JP2010193474A - Device and method for multi-dimensional interpolation, and computer program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To easily materialize a rapid interpolational process such as conversion in color space and the like in as small circuit structure as possible. <P>SOLUTION: An N-dimensional digital image signal 205 is divided to a higher bit signal 212 and a lower bit signal 214 by means of a data divider 210. A referential value corresponding to all the combination of the divided higher bit signal is divisionally stored without overlapping into 2<SP>N-1</SP>pieces of submemories (multi-dimensional LUT 260), and at the same time (N+1) pieces of referential values necessary for the interpolational process are simultaneously read out by a referential-value reader 250. As a result, the interpolational process can be rapidly performed, and an address can be produced without divider for accessing to the submemory. Therefore, the interpolational process at high speed such as the conversion in color space and the like can be materialized in as small circuit structure as possible. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a multidimensional interpolation device, a multidimensional interpolation method, and a computer program, and is particularly suitable for use in interpolating an N-dimensional vector signal.

  In recent years, scanners, video cameras, and the like have become widespread as input devices. On the other hand, as an output device, various color printers using an ink jet, dye thermal sublimation type, or electrophotographic method are widely used. Each of these color input devices and output devices has a unique color space. For example, when a color image obtained by a certain scanner is transferred to another color printer as it is and printed, the color of the printed color image is Is unlikely to match the color of the original color image read by the scanner.

  In order to solve the problem of color reproducibility between devices for such color images, a process of converting the color space of the input device to the color space of the output device (called "color space conversion") is required. In order to improve the color reproduction performance of the input device and the output device, a color space conversion function is installed in the input device and the output device.

  Here, the color space conversion specifically refers to an entire series of image processing such as input γ correction, luminance density conversion, masking, black generation, UCR, output γ correction, or a certain process in the image processing. In general, the digital image signals of three colors (for example, three colors of red, blue, and green, hereinafter abbreviated as “RGB”) of the input device are simultaneously referred to, and the three colors (for example, cyan) of the output device are simultaneously referred to. , Magenta and yellow (hereinafter abbreviated as “CMY”) or 4 colors (for example, cyan, magenta, yellow and black, hereinafter abbreviated as “CMYK”). There are many cases. Further, in the case of an electrophotographic copying machine, the engine characteristics of the printer change with the operating time, so that periodic calibration is required. In such a case, conversion from four colors (for example, “CMYK”) of the input device to four colors (for example, “CMYK”) of the output device is also necessary.

  As a first conventional technique for realizing the color space conversion method as described above, a 3D-LUT (three-dimensional lookup table) in which the result of color conversion is stored is divided into eight memories and stored. A technique is disclosed in which a three-dimensional cube [trilinear] interpolation method or a three-dimensional tetrahedral interpolation method is executed at high speed by reading reference values in parallel from this 3D-LUT (see Patent Document 1).

In such a technique, the reference values corresponding to the vertices of the unit hypercube used for interpolation are all stored in different memories so that there is no access contention when reading the reference values from the memory. ing. Further, the 3D-LUT is equally divided and stored in 8 memories without overlapping (hereinafter, the multidimensional LUT is divided into a plurality of memories in order to read the reference values from the multidimensional LUT in parallel). When stored, these multiple memories will be called "sub-memory"). In the first prior art, the description is limited to the three-dimensional interpolation method. However, when multidimensional interpolation is performed using the first prior art, the multidimensional LUT is changed to 2 ^N sub-intervals. What is necessary is just to divide and memorize | store in memory.

Further, a second prior art is disclosed in which a 3D-LUT is divided and stored in four sub-memory without overlapping, and a reference value is read out in parallel to execute a three-dimensional tetrahedral interpolation at high speed. (See Patent Document 2).
In such a technique, in the case of the three-dimensional tetrahedral interpolation method, four reference values corresponding to the vertices of the selected tetrahedron are read in parallel so that the interpolation calculation can be speeded up. Also in the second prior art, as in the first prior art described above, the 3D-LUT is equally divided and stored in four sub-memory without overlapping, and is referred to from each sub-memory. There is no contention for access when reading a value.
Further, the LUT in the second prior art is divided into multi-dimensions. Therefore, when an N-dimensional input supertetrahedral (N + 1) plane interpolation method is performed, a multidimensional (N-dimensional) LUT can be divided and stored in (N + 1) sub-memory.
Thus, the second prior art is superior to the first prior art described above in that the number of sub-memory used is small.

U.S. Pat. No. 4,837,722 Japanese Patent Laid-Open No. 10-307911

In the first prior art described above, N reference values corresponding to the vertices of the unit super-solid are stored in N independent sub memories, respectively, and the multidimensional LUT is divided into 2 ^N pieces. Therefore, it has the advantage that it can be realized even by using an interpolation method other than the multidimensional tetrahedral interpolation method (for example, multidimensional solid [trilinear] interpolation method, multidimensional triangular prism [prism] interpolation method, etc.). ing.
However, when it is assumed that the multidimensional tetrahedral interpolation method is adopted, in order to read out (N + 1) reference values necessary for the interpolation operation from the sub memory in parallel, a memory peripheral circuit such as a memory interface is used. There is a problem that it is necessary to have 2 ^N pieces, and there are many surplus circuits.

In the second prior art described above, the multi-dimensional (N-dimensional) LUT is divided into a minimum number ((N + 1)) sub-memory by storing only the multi-dimensional tetrahedral interpolation method. is doing. Therefore, the memory peripheral circuit has an advantage that it is an optimal scale.
However, when N is not (power of 2−1), the number of sub-memory is not power of 2. For example, when N is 4, the number of sub memories is 5. When the second prior art is implemented as hardware, it cannot be handled in binary. Therefore, when N is not (power of 2−1), a divider is required when generating an address for accessing the sub memory, and there arises a problem that the address cannot be generated only by a shift operation or a bit mask. In general, the memory is commercialized with the number of words raised to the power of 2. Therefore, when the sub-memory is configured using the memory distributed in the market, the second conventional technique has the number of words. The problem arises that extra memory must be employed.

  The present invention has been made in view of the above problems, and it is possible to easily implement interpolation operations such as color space conversion at high speed with a circuit size as small as possible. The purpose is to.

The multidimensional interpolation apparatus of the present invention is a multidimensional interpolation apparatus that outputs a scalar signal from an input N-dimensional vector signal (N is an integer of 4 or more), and each component of the input N-dimensional vector signal value is obtained. Dividing means for dividing the high-order bits and low-order bits, memory means for storing a reference value corresponding to the combination of the data of the high-order bits divided by the dividing means, reading means for reading the reference value, and the reading Interpolation means for interpolating the input N-dimensional vector signal based on the reference value read by the means and the lower bit data divided by the dividing means, and the memory means The reference value is divided and stored in 2 ^N-1 sub-memory.
The multi-dimensional interpolation method of the present invention is a multi-dimensional interpolation method for outputting a scalar signal from an input N-dimensional vector signal (N is an integer of 4 or more), wherein the dividing means calculates the input N-dimensional vector signal value. A division step of dividing each component into upper bits and lower bits is performed, and a reading unit performs a reading step of reading a reference value corresponding to the combination of data of the upper bits divided by the division step, and an interpolation operation Means performs an interpolation operation step of performing an interpolation operation on the input N-dimensional vector signal based on the reference value read in the read step and the lower bit data divided in the division step; The reference value is divided and stored in 2 ^N-1 sub-memory.
A computer program according to the present invention is a computer program for causing a computer to output a scalar signal from an input N-dimensional vector signal (N is an integer of 4 or more), and for the input N-dimensional vector signal value. A division step for dividing each component into upper bits and lower bits, a read step for reading a reference value corresponding to a combination of the upper bit data divided by the division step, and a read step read Based on the reference value and the lower bit data divided by the dividing step, the computer executes an interpolation operation step for performing an interpolation operation on the input N-dimensional vector signal, and the reference value is 2 ^N− and wherein the stored divided into ^one sub-memory.

According to the present invention, when performing an interpolation calculation of an N-dimensional hypertetrahedron (N + 1 monohedron), (N + 1) reference values necessary for the interpolation calculation can be read in parallel, and the reference value Since it is possible to reduce the number of memories storing 2 ^N, it is possible to realize interpolation operation at high speed with a small circuit scale.

1 is a block diagram illustrating an example of an overall configuration of a color copying machine according to a first embodiment of this invention. 1 is a block diagram illustrating an example of a circuit configuration of a multidimensional interpolation device according to a first embodiment of this invention. It is a figure which shows the 1st Embodiment of this invention and shows the result of having classified each vertex of the unit solid from a reference point to a diagonal point using the four-dimensional Manhattan distance. FIG. 3 is a block diagram illustrating an example of a circuit configuration of a reference value reading unit and a multidimensional LUT according to the first embodiment of this invention. 1 is a block diagram illustrating an example of a configuration of an address generation circuit according to a first embodiment of this invention. It is a figure which shows the 1st Embodiment of this invention and demonstrates an example of the method in which the bit operation part operates the bit of an N-dimensional reference coordinate signal. FIG. 3 is a diagram illustrating an example of a configuration of a sub memory bank generation circuit according to the first embodiment of this invention. FIG. 3 is a diagram illustrating an example of a configuration of a sub memory address generation circuit according to the first embodiment of this invention. FIG. 6 is a diagram illustrating the first embodiment of the present invention, in which the positions of reference values corresponding to sub memory bank IDs are marked in ascending order of addresses. FIG. 10 is a diagram illustrating an example of a configuration of a generalized sub memory bank generation circuit according to the second embodiment of this invention. FIG. 10 is a diagram illustrating an example of a configuration of a generalized sub memory address generation circuit according to the second embodiment of this invention. It is a figure which shows the 3rd Embodiment of this invention and shows an example of a structure of the multidimensional LUT with a cache mechanism. FIG. 10 is a functional block diagram illustrating a case where a 3D interpolation operation is performed using a 3D-LUT according to the embodiment of this invention. It is a figure which shows embodiment of this invention and shows the state which divided | segmented the three-dimensional input color space (RGB space) into the unit solid. It is a figure which shows embodiment of this invention and demonstrates the selection method of the ridgeline in a four-dimensional supertetrahedral interpolation method.

  In an embodiment of the present invention described below, as a method for realizing color space conversion, the result of color space conversion is stored in advance in a memory as a look-up table (LUT), and an input digital image signal In contrast, the color space is converted using the LUT, and the conversion result is output.

  In this color space conversion method using the LUT, an interpolation operation is also used in order to reduce the memory of the LUT. Color space conversion for a three-color input digital image signal is implemented using a three-dimensional interpolation operation. Color space conversion for a four-color input digital image signal is realized using four-dimensional interpolation.

Therefore, first, a color space conversion method using a three-dimensional lookup table (3D-LUT) and a three-dimensional interpolation operation for a three-color input digital image signal will be described.
The input digital image signal (R, G, B) is separated into upper bits and lower bits. The upper bits are used when extracting a plurality of reference values necessary for the interpolation calculation performed using the 3D-LUT. The lower bits are used for the interpolation calculation as a weighting factor g. Then, an interpolation value is calculated by a product-sum operation of the weight coefficient g and the reference value extracted from the 3D-LUT.

  FIG. 13 shows a state in which a three-dimensional input color space (RGB space) is divided into unit solids by dividing the color space (RGB space) by a limited number in each axial direction. The color data after color space conversion at the vertex of this unit solid is stored in the 3D-LUT as a reference value. Considering the high-order bits of the three digital image signals as coordinates in the color space, a unit solid to be used for the interpolation calculation is selected (for example, the unit solid 1301 shown in FIG. 13), and the reference value corresponding to the vertex of the unit solid is interpolated. Used for.

As a three-dimensional interpolation calculation method used at this time, there is a tetrahedral interpolation method.
As shown in FIGS. 14B to 14G, this interpolation method divides a unit solid (for example, unit solid 1301 shown in FIG. 13) into six tetrahedrons (hereinafter, divided six tetrahedrons). Are respectively referred to as Type 0 to Type 5), and interpolation calculation is performed using the following (Expression 1) to (Expression 6) depending on which tetrahedron the input coordinate belongs to. In the following (Expression 1) to (Expression 6), the reference values corresponding to the vertices of the unit solid shown in FIG. 14A are P0 to P7, respectively, and the weighting factors g are ΔR, ΔG, and ΔB. . Further, which tetrahedron is selected from the tetrahedrons of Type 0 to Type 5 is determined by the magnitude relationship of these weighting factors ΔR, ΔG, ΔB.

When Type 0 (ΔR>ΔG> ΔB)
X = P0 + (P1−P0) × ΔR + (P3−P0) × ΔG + (P7−P0) × ΔB (1)
When Type1 (ΔR>ΔB> ΔG)
X = P0 + (P1−P0) × ΔR + (P7−P0) × ΔG + (P5−P0) × ΔB (Expression 2)
When Type2 (ΔG>ΔR> ΔB)
X = P0 + (P3-P0) × ΔR + (P2−P0) × ΔG + (P7−P0) × ΔB (3)
When Type3 (ΔG>ΔB> ΔR)
X = P0 + (P7−P0) × ΔR + (P2−P0) × ΔG + (P6−P0) × ΔB (Expression 4)
For Type 4 (ΔB>ΔR> ΔG)
X = P0 + (P5−P0) × ΔR + (P7−P0) × ΔG + (P4−P0) × ΔB (Expression 5)
When Type5 (ΔB>ΔG> ΔR)
X = P0 + (P7−P0) × ΔR + (P6−P0) × ΔG + (P4−P0) × ΔB (Expression 6)

It is also possible to make the three-dimensional tetrahedral interpolation method described so far multi-dimensional. The multidimensional hypertetrahedral interpolation method will be described below.
When the space fixed to the three-dimensional orthogonal coordinates shown in FIG. 13 is multidimensionalized into a space fixed to the N-dimensional orthogonal coordinates, the unit solid shown in FIG. Multi-dimensional). Then, the tetrahedron used in the three-dimensional tetrahedron interpolation method is multidimensionalized into an (N + 1) plane called a supertetrahedron by this multidimensionalization.

  This supertetrahedron is formed by a line connecting a reference point and a diagonal point of a unit hypersolid and N ridge lines that are orthogonal to and connected to the supersolid. Then, one unit super solid is divided into N! Individual supertetrahedrons can be created. The N ridge lines selected in the process of cutting out the supertetrahedron from the unit hypersolid are perpendicular to each other and parallel to the respective axial directions of the N-dimensional orthogonal coordinates. For this reason, a value obtained by adding a linear sum of output displacement amounts in the respective axis directions to the reference point output value is an output of the multidimensional interpolation calculation. Therefore, the interpolation value is calculated by multiplying the difference value of the reference values at the two end points of the selected ridge line by the distance from the reference point on each ridge line, and accumulating the multiplied value to the reference value of the reference point. Yes.

Next, taking a four-dimensional space (X ₀ , X ₁ , X ₂ , X ₃ ) as an example, a method for selecting edges in the four-dimensional supertetrahedral interpolation method will be described. In FIG. 15, the input coordinates are (X ₀ <i> + ΔX ₀ , X ₁ <j> + ΔX ₁ , X ₂ <k> + ΔX ₂ , X ₃ <l> + ΔX ₃ ). The magnitude relationship of the difference (ΔX ₀ , ΔX ₁ , ΔX ₂ , ΔX ₃ ) from the reference point coordinates (X ₀ <i>, X ₁ <j>, X ₂ <k>, X ₃ <l>) is { In the case of ΔX ₀ > ΔX ₁ > ΔX ₂ > ΔX ₃ }, a method of selecting a ridge line is shown.

First, as shown in FIG. 15B, a ridge line parallel to the X ₀ axis corresponding to ΔX ₀ having the largest transition amount between the input coordinates and the reference point coordinates and having a reference point as an end point, that is, a point (X ₀ <i>, X ₁ <j>, X ₂ <k>, X ₃ <l>) and point (X ₀ <i + 1>, X ₁ <j>, X ₂ <k>, X ₃ <l >) Is selected at both end points.
Next, as shown in FIG. 15C, the amount of change between the input coordinates and the reference point coordinates is parallel to the X ₁ axis corresponding to ΔX ₁ which is the second largest, and the point (X ₀ <i + 1> , X ₁ <j>, X ₂ <k>, X ₃ <l>), that is, an edge (X ₀ <i + 1>, X ₁ <j>, X ₂ <k>, X ₃ <l>) and a point (X ₀ <i + 1>, X ₁ <j + 1>, X ₂ <k>, X ₃ <l>) at both end points are selected.

Next, FIG. 15 (d), the parallel to the X ₂ axis corresponding to [Delta] X ₂ displacement amount is large in the third and the input coordinates and the reference point coordinates and the point (X ₀ <i + 1> , X ₁ <j + 1>, X ₂ <k>, X ₃ <l>), that is, a point (X ₀ <i + 1>, X ₁ <j + 1>, X ₂ <k>, X ₃ <l>) and a point (X ₀ <i + 1>, X ₁ <j + 1>, X ₂ <k + 1>, X ₃ <l>) Selected.
Finally, as shown in FIG. 15E, the point (X ₀ <i + 1>, X is parallel to the X ₃ axis corresponding to ΔX ₃ with the smallest amount of transition between the input coordinates and the reference point coordinates. ₁ <j + 1>, X ₂ <k + 1>, X ₃ <l>) at the end, that is, the point (X ₀ <i + 1>, X ₁ <j + 1>, X ₂ < k + 1>, X ₃ <l>) and a point (X ₀ <i + 1>, X ₁ <j + 1>, X ₂ <k + 1>, X ₃ <l + 1>) Is selected.

Finally, the points (X ₀ <i>, X ₁ <j>, X ₂ <k>, X ₃ <l>), (X ₀ <i + 1>, X ₁ <j>, X ₂ <k >, X ₃ <l>), (X ₀ <i + 1>, X ₁ <j + 1>, X ₂ <k>, X ₃ <l>), (X ₀ <i + 1>, X ₁ <j + 1>, X ₂ <k + 1>, X ₃ <l>), (X ₀ <i + 1>, X ₁ <j + 1>, X ₂ <k + 1>, X ₃ <l A supertetrahedron (pentahedron) composed of five vertices of (+1>) is determined. At this time, the point (X ₀ <i + 1>, X ₁ <j + 1>) farthest from the reference point (X ₀ <i>, X ₁ <j>, X ₂ <k>, X ₃ <l>) , X ₂ <k + 1>, X ₃ <l + 1>) is called a diagonal point, and a line connecting the reference point and the diagonal point is called a diagonal line.

The multidimensional supertetrahedral interpolation method always uses a supertetrahedron regardless of the magnitude relationship of the difference between the input coordinate and the reference point coordinate (ΔX ₀ , ΔX ₁ , ΔX ₂ , ΔX ₃ ). It has the feature that it is selected as a vertex of the body. Therefore, the process of selecting the supertetrahedron corresponding to the input coordinates from the unit hypercube starts from the reference point, the diagonal point is the goal, and each ridge line is the path from the start to the goal. It can be considered as a route selection problem for selecting a route to a point.
In this case, the number of “cases” that can be taken by the path is 24 (= 4! = 4 × 3 × 2 × 1), and the unit hypersolid can be divided into 24 supertetrahedra by the four-dimensional interpolation calculation. I understand. Similarly, in the case of the N-dimensional interpolation calculation, the unit super solid is represented by N! It can be seen that it can be divided into individual supertetrahedra.

Next, N-dimensional interpolation calculation will be described. Coordinates (X ₀ <i>, X ₁ <j>, X ₂ <k>, ... X _N on a space (X ₀ , X ₁ , ..., X _N-1 ) fixed to N-dimensional orthogonal coordinates _-2 <a>, X _N-1 <b>) is referred to as P <i, j, k,..., A, b>. Also, the coordinates of the reference point of the hypercube selected by certain input coordinates are (X ₀ <i>, X ₁ <j>, X ₂ <k>,… X _N-2 <a>, X _N-1 <b>), and the difference between the input coordinates and the reference point coordinates is (ΔX ₀ , ΔX ₁ , ΔX ₂ ,..., ΔX _N-2 , ΔX _N-1 ). When the relationship {ΔX ₀ > ΔX ₁ > ΔX ₂ >...> ΔX _N−2 > ΔX _N−1 } holds, the interpolation value X can be obtained using the following (Expression 7).

(First embodiment)
Next, a first embodiment of the present invention will be described.
FIG. 1 shows an example of the overall configuration of a color copying machine.
1, the image reading unit 120 includes a lens 122, a CCD sensor 124, an analog signal processing unit 126, and the like. An image of the original 100 formed on the CCD sensor 124 through the lens 122 is converted into analog electric signals of R (Red), G (Green), and B (Blue) by the CCD sensor 124.

  The image information converted into the analog signal is input to the analog signal processing unit 126, and is subjected to analog-digital conversion (A / D conversion) after correction or the like is performed for each color of R, G, and B. A digitized full color signal (hereinafter referred to as a digital image signal) is input to the image processing unit 130. The image processing unit 130 performs input γ correction, color space conversion, density correction, and screen processing on the digital image signal, and outputs the digital image signal after the above processing to the printer unit 140.

The printer unit 140 includes, for example, an exposure control unit (not shown) made of a laser or the like, an image forming unit (not shown), a transfer paper conveyance control unit (not shown), and the like. An image is recorded on the transfer paper by the image signal.
The CPU circuit unit 110 includes a CPU 112 for arithmetic control, a ROM 114 for storing fixed data and programs, a RAM 116 for temporarily storing data and loading programs, an external storage device 118, and the like. The reading unit 120, the image processing unit 130, the printer unit 140, and the like are controlled, and the sequence of the color copying machine is comprehensively controlled. The external storage device 118 is a medium such as a disk for storing parameters and programs used by the color copying machine, and the data and programs in the RAM 116 may be loaded from the external storage device 118.

The multidimensional interpolation apparatus of the present embodiment realizes color space conversion performed by the image processing unit 130, for example. The present invention is not limited to the color copying machine as shown in FIG. 1, but can be implemented in a printer or a PC (personal computer). Hereinafter, the circuit configuration and processing flow of the multidimensional interpolation apparatus of the present embodiment will be described in detail with reference to FIG.
The multidimensional interpolation apparatus of this embodiment performs an interpolation operation on an N-dimensional input signal and outputs a new signal. First, the digital image signal 205 is input to the multidimensional interpolation device. The digital image signal 205 is an N input signal. The digital image signal 205 is divided by the data dividing unit 210 into an upper bit signal 212 and a lower bit signal 214.

Since the digital image signal 205 is an N-input signal, the upper bit signal 212 and the lower bit signal 214 obtained by dividing the digital image signal 205 are N signals. In general, the bit depth of the upper bit signal 212 is determined according to the number of unit hypersolids used in the interpolation calculation. For example, when there are (2 ^x -1) unit supersolids for an axis in the N-dimensional space, the bit depth of the upper bit signal 212 corresponding to that axis is x bits. In general, the lower bit signal 214 is represented by the remaining bit depth obtained by subtracting the bit depth of the upper bit signal 212 from the bit depth of the digital image signal 205.

  Next, the order determination unit 230 receives the N lower bit signals 214, determines the magnitude relationship of the lower bit signals 214, and outputs the determined result as an order signal 232. This order signal 232 is determined in advance according to a process of selecting a ridge line of a super tetrahedron (N + 1 face) to be used for interpolation calculation from the unit hypersolid described above. It is a signal indicating a street ridge line selection pattern.

  The reference coordinate selection unit 220 determines and determines coordinates for reading (N + 1) reference values 252 necessary for the interpolation calculation from the multidimensional LUT 260 according to the order signal 232 that is the determination result of the order determination unit 230. The coordinates are output as (N + 1) reference coordinate signals 222.

  Similarly, the weighting factor calculation unit 240 calculates a weighting factor signal 242 for interpolation calculation based on the lower-order bit signal 214 and outputs it to the interpolation calculation unit 270. As a method for calculating the weighting factor signal 242 for the interpolation calculation based on the lower bit signal 214, for example, a table serving as a correspondence table is stored in advance in the ROM, and the weighting factor signal 242 for the corresponding interpolation calculation is stored. Can be read out from the ROM to convert the lower bit signal 214 into a weighting coefficient signal 242 for interpolation, or a method of calculating by a predetermined calculation formula.

Further, the reference value reading unit 250 reads (N + 1) reference values from the multidimensional LUT 260 using (N + 1) reference coordinate signals 222 and outputs the read reference values as reference value signals 252. The multidimensional LUT 260 is implemented by a plurality of sub memories. The reference value reading unit 250 outputs a number of address signals 254 equal to the number of sub memories to each sub memory to access each sub memory, and receives a data signal 262 from each sub memory.
The interpolation calculation unit 270 receives the (N + 1) reference value signals 252 and the N interpolation coefficient weight coefficient signals 242, performs the interpolation calculation according to the above-described interpolation calculation formula, and outputs one output signal 275. To do.

Next, an example of a method in which the reference value of the multidimensional LUT 260 is divided and stored in 2 ^N−1 sub-memory without overlapping in the ^N- dimensional supertetrahedral interpolation method will be described.
FIG. 3 shows the result of classifying each vertex of the unit solid from the reference point to the diagonal point using the four-dimensional Manhattan distance according to the four-dimensional hypertetrahedral interpolation method.

Here, the N-dimensional Manhattan distance Dm is expressed as two points A (X ₀ <i ₁ >, X ₁ <j ₁ >, X ₂ <k ₁ >,..., X _N on a space fixed to N-dimensional orthogonal coordinates. _{-2 <a 1>, X N-} 1 and _{<b 1>), B (} X 0 <i 2>, X 1 <j 2>, X 2 <k 2>, ..., X N-2 <a 2 >, X _N-1 <b ₂ >) and is defined by the following (Equation 8).
Dm = | X ₀ <i ₁ > −X ₀ <i ₂ > | + | X ₁ <j ₁ > −X ₁ <j ₂ > | + | X ₂ <k ₁ > −X ₂ <k ₂ > | + _{··· + | X N-2 <a} 1> -X N-2 <a 2> | + | X N-1 <b 1> -X N-1 <b 2> | ··· (8 type)

Then, in the three-dimensional tetrahedral interpolation method, utilizing the fact that store by dividing the multi-dimensional LUT in 2 ² sub memory, the reference point of the four-dimensional unit hyper stereo _{(X 0 <i>, X} 1 <j>, X ₂ <k>, X ₃ <l>) to point (X ₀ <i + 1>, X ₁ <j + 1>, X ₂ <k + 1>, X ₃ <l>) The multi-dimensional LUT division storage method in the three-dimensional tetrahedral interpolation method is applied to 2 ³ vertices (region 810 in FIG. 3).

Since the process of selecting a super tetrahedron to be used for the interpolation operation from the unit hypersolid is considered to be a route selection problem of which vertex to reach the diagonal from the reference point, a group of vertices having the same Manhattan distance Dm Always select only one vertex. That is, when vertex groups having the same Manhattan distance Dm are stored in the same submemory, two different vertices are not read from the submemory at the same time. For this reason, there is no contention for access to the sub memory.
In FIG. 3, the vertices surrounded by the thick lines described as Mem0, Mem1, Mem2, and Mem3 are stored in the same submemory. In the 3D tetrahedral interpolation method, each vertex is divided into four submemory. Can be understood.

Further, the remaining units supersolid two ^three vertices _{(X 0 <i>, X} 1 <j>, X 2 <k>, X 3 <l + 1>) ~ (X 0 <i + 1>, X ₁ <j + 1>, X ₂ <k + 1>, X ₃ <l + 1>), a multi-dimensional LUT partition storage method in the three-dimensional tetrahedral interpolation method is applied (region in FIG. 3). 820).

Finally, the vertices surrounded by the thick lines described as Mem0,..., Mem7 shown in FIG. 3 are stored in the same submemory, and the multidimensional LUT 260 has (2 × 2 ² ) submemory. It is divided into. Similarly, in the N-dimensional hypertetrahedral interpolation method, 2 ³ vertices (X ₀ <i>, X ₁ <j>, X ₂ <k>,...) To (X ₀ <i + 1>, X _{1 <j + 1>, X} 2 <k + 1>, ... each), it may be applied dividing method of storing multi-dimensional LUT in the three-dimensional tetrahedral interpolation.

Since the 2 ³ vertices are 2 ^N−3 sets, and one set is stored in 2 ² sub memories, the multidimensional LUT 260 is divided into 2 ^N−1 sub memories in total. It will be. Thus the number of sub-memories increases constantly 2 ² sub-memory units. Hereinafter, the unit of increase is referred to as a clustering unit, a group of sub memories corresponding to the number of units is referred to as a cluster sub memory, and the 2 ^N-3 sets are referred to as a cluster number. The number of input dimensions to which one cluster submemory is applied is referred to as a base m (“m = 3” in the case of the first embodiment).

Next, the reference value reading unit 250 and the multidimensional LUT 260 will be described in detail with reference to FIG.
In FIG. 4, (N + 1) address generation circuits 310 receive (N + 1) reference coordinate signals 222 and calculate an address signal 312 and a sub memory bank selection signal 314, respectively. The reference value reading unit 250 includes an address selection circuit 320 corresponding to each sub memory. Each address selection circuit 320 receives (N + 1) sets of address signals 312 and a sub memory bank selection signal 314.

  Each address selection circuit 320 generates (N + 1) address signals 312 as an address signal 254 in which the sub memory bank ID “SmB_ID” associated with the address selection circuit 320 is equal to the input sub memory bank selection signal 314. 1 is selected, and the sub-memory 330 is accessed using the selected address signal 254.

  Next, the (N + 1) data selection circuits 340 have the same number of data signals 262 as the number of sub memories output from the sub memory 330 and the sub memory bank selection signals output from the corresponding address generation circuit 310. 314, and selects only the data signal 262 output from the sub memory 330 having the sub memory bank ID “SmB_ID” equal to the sub memory bank selection signal 314, and outputs the selected data signal 262 as the reference value 252. To do.

Next, an example of a method for generating an address will be described. FIG. 5 shows an example of the configuration of the address generation circuit 310.
5, the address generation circuit 310 includes a bit operation unit 410, a sub memory bank generation circuit 420, and a sub memory address generation circuit 430, and includes an N-dimensional reference coordinate signal 222 ({S ₀ , S ₁ ,..., S _t ,..., S _N-1 }) are input from the reference coordinate selection unit 220.

FIG. 6 shows an example of a method in which the bit operation unit 410 operates the bits of the N-dimensional reference coordinate signal 222.
In FIG. 6, the bit operation unit 410 rearranges the N N-dimensional reference coordinate signals 222 each having a bit depth of w bits into a predetermined format from the lower bits, and converts the signal 515 of (N × w) bits into create. In the example illustrated in FIG. 6, first, the bit operation unit 410 extracts only the lower first bit of the N reference coordinate signals 222 and outputs an N-bit signal [S _N−1 [0] _,. [0],..., S ₁ [0], S ₀ [0]] are created.

Next, the bit operation unit 410 extracts only the lower second bit of the N N-dimensional reference coordinate signals 222 and outputs N-bit signals [S _N-1 [1],..., _St [1], ..., S ₁ [1], S ₀ [1]] are created. Further, the bit operation unit 410 extracts only the lower u-th bit of the N reference coordinate signals 222, and outputs N-bit signals [S _N-1 [u],..., _St [u],. , S ₁ [u], S ₀ [u]]. Thus, the most significant N-bit signal [S _N-1 [w-1],..., S _t [w-1],..., S ₁ [w-1] of the N reference coordinate signals 222. ], S ₀ [w-1]], and the N-bit signals are concatenated to form an (N × w) -bit signal 515 as shown in FIG.

In this way, by manipulating the bits of the N N-dimensional reference coordinate signals 222 (bit manipulation), the number of unit hypersolids of the multi-dimensional LUT 260 can be easily increased. For example, when extending the multidimensional LUT 260 in a multidimensional interpolation apparatus that may have a large capacity of the sub memory 330, the bit depth of the N-dimensional reference coordinate signal 222 is changed from w bits to (w + 1) bits, An N-bit signal [S _N-1 [w],..., S _t [w],..., S ₁ [w], S ₀ [w]] is connected to the output of the bit operation unit 410. do it. In addition, since the extended bits are connected to the upper part of the address before extension (the address generation method will be described later), the address and the multi-dimensional interpolator before extension and the multi-dimensional interpolator after extension Compatibility with data can be made compatible.

Next, details of the sub memory bank generation circuit 420 will be described with reference to FIG.
In FIG. 7, a lower (3 + N) bit signal 605 of the (N × w) bit signal 515 output from the bit operation unit 410 is input to the sub memory bank generation circuit 420. The lower 3 bits signal 608 of the signal 605 is added by the adder 630. The added value (S ₂ [0] + S ₁ [0] + S ₀ [0]) is output as a signal 632.

Similarly, the upper 3 bits signal 606 is added by the adder 610. The added value (S ₂ [1] + S ₁ [1] + S ₀ [1]) 612 is shifted left by 1 bit by the left shifter 640 (added value 612 << 1) and shifted. The value is output as signal 642.
The cluster number generation circuit 620 shifts 1 to the left by the input value 607 (1 << input value 607), and outputs the shifted value as a signal 622.
Further, after the signals 622, 632, and 642 are added by the adder 650, only the lower (N-1) bits are extracted by the sub memory bank selection signal generation circuit 660, and the extracted signals are This is output as a sub memory bank selection signal 314.

Next, details of the sub memory address generation circuit 430 will be described with reference to FIG.
The sub memory address generation circuit 430 generates an address signal 312 for accessing the sub memory 330 based on the (N × w) bit signal 515 output from the bit operation unit 410.
Of the lower (3 + N) bits of the (N × w) bit signal 515 output from the bit operation unit 410, the 6-bit signal 709 is input to the lower address generation circuit 710, and the 4-bit address lower signal 712 is output. Created. The remaining bits are used to generate the sub memory bank signal 314 and are not used by the sub memory address generation circuit 430.
Finally, using the bit coupler 720, the address high-order signal 708 and the 4-bit address low-order signal 712 are bit-connected, and the concatenated signal is output as the address signal 312.

Here, the lower address generation circuit 710 will be described in more detail.
The 6-bit signal 709 input to the lower address generation circuit 710 is formed by the lower 2 bits of the three-dimensional input signal {S ₀ , S ₁ , S ₂ } and corresponds to the coordinates of the 3D-LUT. ing. As described above, since the 4 × 4 × 4 3D-LUT is divided into four sub memories 330, the 6-bit signal 709 corresponds to all the reference values in the four sub memories 330. is doing.

  However, since the required lower address signal 712 is an address signal of lower 4 bits for each sub memory 330, these 6 bits cannot be used as they are. Table 1 shows a correspondence relationship between the 6-bit input signal 709 classified for each sub memory 330 and the 4-bit address lower signal 712 to be output.

  FIG. 9 is a diagram showing the positions of the reference values corresponding to the sub memory bank ID (“SmB_ID = 0”) in the 4 × 4 × 4 3D-LUT in ascending order of address ( Circled numbers 0 to 15 in FIG. 9). In FIG. 9A, the reference value of the part that is not marked means that it is stored in another sub memory 330.

  By creating a table as shown in Table 1 and mounting a 4 × 4 × 4 3D-LUT as a table ROM in the circuit, a 6-bit input signal 709 can be easily converted into a 4-bit address lower signal 712. .

Further, if the scattered reference points in the 4 × 4 × 4 3D-LUT are arranged in advance and are grouped into 4 × 4 reference points as shown in FIG. 9B and FIG. ROM becomes unnecessary. When the reference points are arranged as shown in FIG. 9B, the address low-order signal 712 includes [S ₂ [1], S ₂ [0], S ₁ [1], S ₁ [0]]. The signal is a bit concatenated in this order. Further, when the reference points are arranged as shown in FIG. 9C, the address lower signal 712 includes [S ₂ [1], S ₁ [1], S ₂ [0], S ₁ [0]. ]] In the order of bit concatenation.

  Here, Table 2 shows a result of comparing the first and second conventional techniques described above with the present embodiment.

  As can be seen from Table 2, in this embodiment, the number of sub-memory can be reduced to half that of the first conventional technique. Further, unlike the second conventional technique, in the present embodiment, the number of reference values stored per divided sub memory can be always raised to a power of two. Further, in the supertetrahedral interpolation method of N-dimensional input (N is an integer of 4 or more), the lower 4 bits of the address are always determined from the basic three-dimensional input. Therefore, the circuit that generates the lower 4 bits of the address is the same circuit even if the value of N changes.

  In the second prior art, a divider is required to generate an address. Therefore, in order to reduce the size of the address generation circuit, a device such as changing the contents of the table ROM for each value of N is required. , It is necessary to redesign the address generation circuit for each value of N. On the other hand, as described above, in this embodiment, the address generation circuit can be reused even if the value of N changes, and it is possible to easily cope with multi-dimensionalization. Furthermore, even when the bit depth of the N-dimensional input signal increases with the increase in the number of unit hypersolids of the multidimensional LUT, it can be easily handled by performing bit operations at the time of address generation. That is, in the present embodiment, it is possible to realize a multidimensional interpolation device that is scalable with respect to both the number of N-dimensional input signals and the bit depth.

As described above, according to this embodiment, in the supertetrahedral interpolation method of N-dimensional input (N is an integer of 4 or more), the lower 4 bits of the address are always determined from the basic three-dimensional input, and the address The circuits that generate the lower 4 bits are the same circuit even if the value of N changes. For this reason, the address generation circuit can be reused, and multi-dimensionalization can be easily handled. In addition, even when the bit depth of the N-dimensional input signal increases as the number of unit hypersolids of the multi-dimensional LUT increases, the bit depth of the N-dimensional input signal can be obtained only by performing a bit operation when generating an address. Can be easily accommodated. That is, according to the present embodiment, it is possible to provide a multidimensional interpolation device that is scalable with respect to both the number of N-dimensional input signals and the bit depth. In particular, when the multidimensional interpolation device of this embodiment is applied to the four-dimensional supertetrahedral (pentahedral) interpolation method, the number of sub-memory 330 is larger than the essential five, but the smallest 2. The effect is high because it is a power of 2 (= 2 ³ ).

(Second Embodiment)
Next, a second embodiment of the present invention will be described. In the first embodiment described above, with respect to the underlying three-dimensional input (base m = 3), by applying the sub-memory 2 ^two units (clustering unit = 2 ^2), by dividing the multi-dimensional LUT I remembered it. When applied to the four-dimensional supertetrahedral interpolation method, two sets of sub-memory (cluster sub-memory) of the unit (number of clusters = 2 ^4-3 ) are required. The method of storing the sub-memory divided into the multi-dimensional LUT is not limited to the method described in the first embodiment, and there is a problem that the address generation circuit becomes complicated, but it can be generalized. In the description of the present embodiment, the same parts as those in the first embodiment described above are denoted by the same reference numerals as those in FIGS.

The clustering unit is (m + 1) for the base m, which is the basic number of input dimensions. When applied to the N-dimensional supertetrahedral interpolation method, the number of clusters is 2 ^Nm . The total number of sub-memory NSmB necessary for the N-dimensional supertetrahedral interpolation method is as shown in the following (formula 9) when N> m is satisfied.
NSmB = (m + 1) × 2 ^Nm (9 formulas)

  Table 3 shows the correspondence relationship between the base m, which is the basic number of input dimensions, and the total number of sub memories when the N-dimensional hypertetrahedral interpolation method is used.

It is also possible to implement a method of generalizing the division and storage of the sub-memory in the multidimensional LUT.
FIG. 10 shows a generalized sub memory bank generation circuit 1420.
In FIG. 10, the lower 1 (m + N × u) bit signal 1605 of the (N × w) bit signal 515 output from the bit operation unit 410 is input. Here, u indicates a bit depth at which (m + 1) can be expressed in binary. That is, it is the smallest integer greater than or equal to log ₂ (m + 1). The signal 1608 of lower m bits of the signal 1605 is added by the adder 1630. The added value (S _m−1 [0] +... + S ₁ [0] + S ₀ [0]) is output as a signal 1632.

Similarly, the upper m-bit signal 1606 is added by the adder 1610. The added value (S _m−1 [1] +... + S ₁ [1] + S ₀ [1]) is left shifted by 1 bit by the left shifter 1640, and the shifted value is output as a signal 1642. Is done. Then, as shown in FIG. 10, the same operation is performed until ((S _m−1 [u−1] +... + S ₁ [u−1] + S ₀ [u−1]) << u). .

Further, the signal 1632 obtained by the addition process described above and the shifted signal 1642 are added by an adder 1650 described later. Of the N reference coordinate signals (S ₀ , S ₁ ,..., S _t ,..., S _N-1 ) input, the signal (S ₀ , S ₁ ,..., S _m−1 ) means that only the lower u bits are extracted and the sum is obtained. In the example of the sub memory / bank generation circuit 1420 shown in FIG. 10, since processing is performed on the output of the bit operation unit 410, addition processing and shift processing are performed.

Further, the cluster number generation circuit 1620 outputs a value obtained by multiplying the clustering unit (m + 1) by the input value 1607 as an integer, as a signal 1622.
Further, the signals 1632, 1622 and 1642 are added by an adder 1650 and output as a signal 1652. Then, the sub memory bank selection signal generation circuit 1660 divides the signal 1652 by (m + 1) × 2 ^Nm , and the remainder is output as the sub memory bank selection signal 1314.

  As described above, the sub memory bank generation circuit 1420 according to the present embodiment is a circuit that performs the calculation based on the following (Equation 10) to generate the sub memory bank selection signal 1314.

However, << represents a bit shift to the left, and% represents a remainder of division (for example, a% b represents a remainder obtained by dividing a by b).
In the first term “ΣX _j % (M + 1)” of (Equation 10), IDs 0 to M are created, and the second term “(M + 1) Σ [(X% 2) <<(jM)], (M + 1) * 0, (M + 1) * 1, (M + 1) * 2,.
In FIG. 10, the cluster number generation circuit 1620 determines whether each of the (N−M) components not selected is odd or even, and adds a multiple of (M + 1) based on the determination result to the remainder. The value obtained by the above is used as an index value, and the vertices are classified based on the index value. In the second embodiment, an n-dimensional input (S ₀ , S ₁ ,..., S _m−1 , S _m , S _{m + 1 ,. -1} ) of the signals, paying attention to the signals of (S _m , S _{m + 1} ,..., S _n-1 ), and their LSBs (S _m [0], S _{m + 1} [0],. S _n-1 [0]) is used, so that it is determined whether the signal of (S _m , S _{m + 1} ,..., S _n-1 ) is odd or even.
In the second embodiment, the presence / absence of the addition of the bank signal offset is determined based on the LSB of the signal of interest described above.
(S _m [1], S _{m + 1} [1],…, S _n-1 [1])
(S _m [2], S _{m + 1} [2],…, S _n-1 [2])
...
(S _m [w-1], S _{m + 1} [w-1],…, S _n-1 [w-1])
Instead of LSB, a value obtained by multiplying a clustering unit (m + 1) by an integer based on on / off of a predetermined bit position in (S _m , S _{m + 1} ,..., S _n-1 ) The signal 1622 may be output.

Next, a generalized sub memory address generation circuit 1430 will be described with reference to FIG.
As shown in FIG. 11A, the sub memory address generation circuit 1430 accesses an address for accessing the sub memory 330 based on the (N × w) bit signal 1515 output from the bit operation unit 410. A signal 1312 is generated.

Of the lower (m + N × u) bits of the (N × w) bit signal 1515 output from the bit operation unit 410, the (m × u) bit signal 1709 is input to the lower address generation circuit 1710, and {( An m−1) × u} bit address lower signal 1712 is generated. The remaining bits are used to generate the sub memory bank signal 1314 and are not used in the sub memory address generation circuit 1430.
Finally, the bit higher rank signal 1708 and the {(m−1) × u} bit lower address signal 1712 are bit-linked using the bit coupler 1720, and the linked signal is output as the address signal 1312.

As shown in FIG. 11B, the lower address generation circuit 1710 generates an address lower signal 1712 using the table ROM 1710a. In this case, as the base m becomes larger, the required size of the table ROM becomes larger.
Therefore, when the base m is a value of power of −1 (for example, the base m is 3 and 7 in Table 3), the low-order address generation circuit 1710 has a bit as shown in FIG. This can be realized using the circuit 1710b that performs only the operation. As can be seen from Table 3, in the case of the 8-dimensional supertetrahedral interpolation method, if a configuration with a base m of 7 is used, an address can be easily generated, and an 8-dimensional supertetrahedral interpolation device (multi Dimensional interpolation device).

As described above, according to the present embodiment, in the supertetrahedral interpolation method of N-dimensional input (N is an integer of 3 or more), if the base m, which is the basic number of input dimensions, is (N−1). The clustering unit (m + 1) is always N. Therefore, the number of clusters is 2 ^Nm = 2 ¹ , and the multidimensional LUT can be divided and stored in (N × 2) sub-memory.

(Third embodiment)
Next, a third embodiment of the present invention will be described. In this embodiment, the multidimensional LUT 260 in the first embodiment described above is realized by using an external memory and a cache mechanism. In addition, the bit depth of the reference value of the multidimensional LUT in this embodiment is 8 bits. In the description of the present embodiment, the same portions as those in the first and second embodiments described above are denoted by the same reference numerals as those in FIGS. 1 to 11 and detailed description thereof is omitted.

  The cache mechanism according to the present embodiment makes it possible to update data at high speed by performing burst access to the external memory. Here, burst access is a method in which data corresponding to a plurality of consecutive addresses is read in a batch rather than reading data from the external memory word by word. Generally, general-purpose memories such as SDRAM and DDR-SDRAM support burst access in units of 16 bytes (128 bits). In this embodiment, the unit of burst access is 16 bytes.

  Therefore, the cache mechanism accesses the external memory by using data of 8 bits (bit depth of the reference value) × 16 words as one unit. As can be seen from FIG. 8, the address upper signal 708 becomes an address signal for accessing the external memory, and the address lower signal 712 indicates the number of bytes of the 16-byte data obtained by the burst access. Signal. Hereinafter, the terms “address upper signal” and “address lower signal” are distinguished from each other.

FIG. 12 shows an example of the configuration of the multidimensional LUT in the present embodiment, and the circuit operation of the present embodiment will be described with reference to FIG.
When the address signal 254 is input, the miss hit determination circuit 920 determines whether or not the cache has been hit based on the address high-order signal 708 included in the address signal 254. The miss hit determination circuit 920 stores the upper value of the address corresponding to the data stored in the data buffer 950 in the tag RAM 930.

  The miss-hit determination circuit 920 determines that the cache has been hit if the address upper signal 708 included in the address signal 254 input from the outside is the same as one of the address upper values stored in the tag RAM 930. On the other hand, if the address upper signal 708 is different from any address upper value, it is determined that a miss has occurred in the cache. When it is determined that a miss hit has occurred, the miss hit determination circuit 920 overwrites the upper address signal 708 input from the outside with the upper address value stored in an arbitrary tag number in the tag RAM 930, and sets the tag number to that tag number. The data in the corresponding data buffer 950 is replaced with data updated from the external memory.

  In order to perform this data replacement processing, first, the miss hit determination circuit 920 sends the address high-order signal 708 included in the address signal 254 to the arbitration circuit 975 as the address signal 922, and the arbitration circuit 975, the bus 970, and Data is requested from the external memory 960 via the memory controller 965. The arbitration circuit 975 arbitrates each address signal 922 output from the cache mechanism connected to each sub memory and sends the address signal 922 to the memory controller 965 in order.

  Further, the data sent from the memory controller 965 is returned to the cache mechanism connected to each of the sub memories 910a to 910n in the order of sending the address signal 922. Next, regardless of the determination result of whether or not there is a miss hit in the cache, the miss hit determination circuit 920 includes a flag indicating the determination result, a cache tag number, and a lower address signal 712 included in the address signal 254. Are sent together as a command signal 942 to the delay circuit 940. The number of sub-memory 910 is not limited to that shown in FIG.

  The delay circuit 940 is implemented by FIFO (First-In First-Out), and can store a certain number of command signals 942. For this reason, the miss-hit determination circuit 920 sends out the address signal 922 to the external memory 960 continuously while there is an empty area in the FIFO. The cache controller 945 receives the command signal 942 from the delay circuit 940, and if the determination result is a hit, takes out 16-byte data 952 from the data buffer 950 corresponding to the tag number. Next, 1-byte data to be output is selected by the address lower signal 712, and the selected data is output as the data signal 262.

  If the determination result is a miss hit, the cache controller 945 waits until data 978 is sent from the arbitration circuit 975. When the data 978 is sent, 1-byte data to be output is selected from the data 978 by the address lower signal 712 and output as the data signal 262, and the data 978 is written to the data buffer 950 according to the tag number, The buffer 950 is updated.

Note that the address high-order signal 708 is subjected to the bit operation described with reference to FIG. For this reason, the cache miss-hit rate is substantially uniform for any change in the N-dimensional reference coordinate signals (S ₀ , S ₁ ,..., S _t ,..., S _N-1 ).

  As described above, according to the present embodiment, in the N-dimensional input supertetrahedral interpolation device (N is an integer of 4 or more), even if the value of N changes, the address lower signal 712 is always 4 bits. Even if the value of is changed, it is not necessary to change the cache mechanism, and the same circuit can be reused. Therefore, regardless of the value of N, a multidimensional LUT can be easily realized using an external memory and a cache mechanism.

(Other embodiments of the present invention)
In order to operate various devices to realize the functions of the above-described embodiments, program codes of software for realizing the functions of the above-described embodiments are provided to an apparatus or a computer in the system connected to the various devices. What is implemented by operating the various devices according to a program supplied and stored in a computer (CPU or MPU) of the system or apparatus is also included in the scope of the present invention.

  In this case, the program code of the software itself realizes the functions of the above-described embodiments, and the program code itself and means for supplying the program code to the computer, for example, the program code are stored. The recorded medium constitutes the present invention. As a recording medium for storing the program code, for example, a flexible disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, a magnetic tape, a nonvolatile memory card, a ROM, or the like can be used.

  Further, by executing the program code supplied by the computer, not only the functions of the above-described embodiments are realized, but also the OS (operating system) or other application software in which the program code is running on the computer, etc. It goes without saying that the program code is also included in the embodiment of the present invention even when the functions of the above-described embodiment are realized in cooperation with the embodiment.

  Further, after the supplied program code is stored in the memory provided in the function expansion board of the computer or the function expansion unit connected to the computer, the CPU provided in the function expansion board or function expansion unit based on the instruction of the program code Needless to say, the present invention includes a case where the functions of the above-described embodiment are realized by performing part or all of the actual processing.

As described above, in each embodiment of the present invention, the N-dimensional digital image signal 205 is divided into the upper bit signal 205 and the lower bit signal 214, and all combinations of the divided upper bit signals are supported. The reference value is divided and stored in 2 ^N-1 sub-memory 330 (multi-dimensional LUT 260) without overlapping, and (N + 1) reference values necessary for the interpolation operation are read out simultaneously, thereby interpolating the calculation. And an address for accessing the sub memory 330 can be generated without a divider so that interpolation operations such as color space conversion can be performed at high speed. Can be easily realized with as small a circuit scale as possible. Furthermore, when extending to multi-dimensions, it is not necessary to redesign the address generation circuit or the cache mechanism, and it is possible to provide a multi-dimensional interpolation device that is highly scalable with respect to both the number of inputs and the bit depth.

DESCRIPTION OF SYMBOLS 100 Document 110 CPU circuit unit 120 Image reading unit 130 Image processing unit 140 Printer unit 210 Data division unit 220 Reference coordinate selection unit 230 Order determination unit 240 Weight coefficient calculation unit 250 Reference point reading unit 260 Multidimensional LUT
270 Interpolation calculator

Claims (11)

A multidimensional interpolation device that outputs a scalar signal from an input N-dimensional vector signal (N is an integer of 4 or more),
Dividing means for dividing each component of the input N-dimensional vector signal value into upper bits and lower bits;
Memory means for storing a reference value corresponding to the combination of the upper bit data divided by the dividing means;
Read means for reading the reference value;
Interpolation calculation means for performing interpolation calculation on the input N-dimensional vector signal based on the reference value read by the reading means and lower bit data divided by the dividing means,
The multi-dimensional interpolation apparatus, wherein the memory means stores the reference value divided into 2 ^N-1 sub-memory. Selecting means for selecting predetermined M (M is an integer smaller than N) components from N components in the input N-dimensional vector signal;
First calculation means for calculating the sum of M components selected by the selection means;
Second calculating means for calculating a remainder obtained by dividing the sum calculated by the first calculating means by (M + 1);
Based on the value of each predetermined bit position of (N−M) components not selected by the selection means, a multiple of (M + 1) is added to the remainder calculated by the second calculation means. A third calculating means for calculating an index value;
Classification means for classifying the reference value based on the index value calculated by the third calculation means;
2. The multidimensional interpolation apparatus according to claim 1, wherein the memory unit divides and stores the reference value into 2 ^N−1 sub-memory based on a classification result by the classification unit. A cluster number signal is generated based on the value of each predetermined bit position of the (N-3) component from the higher order data of N components in the input N-dimensional vector signal, and the generated cluster number signal is generated. Sub memory selection means for selecting the sub memory based on
The multi-dimensional interpolation apparatus according to claim 1, wherein the reading unit reads a reference value stored in a sub memory selected by the sub memory selecting unit. The upper bit data of N components in the input N-dimensional vector signal is converted into a digital signal to obtain an upper bit signal,
Among the N components in the input N-dimensional vector signal, a 6-bit signal is generated using lower 2 bits of the higher-order bit signal in specific 3 components, and the 6-bit signal is converted into a 4-bit signal. 4. The multidimensional interpolation apparatus according to claim 1, further comprising lower-order address generation means for converting into a 4-bit address lower-order signal and generating a 4-bit address lower-order signal. 5.   The sub-memory selection means uses the lower 1 bit of the upper bit signal of the (N-3) component from which the lower 2 bits were not extracted by the lower address generation means as a cluster number signal, and based on the cluster number signal, The multidimensional interpolation apparatus according to claim 4, wherein a signal for selecting the sub-memory is generated.   6. The multidimensional interpolation apparatus according to claim 4 or 5, wherein the lower address generation means converts the 6-bit signal into a 4-bit signal using a table stored in a ROM.   6. The multidimensional data according to claim 5, wherein the lower address generating means converts the 6-bit signal into a 4-bit signal by truncating the lower 2 bits of one of the specific three components. Interpolator. The upper bit data of N components in the input N-dimensional vector signal is converted into a digital signal to obtain an upper bit signal,
The operation of taking one bit at the same bit position from the N upper bit signals divided by the dividing means and concatenating the extracted N bits together to make an N bit signal A plurality of N-bit signals obtained by performing the processing from the least significant bit signal to the most significant bit signal, and a higher-order address generating means for generating a signal obtained by removing (N + 3) bits from the concatenated signal as an address higher-order signal The multidimensional interpolation apparatus according to any one of claims 1 to 4, wherein   9. The multi-dimensional interpolation apparatus according to claim 8, further comprising a cache mechanism that determines a cache hit or a miss hit by using an address upper signal generated by the upper address generation means. A multidimensional interpolation method for outputting a scalar signal from an input N-dimensional vector signal (N is an integer of 4 or more),
A dividing means performs a dividing step of dividing each component of the input N-dimensional vector signal value into upper bits and lower bits,
A reading unit performs a reading step of reading a reference value corresponding to the combination of the upper bit data divided by the dividing step,
An interpolation calculation means performs an interpolation calculation step of performing an interpolation calculation on the input N-dimensional vector signal based on the reference value read in the reading step and the lower bit data divided in the division step. ,
The multi-dimensional interpolation method, wherein the reference value is divided and stored in 2 ^N-1 sub-memory. A computer program for causing a computer to output a scalar signal from an input N-dimensional vector signal (N is an integer of 4 or more),
A division step of dividing each component of the input N-dimensional vector signal value into upper bits and lower bits, respectively;
A reading step of reading a reference value corresponding to the combination of the upper bit data divided by the dividing step;
Based on the reference value read in the reading step and the lower bit data divided in the dividing step, the computer executes an interpolation calculation step for performing an interpolation calculation on the input N-dimensional vector signal,
The computer program according to claim ¹ , wherein the reference value is divided and stored in 2 ^N-1 sub-memory.

Images