JPH1051793A - Block matching device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce an operation quantity and to suppress a pixel data transfer quantity in a retrieval area, which is to be transferred, by sub-sampling pixels existing in a template one pixel by one, generating reduced templates and evaluating a difference of a value of the pixels in a corresponding position with a reduced block segmented from a reduced picture for the respective reduced templates. SOLUTION: A reduced template generation means 40 divides the template of four pixels × four lines into a small block of two pixels × two lines and samples a pixels existing in a prescribed position in a small block by one. Thus, the reduced templates ST1-4 are generated. A reduced block evaluation means 50 evaluates the difference of the values of the picture elements in a corresponding position with a reduced block of two pixels × two lines, which is segmented from a reduced picture of five pixels × five lines generated by a reduced picture generation means 30, for the respective reduced templates ST1-ST4 generated by the educed template generation means 40.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a block matching apparatus used for detecting a motion vector in moving picture coding, moving picture recognition, and the like.

[0002]

2. Description of the Related Art A motion vector is a vector indicating from which position in a picture different from a prescribed picture a pixel block composed of a plurality of adjacent pixels in a prescribed picture has moved.

FIG. 7 is a diagram for explaining the operation of a conventional block matching apparatus, and is a diagram showing a relationship between a template T0 to be subjected to motion detection and a search area R0.

The conventional block matching apparatus searches for where a pixel block in a predetermined image has moved in another image, that is, a search area R0 is a pixel block in the other image. , Template T0 is
This is a pixel block in the predetermined image (a pixel block in the current image). In other words, the conventional block matching apparatus searches for a position in the search area R0 in the other image from which the template T0, which is a pixel block cut out from the predetermined image, has moved. Pixel 3 (△) is a pixel in template T0, and pixel 4 (〇) is a pixel in search area R0.

In FIG. 7, since the template T0 is superimposed on the search region R0, the pixel 4 (〇) in the portion where the template T0 overlaps in the search region R0.
Is omitted from the display. In FIG. 7, the size of the rectangular area of the template T0 and the size of the rectangular area of the search area R0 are both 4 pixels × 4 lines, but the size of both rectangular areas is a size other than the above 4 pixels × 4 lines. It may be. In this specification, the unit of the number of pixels in the horizontal direction is called a pixel, and the unit of the number of pixels in the vertical direction is called a line. One pixel is represented by a plurality of bits.

As a method of detecting the motion vector,
Conventionally, a full search block matching method is known. In this full search block matching method, first, the search area R0
, A pixel block 5 of 4 pixels × 4 lines is cut out, and the difference between the value of each pixel of the 4 pixels × 4 lines in the pixel block 5 and the value of the pixel existing at the corresponding position in the template T0 is determined for each pixel. Next, a difference absolute value or a difference square value is calculated, the sum of these calculated values is calculated, and then the pixel block 5 is moved left and right or up and down by one pixel. The difference between the value of each pixel of the pixel × 4 line and the value of the pixel existing at the corresponding position in the template T0 is calculated for each pixel by a difference absolute value or a difference square value. Find the sum. As described above, while sequentially moving the pixel block 5 left and right or up and down by one pixel, the value of each pixel of 4 pixels × 4 lines in the new pixel block 5 after the movement and the correspondence in the template T0 Then, the difference from the value of the pixel existing at the position to be calculated is calculated for each pixel by a difference absolute value, a difference square value, or the like, and an evaluation value that is a sum of these calculated values is obtained.

Then, it is determined that the template T0 has moved from the position corresponding to the pixel block having the smallest evaluation value. In this case, in the example of FIG.
Since 0 is composed of 11 pixels × 11 lines, the template T0 moves eight times in the horizontal direction and also moves eight times in the vertical direction.
Move times. That is, the number of pixel blocks 5 to be evaluated in comparison with the template T0 is 8 × 8 = 64 sets,
The number of evaluation values is 64.

In the example shown in FIG. 7, if the range of the motion vector is -4 to +3 in both the horizontal and vertical directions and, for example, the evaluation value in the pixel block 5 shown in FIG. (2, 2).

In the above-mentioned conventional example, an example is shown in which a motion is detected with single pixel accuracy (a method of detecting the movement when the image of interest moves in units of one pixel). It does not necessarily move in the direction or the vertical direction by an integer number of pixels, and in fact, between the block in the search area R0 where the motion is to be detected and the template T0,
In many cases, a shift of one pixel or less exists. Further, since what is used as the evaluation formula is not the essence of the present invention, a case where the absolute difference value is used as the evaluation formula will be described below for convenience.

In the above-described conventional full search block matching method, in order to evaluate an entire rectangular area, there are 64 sets of pixel blocks 5 to be evaluated in comparison with the template T0. Since pixels are included, it is necessary to calculate the absolute difference value 64 × 16 = 1024 times, and there is a problem that the amount of calculation is large.

In the above-described conventional full search block matching method, pixels in the search area R0 are usually stored in a memory, and transferred from this memory to a block matching device (a device for evaluating a motion vector). In this case, since the search area R0 is composed of 11 pixels × 11 lines, it is necessary to transfer 11 pixels × 11 lines = 121 pixels, and there is a problem that the transfer amount is large.

Therefore, the following method is known to reduce the amount of calculation and the amount of transfer. The method comprises a first step and a second step, wherein in the first step:
Sub-sampling for selecting one pixel from a plurality of pixels is performed, a reduced image is obtained by performing the sub-sampling in this way, motion is evaluated on the reduced image, and in a second step following the first step, In this method, the motion vector obtained in the first step is used as an initial displacement, and only a narrow range is searched on the image before sub-sampling.

FIG. 8 shows two pixels × 2 pixels in the above conventional example.
FIG. 7 is a diagram illustrating pixels extracted when one pixel is sub-sampled from a line.

In this conventional example, the template T0
From the search area R0, the pixel 6
(▲) is a pixel sub-sampled from the template T0, and a pixel 7 (●) is a pixel sub-sampled from the search area R0. Therefore, in the example shown in FIG. 8, in order to evaluate the entire rectangular area on the reduced image, there are 16 sets of areas to be evaluated in comparison with the template T0, and this set is 2 pixels × 2 Since it is composed of lines, it is necessary to perform 16 × 4 = 64 difference absolute value calculations.
It is necessary to transfer the search area pixels 7 (●) of 5 pixels × 5 lines = 25 pixels, which is the number of pixels. Further, when searching for -1 to +1 around the initial transition on the image before subsampling as the second step, the template T0
, The absolute value of the difference is calculated for the 16 pixels included in, and the template T0 is moved by three pixels in the horizontal direction and is moved nine times in total in the vertical direction for three lines to calculate the absolute value of the difference. = 144 difference absolute value calculations are required, and the search area pixels need to be transferred for 36 pixels × 6 lines = 36 pixels to be searched. The sum of the calculation amount and the transfer amount is 64 + 144 = 208 times, and 25 + 36 = 61 pixels.
The sub-sampling method of sub-sampling one pixel from a line greatly reduces the sum of the amount of calculation and the amount of transfer as compared with the above-described conventional full search block matching method.

However, in the above method of obtaining a motion vector from a reduced image, if the image contains noise,
There is a disadvantage that the detection accuracy is deteriorated.

FIG. 9 shows a conventional method proposed to eliminate the disadvantage that the detection accuracy is degraded when an image contains noise. The average value of 2 pixels × 2 lines = 4 pixels is calculated. FIG. 6 is a diagram showing a method of creating a reduced image by using the method.

When a motion vector is obtained using the reduced image, in order to prevent the detection accuracy from deteriorating when the image contains noise, two pixels × 2 lines = 4 pixels as shown in FIG. May be calculated, and a reduced image may be created based on this. Pixel 8 (▼) corresponds to template T0
Is a pixel having a value obtained by calculating an average value of the respective values of 2 pixels × 2 lines = 4 pixels above
(◎) is a pixel having a value obtained by calculating an average value of 2 pixels × 2 lines = 4 pixels on the search area R0.

The motion vector detection method is described in, for example, N. Hayashi, T. Kitsuki, I. Tamitani, H. Honma,
Y.Ooi, T.Miyazaki, K.Oobuchi, "A bidirectional mo
tioncompensation LSI with a compact motion estimat
or "IEICE Trans. Electron, E78-C, 12, pp. 1682-1690,
Dec. 1995.

The above-mentioned conventional example is an example in which a motion vector of single pixel precision is obtained after a motion vector of two pixel precision is obtained on a reduced image obtained by calculating an average value of four pixels. The literature describes an example in which a motion vector with half-pixel accuracy is obtained after obtaining a motion vector with two-pixel accuracy.

[0020]

As described above, the conventional full search block matching method has a problem that the amount of calculation is large, and a search area between a memory and a device for evaluating a motion vector is required. There is a problem that a large amount of pixel data is transferred.

FIG. 10 shows a search area R0 from which a motion vector is detected in the conventional subsampling method.
FIG. 11 is a diagram showing that there is a shift of up to one pixel in the horizontal direction and up to one line in the vertical direction between the middle reduced block BL and the template T0.

In the conventional full search block matching method,
Although there is only a shift of 0.5 pixel in the horizontal direction and 0.5 line in the vertical direction at the maximum, the sub-sampling method searches with an accuracy of 2 pixels, so the image has moved odd pixels in the horizontal direction and odd lines in the vertical direction. In this case, as shown in FIG.
There is a possibility that there is a displacement of one pixel in the horizontal direction and one line in the vertical direction between the reduced block BL in the search area R0 from which the motion vector is to be detected and the template T0. For this reason, the conventional sub-sampling method has a problem that an erroneous motion vector is easily detected as compared with the conventional full search block matching method.

Further, in the conventional sub-sampling method,
There is a problem that a step for obtaining a motion vector with single pixel accuracy is required separately.

According to the present invention, while maintaining the same search accuracy as the conventional full search block matching method, the calculation amount can be reduced to 1/4 of the calculation amount in the conventional full search block matching method. It is another object of the present invention to provide a block matching apparatus capable of suppressing the transfer amount of pixel data in a search area to be transferred from a memory to a block matching apparatus to one-fourth of a conventional full search block matching method. .

[0025]

According to the present invention, a reduced image is created by sub-sampling from a pixel in a search area every predetermined pixel in the horizontal direction and every predetermined line in the vertical direction, and a predetermined position in the template is determined. A plurality of reduced templates are created by sub-sampling the pixels existing in each pixel one by one, and for each reduced template, the difference in the value between the pixels at the corresponding positions between the reduced block cut out from the reduced image and each reduced template Is evaluated.

[0026]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a diagram showing a block matching device BM1 according to one embodiment of the present invention. FIG. 2 is a flowchart illustrating the operation of the block matching device BM1. FIG. 3 shows the search area R1 and the template T in the block matching device BM1.
FIG. 4 is a diagram showing a relationship with the number 1;

The block matching apparatus BM1 has a reduced image generation unit 30, a memory M, a reduced template creation unit 40, a reduced block evaluation unit 50, and a reduced template selection unit 60.

The reduced image creating means 30 is 11 pixels × 11
Sub-sampling is performed by sampling every two pixels in the horizontal direction and every two lines in the vertical direction from pixels in the line search area R1, thereby creating a reduced image SR1 of 5 pixels × 5 lines.

The memory M is a memory for storing data of a pixel in the search area R1, and is a memory for transferring the data of the pixel to the reduced block evaluation means 50 for evaluating a motion vector.

The reduced template creating means 40 divides a template of 4 pixels × 4 lines into small blocks of 2 pixels × 2 lines, and samples pixels existing at predetermined positions in the small blocks one by one. This is means for performing sub-sampling and thereby generating reduced templates ST1, ST2, ST3, ST4 of 2 pixels × 2 lines.
3, 44. Reduced template creation units 41, 4
2, 43 and 44 are reduced templates ST, respectively.
1, ST2, ST3, and ST4 are created.

The reduced block evaluation unit 50 generates a reduced image SR1 of 5 pixels × 5 lines created by the reduced image creation unit 30 for each of the reduced templates ST1, ST2, ST3, and ST4 created by the reduced template creation unit 40. This is a means for evaluating a difference in value between pixels at corresponding positions between a reduced block of 2 pixels × 2 lines cut out from the inside. The reduced block evaluation units 51, 52, 53, and 54 using reduced templates are Have. The reduced block evaluation units 51, 52, 53, and 54 evaluate the reduced blocks using reduced templates ST1, ST2, ST3, and ST4, respectively.

The reduced template selecting means 60 has the smallest evaluation value obtained by the reduced block evaluating means 50 among the four reduced templates ST1, ST2, ST3 and ST4 created by the reduced template creating means 40. This is a means for selecting a reduced template.

In general terms, the block matching device BM1 includes a template composed of pixel blocks of a pixels × b lines in a predetermined image and a c pixels × d lines (c ≧ a, d ≧ b)
A cut out from the image data of the search area composed of
This is a block matching device that evaluates a difference in value between pixels existing at corresponding positions between a pixel block of a pixel × b line.

Then, the reduced image creating means 30 performs sub-sampling by sampling every m pixels in the horizontal direction and every n lines in the vertical direction from the image data of the search area, and obtains (c / m) pixels × (d / N) is an example of a reduced image creating means for creating a reduced image of a line.

The reduced template creating means 40 divides the template into small blocks of m pixels × n lines,
Sub-sampling by sampling pixels existing at a predetermined position in the small block one pixel at a time to create m × n reduced templates of (a / m) pixels × (b / n) lines It is an example of the means.

The reduced block evaluation means 50 is for each of the reduced templates created by the reduced template creation means, (a / m) pixels × (b) cut out from the reduced image created by the reduced image creation means. /
n) An example of reduced block evaluation means for evaluating a difference in value between pixels at corresponding positions between a pixel block of a line and a pixel block of a line.

Next, the operation of the block matching device BM1 will be described.

In the flowchart shown in FIG. 2, first, a reduced image SR1 is created by sub-sampling predetermined pixels from the pixels constituting the search region R1. That is, the image data of the search area R1 is input (S
1) The pixel E11 shown in FIG. 3 is selected from the pixels constituting the search region R1, and the reduced image SR1 is created by sub-sampling the selected pixel E11.

Next, the reduced templates ST1, ST2,
ST3 and ST4 are created. That is, the template T1
(S11), the first reduced template ST1 is created by sub-sampling the pixel E1 from the pixels forming the template T1 (S12), and the pixel E2 is converted from the pixels forming the template T1. The second reduced template ST2 is created by sub-sampling (S13), and the third reduced template ST3 is created by sub-sampling the pixel E3 from the pixels forming the template T1 (S14). A fourth reduced template ST4 is created by sub-sampling the pixel E4 from the pixels constituting the template T1 (S15). Then, a function n indicating the number of evaluations is set to 1 (S1).
6).

FIG. 4 is a diagram showing an example of creating four reduced templates ST1 to ST4 in the block matching device BM1.

In the block matching device BM1,
The pixel 7 (•) is sub-sampled from the pixels in the search region R1 in the same manner as the sub-sampling method in the above-described conventional example, thereby creating a reduced image SR1. That is, a sub-sampled image SR1 of 5 pixels × 5 lines is created by subsampling every two pixels in the horizontal direction and every two lines in the vertical direction from the pixels in the search region R1 of 11 pixels × 11 lines.

On the other hand, the pixels in the template T1 are shown in FIG.
As shown in (1), first, pixels E1 (（), E2 (▽), and E3 constituting two pixels × 2 lines = 4 pixels adjacent to each other
Of the (★) and E4 (☆), the upper left pixel E1 (▲) is sub-sampled to create a reduced template ST1. Similarly, the upper right pixel E2 (▽), the lower left pixel E3 (★), and the lower right pixel E4 (☆) are sub-sampled, and the reduced templates ST2, ST3, ST
Create 4.

Next, the reduced image SR1 is evaluated using the reduced templates ST1 to ST4. In other words, to create a reduced block SI1 from the reduced image SR1 2 pixels × 2 lines (S20), evaluates the reduced block SI1 by the reduction template ST1, obtain the evaluation value EV ₁₁ (S2
1), reduced block S by reduced template ST2
Evaluate the I1, the evaluation value EV ₁₂ obtained (S22), evaluates the reduced block SI1 by the reduction template ST3, to evaluate the reduction unit SI1 the evaluation value EV ₁₃ obtained (S23), by the reduction template ST4, the obtaining an evaluation value EV ₁₄ (S24). Then, n is incremented by 1 (S25), the process returns to step S20, and the reduced image S
R1 to create a 2-pixel × 2 lines reduced block SI2, to obtain an evaluation value _{EV 21, EV 22, EV 23} , EV 24. Then, the operation of steps S20 to S25, repeated a predetermined number of times, evaluation value _{EV 31, EV 32, EV 33} , EV 34 ~E
_Vn1 , _EVn2 , _EVn3 , and _EVn4 are obtained (S26).

In the above embodiment, the reduced blocks SI1 to SIn are sequentially evaluated by each of the four reduced templates ST1 to ST4 to obtain an evaluation value, but one of the reduced blocks SI1 to SIn is obtained. , The evaluation using the four reduced templates ST1 to ST4 may be performed simultaneously.

Then, the reduced template selecting means 60
But selects the minimum evaluation value among the evaluation values EV ₁₁ ~EV _n4 (S31), and outputs the reduced block corresponding to the evaluation value of the minimum (pixel blocks) (S32), the above minimum evaluation value Select the motion vector of the corresponding reduced template. The reduced templates ST1, ST2, ST
3 and ST4 are (2x, 2
y), (2x + 1, 2y), (2x, 2y + 1), (2
x + 1,2y + 1). The range between x and y is
They are respectively -2 to +1.

In the above embodiment, the reduced template S
Operation for creating T1, ST2, ST3, ST4 (S11
To S15) are changed to the operation of creating the reduced image SR1 (S1,
It may be performed prior to S2), and the operations of creating reduced templates ST1, ST2, ST3, ST4 (S11 to S15) and the operations of creating reduced image SR1 (S1, S2) are performed simultaneously. You may do so.

FIG. 5 shows four reduced templates ST1, ST2, ST in the block matching device BM1.
FIG. 3 is a diagram illustrating that one of ST4 has a deviation of the same size as that of the full search block matching method.

FIG. 5 shows a case where there is a displacement of 0.3 pixels in both the horizontal and vertical directions between the reduced block BL1 in the search area R1 in which motion is to be detected and the reduced template ST4. In the above embodiment, although the number of evaluation pixels is reduced to 1/4 as compared with the full search block matching method, the size of the deviation between the reduced block SI1 and the reduced template ST4 is reduced by the full search block matching method. Is the same as the size of the displacement between the template and the pixel block when the search is made in step (1).

In the above embodiment, the reduced image is evaluated for each reduced template. In this case, in the above embodiment, among the pixels in the search area R1, the number of pixels in the horizontal direction is reduced by 1 /.
Since the reduced image is obtained by reducing the number of lines to ２ and reducing the number of lines by 、, when performing block matching on the reduced image, it is 1 more than block matching on the search region R1.
The evaluation is performed with the number of pixels of / 4. That is, in the above embodiment, the evaluation can be performed with 1/4 pixel number of the conventional full search block matching method.

Further, in the above embodiment, the magnitude of the displacement between one of the four reduced templates ST1 to ST4 and the reduced block SIn always corresponds to the template T0 and the pixel block in the full search block matching method. 5 is the same as the magnitude of the deviation. Therefore, the above embodiment can obtain the same level of accuracy as the conventional full search block matching method.

In the above conventional example, when an image contains noise, a low-frequency pass filter is formed by calculating an average value of 2 pixels × 2 lines = 4 pixels to create a reduced image. To prevent the search accuracy from deteriorating. Also in the above embodiment, if the image contains noise, the sub-sampling pixel E11 in the search region R1
When noise is included in (●), search accuracy deteriorates.

FIG. 6 is an explanatory diagram of a method of performing block matching by the same method as described above, using pixels having, as data, an average value of data of four adjacent pixels in the above embodiment.

That is, in the search area R1, of the pixels 22 obtained by averaging the data of four adjacent pixels and the pixels 23 obtained by averaging the data of four adjacent pixels in the same manner, Create a reduced image by subsampling.

On the other hand, the template T1 has a pixel 24 (▼) at the upper left, a pixel 25 (▽) at the upper right, a pixel 26 (★) at the lower left, and a pixel 2 at the lower right among 2 pixels × 2 lines = 4 pixels.
7 (☆ 27) is subsampled to create four reduced templates. The reduced images are evaluated simultaneously with these four reduced templates (there is no need to evaluate the reduced images at the same time), the evaluation values are obtained, and among these evaluation values, the reduced template having the smallest evaluation value is selected. Then, a pixel block is specified, and its motion vector is obtained.

According to the above embodiment, when a motion vector is obtained by block matching, the amount of calculation in block matching is reduced while the search accuracy in block matching is maintained at the same level as in the conventional full search block matching method. 1 / of the full search block matching method
4 can be reduced. In addition, the amount of pixels to be transferred from the memory M in the search area can be reduced to one-fourth that of the conventional full search block matching method, and a motion vector with single pixel accuracy can be directly obtained.

In the above embodiment, the low-pass filter is constructed by averaging the data of a plurality of adjacent pixels. May be adopted.

In each of the above embodiments, only the case of subsampling at 4: 1 has been described.
A reduced image or reduced block may be created by performing subsampling at another ratio such as 1, 16: 1. In this case, the same effect as above can be obtained. Further, in the above-described embodiment, an example has been described in which the accuracy is prevented from deteriorating by searching on an image composed of pixels having a value obtained by averaging data of four adjacent pixels.
The search may be performed by applying an arbitrary low-pass filter other than the above-mentioned low-pass filter, whereby deterioration in accuracy can be prevented.

[0058]

According to the first and second aspects of the present invention, in an apparatus for obtaining a motion vector by block matching, block search is performed while maintaining search accuracy similar to that of the conventional full search block matching method. The amount of calculation in matching can be reduced to 1/4 of the conventional full search block matching method, and the transfer amount of pixels in the search area can be reduced to 1/4 of the conventional full search block matching method. There is an effect that a motion vector with single pixel accuracy can be directly obtained.

According to the third aspect of the present invention, there is an effect that even if the image contains noise, the search accuracy does not deteriorate.

[Brief description of the drawings]

FIG. 1 is a diagram showing a block matching device BM1 according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating an operation of the block matching device BM1.

FIG. 3 is a diagram showing a relationship between a search region R1 and a template T1 in the block matching device BM1.

FIG. 4 is a diagram showing an example of creating four reduced templates ST1 to ST4 in the block matching device BM1.

FIG. 5 is a diagram for explaining that one of four reduced templates ST1, ST2, ST3, and ST4 has a displacement of the same size as that of the full search block matching method in the block matching device BM1.

FIG. 6 is an explanatory diagram of a method of performing block matching using pixels having, as data, a value obtained by averaging data of four adjacent pixels in the embodiment.

FIG. 7 is a diagram for explaining the operation of the conventional block matching device, and is a diagram showing a relationship between a template T0 to be subjected to motion detection and a search region R0.

FIG. 8 is a diagram showing an example in which 2 pixels × 2 lines correspond to 1
FIG. 7 is a diagram illustrating pixels extracted when sub-sampling pixels.

FIG. 9 is a conventional method proposed to eliminate a disadvantage that detection accuracy is deteriorated when an image includes noise, and calculates an average value of 2 pixels × 2 lines = 4 pixels to reduce a reduced image. FIG. 6 is a diagram showing a method of creating a.

FIG. 10 shows a conventional subsampling method, in which a displacement of up to one pixel in the horizontal direction and up to one line in the vertical direction between a reduced block BL in a search area R0 from which a motion vector is to be detected and a template T0. It is a figure showing that exists.

[Explanation of symbols]

 BM1: block matching device, 30: reduced image generation unit, M: memory, 40: reduced template creation unit, 41 to 44: reduced template creation unit, ST1 to ST4: reduced template, 50: reduced block evaluation unit, 51 to 54 ... Reduced block evaluator 60. Reduced template selecting means.

Claims (3)

[Claims] 1. A template composed of a pixel block of a pixel × b line in a predetermined image, and a c pixel × d line (c ≧ a, d ≧ b) in an image other than the predetermined image
A cut out from the image data of the search area composed of
In a block matching apparatus for evaluating a difference in a value between pixels existing at corresponding positions between a pixel block of a pixel × b line and a pixel block, the image data of the search area includes n pixels in the horizontal direction and n pixels in the vertical direction. A reduced image creating means for creating a reduced image of (c / m) pixels × (d / n) lines by sub-sampling by sampling every line; and dividing the template into small blocks of m pixels × n lines Sub-sampling by sampling pixels present at predetermined positions in the small block one by one,
(A / m) pixel × (b / n) reduced template creating means for creating m × n reduced templates; and reduced image creating means for each reduced template created by the reduced template creating means. (A / m) pixels cut out from the created reduced image ×
(B / n) a reduced block evaluation unit that evaluates a difference in value between pixels at corresponding positions between a pixel block on a line and a reduced block evaluation unit. 2. A reduced template having the smallest evaluation value obtained by said reduced block evaluation means is selected from said m × n reduced templates created by said reduced template creation means. A block matching apparatus, comprising: 3. The search area image according to claim 1, wherein, for each pixel included in the search area image, a search area image in which a high-frequency component is reduced by a low-pass filter; A block matching device, comprising: inputting a template whose high-frequency components have been reduced by a pass filter to the block matching device.