
4, 105 and 106 are shown. That is, in FIG. 2, the upper 3-bit signal is 0 to 7, 0 to 7, 0 from the unit rectangular parallelepiped including the input color as the position coordinates of the point a closest to the origin in the G, R, and B axial directions, respectively. It is expressed by a numerical value up to 7, and the lower 5 bits signal is point a in each unit rectangular parallelepiped.
The position of the reference point and the input color is represented by numerical values from 0 to 31, 0 to 31, and 0 to 31 in the G, R, and B axial directions. Therefore, the lower 5 bit signal is the ΔG, ΔR,
It means to express ΔB.

First, the ΔG signal is weighted complement calculation means 133.
(1-ΔG) is calculated. △ R signal and △ B
The signals 105 and 106 are input to the triangular prism determination unit 107, and 1
The bit-shaped triangular prism determination signal 108 is output. According to (Equation 9), whether the input color point is included in the two triangular prisms obtained by dividing the rectangular parallelepiped is determined by outputting 1 or 0 to determine whether the input color point is included. Now, it is assumed that the input color is included in the triangular prism abc-efg as a result of this determination. At the same time, the ΔR and ΔB signals 105 and 106 are sent to the multiplier 11
9 and 120 and multipliers 126 and 127, respectively. On the other hand, the higher-order 3-bit signal pairs 101, 102, and 10
3 and the triangular prism determination signal 108 are collectively (3 + 3 + 3 +
Assuming that 1 =) 10-bit memory address signal (109), this memory address signal 109 is sent to each color conversion table memory 110-115. Each color conversion table memory 110-115 stores in advance information of terms necessary for interpolation when one triangular prism is designated, and outputs the stored value in parallel with the memory address input. When the triangular prism abc-efg is specified as described above, the plane 0 is used among the color conversion table memories having two planes. In this case, the color conversion table memories 110, 111, and 112 are respectively defined in (Equation 4). The output value (a), the output first difference value {(b)-(a)}, and the output second difference value {(c)-(a)} are output. These are shown at 116, 117, 118 in FIG. According to the expression in [] in the first half of (Expression 4), the color conversion table memory 1
Output first difference value {(b)-(a)} 1 which is the output of 11
17 and the ΔR signal 105 are multiplied by the multiplier 119, and the output second difference value {(c) − (a)} 118 output from the color conversion table memory 112 and the ΔB signal 106 are multiplied by the multiplier 120. They are multiplied, and the two multiplication results are added by the adder 12
1 is added, and the result is the color conversion table memory 11
The output (a) 116, which is an output of 0, is added by the adder 122.

This result is multiplied by the weighting coefficient (1−ΔG) in the multiplier 135 to generate the first half term of (Equation 4). On the other hand, the plane 0 of the color conversion table memories 113, 114, and 115 is the output value (e), the output first difference value {(f)-(e)}, and the output second difference value {(g) in (Equation 4), respectively. −
(F))} is output. These are 123 and 12 in FIG.
4, 125. According to the expression in [] in the latter half of (Expression 4), the output first difference value {(f) − (e)} 124 which is the output of the color conversion table memory 114 and the R lower 5 bit signal 105 are multiplied by the multiplier 126. And output second difference value {(g) − which is the output of the color conversion table memory 115.
(F))} 125 and the B lower 5 bit signal 106 are multiplied by the multiplier 127, and the two multiplication results are added by the adder 128.
And the result is added by the adder 129 to (e) 123 which is the output of the color conversion table memory 113. The result is the G lower 5 bit signal indicated by 104.
G is multiplied by the multiplier 130, the second half term of (Equation 4) is completed, and the result of the first half term is further added by the adder 131, and the final result (R'132) of (Equation 4) is output. To be done. When the input color is included in the triangular prism acd-egh as a result of the triangular prism determination unit 108, the table memory is accessed in a state where only the most significant 1 bit of the triangular prism determination signal 108 in the memory address signal 109 is different. That is, the plane 1 of the two planes is used for all the six color conversion table memories 110 to 115 existing in parallel, at which time the output values of the table memories 110 to 115 are expressed by (Equation 8). The values become the values shown in the respective terms, and the interpolation is performed by operating so as to calculate (Equation 8).

The number of bits of each signal and the bit allocation of the higher-order signal and the lower-order signal shown in this embodiment are examples, and other values may be used.

[0037]

As described above, according to the present invention, the input color is divided into two triangular prisms when the rectangular parallelepiped is divided into two triangular prisms. A color conversion table memory that stores the output first difference value and the output second difference value, which are the difference values between the output value corresponding to the reference vertex position and the output values at two different vertex positions on the triangle base. By providing the output first difference value and the output second difference value with a multiplier and an adder, it is possible to realize an arbitrary color conversion process in real time.

Further, although the conventional linear interpolation method using tetrahedron division, which is a three-dimensional color signal interpolation method, has a large memory waste and cannot effectively utilize the memory, the present invention uses a divided solid body of triangular prisms. As a result, the three-dimensional color space cube, which has been a problem in the past, is not divided into a number of regions other than the power of 2 and memory is not wasted. Therefore, an excellent color conversion device in the real-time color image processing field can be provided. It can be realized.

[Brief description of drawings]

FIG. 1 is a block diagram of a main part of a color conversion device according to an embodiment of the present invention.

FIG. 2 is a conceptual diagram in which a three-dimensional space created by G, R, and B in the same color conversion device is evenly divided into a plurality of cubes.

FIG. 3 is a conceptual diagram in which a three-dimensional space created by G, R, and B is unevenly divided into a plurality of rectangular parallelepipeds in the same color conversion device.

FIG. 4 is a conceptual diagram in which a rectangular parallelepiped is divided into two triangular prisms in the same color conversion device.

FIG. 5 is a conceptual diagram of a triangular prism abc-efg in the same color conversion device.

FIG. 6 is a conceptual diagram in which a triangle abc-efg in the same color conversion device is observed from the G axis direction.

FIG. 7 is a conceptual diagram of a triangular prism acd-egh in the same color conversion device.

FIG. 8 is a conceptual diagram in which a triangular prism acd-egh in the same color conversion device is observed from the G axis direction.

FIG. 9 is a block connection diagram of a conventional color conversion device.

FIG. 10 is a conceptual diagram in which a color cube in the same color conversion device is divided into six tetrahedra.

FIG. 11 is a conceptual diagram showing a tetrahedron that is a unit interpolation area in the same color conversion apparatus.

[Explanation of symbols]

107 triangular prism determination unit 110-115 color conversion table memory 119, 120, 126, 127, 130, 135 multiplier 121, 122, 128, 128, 131 adder 133 weight complement calculation means

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Rika Iikawa 3-10-1 Higashisanda, Tama-ku, Kawasaki City, Kanagawa Prefecture Matsushita Giken Co., Ltd. No. 10-1 Matsushita Giken Co., Ltd.

Claims (3)

[Claims] 1. A triangular prism that divides a three-dimensional space formed by three-color separated input color signals into a plurality of rectangular parallelepipeds and determines which triangular prism the input color is included in when the rectangular parallelepiped is divided into two triangular prisms. For the determination unit and the two triangular bases of the triangular prism area, the output value corresponding to the reference vertex position and the output first difference value that is the difference value between the output values at two different vertex positions, the output second A color conversion device, comprising: a color conversion table memory that stores difference values; a multiplier that multiplies and adds the output first difference value and the output second difference value; and an adder. 2. The color conversion device according to claim 1, wherein the three-color separated input color signals are divided into different roughnesses to create a color conversion table memory. 3. The base of the triangular prism region is constituted by any two signals of the three-color separated signals.
The described color conversion device.