JPH08138030A - Method and device for data conversion - Google Patents

Abstract

PURPOSE: To reduce the interpolation error in a color converter where data converted by a table memory is subjected to interpolation to perform the color conversion. CONSTITUTION: Luminance signals R, G, and B from a scanner or a personal computer are nonlinearly converted by look-up tables 1 to 3. Intervals of lattice point addresses of table memories 4 to 7 indicated by upper bit signals RU, GU, and BU of obtained signals R', G', and B' are not uniform in the color space generated by signals R, G, and B but are longer in bright parts of signals R, G, and B and are shorter in dark parts. Thus, the precision of interpolation in parts where these intervals are shorter is relatively improved, and the interpolation error is reduced as the whole.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data conversion device and a data conversion method, and more particularly to a color conversion process for converting three signal values representing color signals into output signal values which are different color signals. It is a thing.

[0002]

2. Description of the Related Art Conventionally, as a color conversion device of this kind, one shown in FIG. 10 has been known. This device uses three primary colors to print out the data obtained by color-separating an original image into R, G, and B colors and the image data artificially synthesized on a computer with an inkjet printer or the like. It is for converting into a certain C, M, Y signal.

In FIG. 8, signal processing circuits 81, 82 and 83 generate output color signals C, M and Y from input three primary color signals R, G and B, respectively. For example, the signal processing circuit 8
In so-called masking calculation as shown in the following equations, the output color signals C, M and Y are respectively obtained at 1, 82 and 83.

[0004]

## EQU1 ## C = A ₁₁ × R + A ₁₂ × G + A ₁₃ × B M = A ₂₁ × R + A ₂₂ × G + A ₂₃ × BY Y = A ₃₁ × R + A ₃₂ × G + A ₃₃ × B where A _ij is the characteristic of the output device. It is a coefficient determined according to the above.

However, a circuit for performing such an operation has a problem that it is relatively large-scaled, and therefore, an alternative to this is that a table using the following table is conventionally known.

That is, instead of performing the sum-of-products operation by the signal processing circuit as described above, the operation result is stored in advance in the table memory in correspondence with each value of R, G, B, and input. It is a device for reading out the calculation result of the R, G, B signal values to be output from the table memory and outputting it.
However, in the case of this device, if the input signal is represented by 8 bits for each color, 2 ²⁴ addresses, that is, a storage area of 16 million or more is required, which is not realistic.

Therefore, as still another conventional example, FIG.
Those shown in are also known. Here, the input signals R, G, B are divided into upper bit data 91 and lower bit data 92, only the operation result for the upper bits is stored in the table memory 93, and each upper bit from the table memory 93 is stored. Output value 94 corresponding to the data
Can be linearly interpolated by the interpolating circuit 95 based on the lower bit data to obtain the output signal 96. According to such a configuration, the storage area of the table memory needs to have only the number of addresses determined by the number of upper bits, and for example, if the number of upper bits of each signal is 3 bits for each color. 2 ⁹ addresses, ie 51
Only two storage areas are required, and the amount of storage area can be significantly reduced as compared with the case of inputting the above 8-bit data.

[0008]

However, in the color conversion in which the table memory and the interpolation circuit are combined as described above, there is an advantage that a complicated non-linear conversion can be realized relatively easily, but there is a difference in the degree. However, the problem that interpolation error occurs (conversion accuracy decreases) is inevitable.

In order to reduce the interpolation error, it is generally most effective to increase the number of upper bits by division, but if the number of bits is increased by 1 bit, the input signals R, G and B are obtained. In the case of 3 inputs, the storage area of the table memory is required to be 2 ³ times, that is, 8 times.

As described above, there is a trade-off relationship between the conversion accuracy and the amount of storage area, and in order to reduce the memory amount, it is unavoidable to set the number of bits while sacrificing the conversion accuracy to some extent. there were.

The present invention has been made in view of the above problems, and an object thereof is to obtain a sufficient conversion accuracy by performing linear interpolation in a virtual data space in which an interpolation error is unlikely to occur. A data conversion device and a data conversion method are provided.

[0012]

Therefore, according to the present invention,
In a data conversion device for converting input data into another data, first conversion means for performing non-linear conversion on the input data,
The table memory means for outputting the function value data corresponding to the grid point address data generated based on the data converted by the first converting means, and the function value data output from the table memory means are the first An interpolation means for interpolating the weighting coefficient data generated based on the data converted by the converting means and outputting another data as the interpolation result.

Further, in the data conversion method for converting the input data into another data, the first non-linear conversion is performed on the input data, and the grid point address data generated based on the data converted by the conversion is stored in the table memory. To output the function value data corresponding to the grid point address data, and the function value data output from the table memory is generated based on the data converted by the first nonlinear conversion. Interpolated by the data,
A second non-linear conversion is performed on the data of the interpolation result, and another data is output, and each step is provided.

[0014]

According to the above construction, the non-linear conversion of the first conversion means makes it possible to make the intervals of the grid point address data relative to the input data space formed by the input data nonuniform. It is possible to control and improve the interpolation accuracy determined by the interval of the point address data.

[0015]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a block diagram showing a color conversion circuit according to an embodiment of the present invention.

In FIG. 1, reference numerals 1 to 3 perform non-linear conversion on R, G, and B signals, respectively, as will be described later, and R ',
It is a look-up table that outputs G ', B'signals, and these outputs R', G ', B'form a virtual color space. Reference numerals 4 to 7 are table memories for reading out grid point data in the interpolation calculation by using the upper bits of the converted signals R ', G', B'as addresses, and 8 is a table memory of the converted signals R ', G', B '. It is an arithmetic unit that calculates a weighting coefficient from the lower bits as a coefficient of interpolation calculation. An interpolation calculation circuit 9 linearly interpolates the grid point data from the table memories 4 to 7 by the weighting coefficient from the calculator 8 based on the lower bits. Reference numerals 10 to 13 are lookup tables for performing a non-linear conversion as will be described later on the interpolation calculation result.

FIG. 2 is a schematic diagram conceptually showing the non-linear conversion in the lookup table 1.

The look-up table 1 converts the signal R into a signal R'constituting a virtual color space with few errors in the above linear interpolation. Specifically, as shown in FIG.
Similarly, the signal R indicating the density value of any of the 6 gradations is set to 256.
It is to be converted into a signal R'representing any density value of gradation, and the intervals of the addresses (hereinafter, referred to as grid point addresses) of the table memories 4 to 7 indicated by the upper bits of the signal R ', The conversion is performed so that the bright portion is relatively large (sparsely) and the dark portion is relatively small (dense) with respect to the space spanned by the signal R. That is, the intervals between the lattice point addresses represented by the upper bits of the signal R ′ are uniform,
The space formed by the signal R relatively changes according to the lightness and darkness of the signal R, and as a result, the interpolation calculation with higher accuracy and less error is performed in the portion where the intervals of the grid point addresses are relatively dense. Will be able to do. As the calculation for such conversion, as can be understood from the curve shown in FIG. 2, for example, logarithm or 1 / nth power calculation can be performed.

The same conversion is performed in the look-up tables 2 and 3, so that the signals G and B are converted into signals G'and B '.

Signals R ', G', B'converted in Tables 1, 2 and 3 are represented by the following equation (1): upper bit data RU, GU, BU and lower bit data R.
It is divided into L, GL and BL. Where R ', G',
The number of bits of the B ′ signal is Ni, and the number of upper bits is N
U and the number of lower bits are NL. In this case, obviously Ni = NU + NL.

[0022]

R ′ = RU × 2 ^NL + RL G ′ = GU × 2 ^NL + GL B ′ = BU × 2 ^NL + BL (1) When formula (1) is arranged for each upper bit data,

[0023]

RU = R ′ / 2 ^NL GU = G ′ / 2 ^NL BU = B ′ / 2 ^{NL Further} , when rearranging each lower bit data,

[0024]

RL = R′−RU × 2 ^NL GL = G′−GU × 2 ^NL BL = B′−BU × 2 ^NL Table memories 4, 5, 6 and 7 have R ′, G ′,
Output values for all combinations of upper bits of B ', that is, grid point data corresponding to grid points (addresses) when performing interpolation calculation are written. If we write this as C1 (i, j, k),

[0025]

C1 = (i, j, k) = F (i × 2 ^NL , j × 2 ^NL , k × 2 ^NL ) (2) Here, F is a color conversion function and interpolation This is the target function. Further, i, j, and k are 0 to 2 respectively.
It is an integer up to ^NU .

The storage capacity required in each of the table memories 4 to 7 is (2 ^NU +1) ³ addresses, for example, 729 addresses when NU = 3, as is clear from the equation (2).

Looking at the relation of the above equation (2) in the three-dimensional space formed by R ', G', B ', as shown in FIG. 3, the input signals RU, BU, GU are at 2 ^NL step intervals. The function value C1 is written on the grid points corresponding to the grid points arranged evenly. The virtual color space of R ', G', B'shown in FIG. 3 is converted into R, G, B before conversion.
When viewed from the three-dimensional color space, the lattice points are distorted in the respective axial directions. That is, by the conversion shown in FIG. 2, the grid point spacing shown in FIG. 3 becomes smaller in a region where an error in interpolation is likely to occur in the R, G, and B color spaces or in a region where visual color discrimination is severe. It can be said that

Next, the method of interpolating the lattice point data in the interpolating circuit 9 will be described with reference to FIG.

The interpolation shown in FIG. 4 is a so-called 8-point interpolation, and includes a cube inside which the points indicated by the signals R ', G,', and B'are contained, and each of which has eight vertices corresponding to a lattice point. The interpolation space is used. Specifically, from each of the table memories 4 to 7, the grid point data C11 to C18 of 8 neighborhoods including the points indicated by the signals R ', G,', and B'inside are read out. The grid point data C11 to C18 are R ', G,',
The upper bit data of B'can be applied to the equation (2) and expressed as follows.

[0030]

## EQU00006 ## C11 = C1 (RU, GU, BU) C12 = C1 (RU + 1, GU, BU) C13 = C1 (RU, GU + 1, BU) C14 = C1 (RU + 1, GU + 1, BU) C15 = C1 (RU, GU, BU + 1) C16 = C1 (RU + 1, GU, BU + 1) C17 = C1 (RU, GU + 1, BU + 1) C18 = C1 (RU + 1, GU + 1, BU + 1) (3) These grid point data C11 to C18 have a table memory 1 In the case of individual pieces, all of them cannot be read out at the same time, so they are read out in time series. Alternatively, eight table memories having the same content may be prepared and the above data may be read simultaneously.

Each of these eight grid point data and the distance D between each grid point and the point indicated by the input signals R ', G', B '.
Since 11 to D18 can be calculated by the calculator 8 based on the lower bit data of the input signals R ', G', B ', the grid point data Cij is weighted by the calculated distance Dji (weighting coefficient). The output signals C ', M', Y ', K'can be obtained by the interpolation calculation shown in the following equation (4).

[0032]

[Equation 7] C '= (D11 x C11 + D12 x C12 + D13 x C13 + D14 x C14 + D15 x C15 + D16 x C16 + D17 x C17 + D18 x C18) ÷ (D11 + D12 + D13 + D14 + D15 + D16 + D17 + D18 + D18 ... D18). The signals C ', M', Y'and K'are converted by the look-up tables 10 to 13 into the signals C, M, Y and K for actual printing.

FIG. 5 is a schematic diagram conceptually showing this conversion.

That is, it is desirable that the signal used by the printer or the like be one whose density indicated by it changes linearly, but the signals C ', M', and
Generally, the linearity of Y'and K'is not maintained due to the effect of conversion by the lookup tables 1 to 3. Therefore, the lookup tables 10 to 13 in FIG.
The linearity is corrected by performing the conversion shown in.
Specifically, the low-density portion where the linearity is easily damaged is densely converted, and the high-density portion where the linearity is hard to be damaged is converted to sparse.

According to the embodiment described above, the intervals between the grid point addresses input to the table memory are determined by the input signals R,
It is not uniform in the space spanned by G and B, and is dense or sparse depending on the easiness of occurrence of an interpolation error or the conspicuousness of the error. Therefore, the error caused by interpolation can be reduced as a whole. it can.

FIG. 6 is a block diagram showing a color conversion circuit according to a modification of the above embodiment.

In this modification, so-called 5-point interpolation is performed. In FIG. 6, 201 to 203 are look-up tables that perform a non-linear conversion to form a virtual color space, and 204 to 207 are R ′ and R ′, respectively.
A color processing circuit for converting G ', B'into C', M ', Y', K ', and 208 to 211 are look-up tables for performing non-linear conversion to correct linearity as in the above embodiment. The color processing circuit 204 uses the first table memory 21.
2, second table memory 213 and interpolation circuit 214
Have. Although not shown, the color processing circuits 205 to 207
Has the same configuration.

The data represented by the equation (2) of the above embodiment, the signal R ', G', when viewed in the three-dimensional space of the B ', as shown in FIG. 3, 2 ^NL step interval for the input signal In this state, the grid point data C1 is written on the grid points that are evenly arrayed in 1., and this state is stored in the first table memory 212.

On the other hand, as shown in FIG. 7, the second table memory 213 stores grid point data C2 of points that fill the spaces between these grid points. That is, with a certain offset value as δ, the function value C2 as in the following equation is stored.

[0040]

C2 (i, j, k) = F (i × 2 ^NL + δ, j × 2 ^NL + δ, k × 2 ^NL + δ) where i, j, and k are integers from 0 to 2 ^NU −1, respectively. (5) The value of δ needs to be set so that the grid point data C2 is inside the grid point in the equation (2), but for that purpose, the following condition may be satisfied.

[0041]

[Equation 9] 0 <δ <2 ^NL (6) However, the following equation is set as the value of δ in order to simplify the selection of lattice points described later.

[0042]

## ^EQU10 ## δ = 2 ^NL / 2 (7) The storage area required for the second table memory 213 is (5)
As is clear from the equation, there are (2 ^NU ) ³ addresses, for example, 512 addresses when NU = 3.

Next, a data interpolation method in the case of obtaining an output signal from the contents of the table memory thus constructed will be described with reference to FIG.

The signals R ', G', B'are divided into upper bit data and lower bit data as shown in the equation (1). From the first table memory 212, grid point data C11 to C14 in the four neighborhoods closest to the values of the input signals R ', G', B'are read out. This is the grid point data of the four vertices of the surface closest to the input R ', G', B'signals, as indicated by the white circles in FIG. As the surface closest to the input signal, one of the 6 surfaces surrounding the input signal will be selected, but the distance between the input signal value and the 6 surfaces should be calculated and the one with the smallest value should be selected. To

When the surface closest to the input signal is the surface shown in FIG. 8, C11 to C14 can be expressed as follows by applying the higher order bit data of R ', G', and B'to the equation (2).

[0046]

C11 = C1 (RU, GU, BU) C12 = C1 (RU + 1, GU, BU) C13 = C1 (RU, GU + 1, BU) C14 = C1 (RU + 1, GU + 1, BU) (8) C11 ... Since all of C14 cannot be read at the same time, they may be read in time series, or four of the same table memories may be prepared and read in parallel.

Next, the grid point data C21 closest to the input R ', G', and B'signals from the second table memory 213.
Read out. This is indicated by a black circle in FIG. 8 and can be expressed as follows by applying it to the equation (5).

[0048]

[Equation 12] C21 = C2 (RU, GU, BU) (9) A total of 5 grid point data are obtained by the above procedure, but δ
Is defined by the equation (7), the values of the input signals R ', G', and B'become inside the quadrangular pyramid formed by these five points, as shown in FIG. Therefore, the distance D11 to D1 between each of these five points and the input signal value
If D4 and D21 are obtained, the output signal C ′ can be obtained by the interpolation calculation as shown in the following equation in which the grid point data Cij is weighted at the obtained distance Dij.

[0049]

[Equation 13] C '= (D11 x C11 + D12 x C12 + D13 x C13 + D14 x C14 + D21 x C21) / (D11 + D12 + D13 + D14 + D15 + D21) (10)

[0050]

As described above, according to the present invention,
By the non-linear conversion of the first conversion means, the intervals of the grid point address data relative to the input data space created by the input data can be made non-uniform, so that the interpolation accuracy determined by the intervals of the grid point address data can be obtained. It is possible to control and improve this.

As a result, for example, it is possible to perform color reproduction with less error and high accuracy at low cost.

[Brief description of drawings]

FIG. 1 is a block diagram showing a color conversion circuit according to an embodiment of the present invention.

FIG. 2 is a diagram conceptually showing conversion by a look-up table used in the above circuit.

FIG. 3 is an explanatory diagram illustrating an interpolation calculation in the above circuit.

FIG. 4 is a schematic diagram showing an interpolation space of the interpolation calculation.

FIG. 5 is a diagram conceptually showing conversion by another lookup table used in the above circuit.

FIG. 6 is a block diagram showing a color conversion circuit according to a modification of the above embodiment.

FIG. 7 is an explanatory diagram illustrating an interpolation calculation in the color conversion circuit of the modified example.

FIG. 8 is an explanatory diagram illustrating an interpolation calculation in the color conversion circuit according to the modified example.

9A and 9B are schematic diagrams showing an interpolation space in the modification.

FIG. 10 is a block diagram showing a conventional example of a color conversion circuit.

FIG. 11 is a block diagram showing another conventional example of a color conversion circuit.

[Explanation of symbols]

1, 2, 3, 201, 202, 203 Look-up table 4, 5, 6, 7, 212, 213 Table memory 8 Weight coefficient generation calculator 9, 214 Interpolation circuit 10, 11, 12, 13, 208, 209, 210, 2
11 Look-up table 204, 205, 206, 207 Color processing circuit

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical indication H04N 1/40 D

Claims (6)

[Claims] 1. A data conversion device for converting input data into another data, first conversion means for performing non-linear conversion on the input data, and grid points generated based on the data converted by the first conversion means. Table memory means for outputting the function value data corresponding to the address data, and function value data output from the table memory means,
A data conversion device comprising: an interpolation unit that interpolates with weighting factor data generated based on the data converted by the first conversion unit and outputs another data as the interpolation result. 2. A data conversion device for converting input data into another data, first conversion means for performing non-linear conversion on the input data, and grid points generated based on the data converted by the first conversion means. Table memory means for outputting the function value data corresponding to the address data, and function value data output from the table memory means,
Interpolating means for interpolating with the weighting factor data generated based on the data converted by the first converting means and outputting the data as the interpolation result, and non-linear conversion for the data output by the interpolating means, and another data And a second conversion means for outputting as a data conversion device. 3. The first conversion means performs a non-linear conversion in which an interval of the grid point address data generated from the conversion data is not uniform with respect to an input data space formed by the input data. Item 1. The data conversion device according to item 1 or 2. 4. The first conversion means performs a non-linear conversion in which an interval of the grid point address data generated from the conversion data is not uniform with respect to an input data space created by the input data, and the second conversion. The data conversion device according to claim 2, wherein the means performs a conversion for correcting the linearity of the output different data. 5. The data conversion apparatus according to claim 1, wherein the input data is three digital color separation signals. 6. A data conversion method for converting input data into another data, wherein a first non-linear conversion is performed on the input data, and grid point address data generated based on the converted data is stored in a table memory. To output the function value data corresponding to the grid point address data, and the function value data output from the table memory is generated based on the data converted by the first nonlinear conversion. A data conversion method comprising the steps of: interpolating with data, performing a second non-linear conversion on the interpolation result data, and outputting another data.