JP2015005233A - Image data matching processing method, processing device thereof, and image inspection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve efficiency and speed of matching process between a master image and an inspection object image.SOLUTION: A master image and an inspection object image G3 are divided into a plurality of blocks. For at least two blocks of G3(n+1, m), G3(n+2, m) of the inspection object image G3, a deviation amount between start point coordinates (n+1, m), (n+2, m) of the respective blocks of the inspection object image G3 and start point coordinates of the respective blocks of the master image corresponding to the respective blocks is acquired by calculation. When respective block data of the inspection object image G3 are memory-mapped in a simultaneous parallel calculation object space, a coordinate which is obtained by shifting the start point coordinate of the inspection object image G3 by an amount corresponding to a deviation amount is defined as load start addresses (n+1', m'), (n+2', m'), and from the load start address, the data of pixels of the respective blocks of the inspection object image G3 are loaded. Then, a matching process is simultaneously performed for the data of the pixels in the respective blocks.

Description

The present invention relates to a matching processing method of image data used for image processing, image matching, pattern matching, image inspection, and the like, a processing apparatus therefor, and an image inspection apparatus.
This technique can be applied to an image inspection device, a printed matter inspection device, a WEB inspection device, a liquid crystal inspection device, a semiconductor inspection device, and the like.

  Image of inspection object such as printed matter, semiconductor, etc., compare the inspection object image with master image, and check whether data of each pixel of inspection object image and corresponding pixel data of master image match A matching processing method for image data performed by processing is known.

  In this type of image data matching processing, it is necessary to perform alignment between the master image and the inspection target image. For example, the master image and the inspection target image are rectangular images, and the positions of the vertical and horizontal four corners are aligned, and it is necessary to inspect whether the data matches for each pixel of the images to be compared. .

However, even if one corner of the master image is aligned with one corner of the inspection target image, the remaining three corners often do not match.
This is because there is a difference in size between the master image and the image to be inspected due to the expansion, contraction, distortion, etc. of the image to be inspected, and the variation in the photographing distance during imaging, etc. This is because of changes.

  Therefore, the entire image area of the master image and the entire image area of the inspection target image are equally divided into a plurality of blocks corresponding to each other, and each pixel data of the master image and each of the inspection target images are divided for each block. Techniques for performing matching processing with pixel data have been developed (see Patent Document 1, Patent Document 2, and Patent Document 3).

  In Patent Documents 1 to 3, the entire positional deviation of the inspection target image with respect to the master image is converted into a positional deviation for each block, and the amount of positional deviation of each pixel in the block is assumed to be equal. By performing the processing, matching processing of data of each pixel is performed.

  According to these techniques, in order to eliminate the overall misalignment between the master image and the image to be inspected, the entire image area is divided into a plurality of image areas, and the misalignment amount for each block is collectively processed by a parallel computer. Since the correction is performed by processing, the entire inspection process can be made more efficient and faster.

  However, the size of the simultaneous parallel calculation target space of a parallel computer (for example, SIMD processor) actually used is finite. This is because the size of the simultaneous parallel calculation target space is realized as an LSI, and thus the size of the simultaneous parallel calculation target space is almost fixed.

  For example, assuming that the number of processor elements (PE) of the SIMD processor is 352, the number of pixels in the horizontal direction of each block (the number of pixels in the main scanning direction) is considered in consideration of the accuracy of inspection of the master image and the inspection target image. ) Is defined as 150 pixels.

  In this case, even if the size of the SIMD processor simultaneous parallel calculation target space, that is, the number of processor elements (PE) of the SIMD processor is 352, the positional deviation differs for each block, so at least two or more blocks. The pixel data in the memory cannot be mapped to the simultaneous parallel calculation target space and operated simultaneously.

  This is because if the position shift is different for each block, the same calculation processing cannot be applied to each block at the same time in the simultaneous parallel calculation target space, and only 150 pixels are processed simultaneously in one calculation. It is because it is not possible.

SIMD refers to a method of processing a plurality of data with one instruction, and SIMD processor refers to one of parallel computers that perform such processing.
Memory mapping is one of computer processes for associating a virtual address space with a physical address space.

  Generally, in the simultaneous parallel calculation processing of the SIMD processor having N processor elements (PE), N processor elements (PE) can be used, and the utilization efficiency in this case is 100%.

  However, in the above case, only 150 processor elements (PE) are used out of 352 processor elements (PE), so the usage efficiency is 150 / 352≈0.43, that is, about 43%. However, it becomes a bottleneck for improving the efficiency and speed of the matching process.

  The present invention equally divides the entire image area of the master image and the entire image area of the inspection target image into a plurality of blocks, and increases the efficiency and speed of matching processing between the master image and the inspection target image for each block. It is an object of the present invention to provide a matching processing method for image data that can achieve the above.

According to the image data matching processing method of the present invention, the entire image area of the master image and the entire image area of the inspection target image stored in the frame memory are equally divided into a plurality of blocks. For two or more blocks, a step of calculating a deviation amount between the start point coordinates of each block of the image to be inspected and the start point coordinates of each block of the master image respectively corresponding to each block;
When the pixel data of each block of the inspection target image is subjected to memory mapping from the frame memory to the simultaneous parallel calculation target space of a parallel computer, the start point coordinates of each block of the inspection target image are shifted by an amount corresponding to the shift amount. The coordinate data is used as a load start address, the pixel data of each block of the inspection target image is loaded from the load start address, and the pixel data in at least two blocks of the inspection target image are each corresponding to the master image. And simultaneously performing matching processing with pixel data in the block.

  According to the present invention, when loading a plurality of blocks of the inspection target image into the simultaneous parallel calculation target space, the load start address for each block is shifted to an address corresponding to the shift amount for each block. When the memory mapping of the pixel data in each block of the inspection target image is completed in the parallel calculation target space, the position shift of each block corresponding to the inspection target image with respect to each block of the master image is eliminated.

FIG. 1 is a block diagram of a processing apparatus used in the image data matching processing method according to the present invention, and shows a printer having an image inspection apparatus for inspecting an image of a printed matter. FIG. 2 is an explanatory diagram illustrating an example of an inspection target image printed by a printer. FIG. 3 is an explanatory diagram showing an example in which the image area of the inspection target image shown in FIG. 2 is equally divided into 8 blocks in the horizontal direction (main scanning direction) and equally divided into 18 blocks in the vertical direction (sub-scanning direction). . FIG. 4 is an explanatory diagram conceptually showing a positional deviation of the inspection target image with respect to the master image. FIG. 5 is an explanatory diagram conceptually showing misalignment correction for each block of the master image for each block of the inspection target image used in the conventional image data matching processing method. FIG. 6 is an explanatory view conceptually showing misalignment correction with respect to the block of the master image for each block of the inspection target image used in the image data matching processing method according to the present invention. It is explanatory drawing which shows notionally the memory mapping by loading the pixel data in the block of an image. FIG. 7 is an explanatory diagram conceptually showing the positional deviation correction of each block of the inspection target image used in the image data matching processing method according to the present invention, wherein the starting point coordinates of the block of the master image and the starting point coordinates of the inspection target image It is a conceptual diagram which shows typically the relationship. FIG. 8 is a flowchart showing an example of a conventional image data matching processing method. FIG. 9 is a flowchart showing an example of the image data matching processing method of the present invention.

A processing apparatus used in the image data matching processing method according to the present invention will be described below with reference to the printer shown in FIG.
In FIG. 1, reference numeral 1 denotes a printer body. The printer main body 1 includes a printer control unit 2, a print engine 3, and a mechanical mechanism (not shown).

  The original image data G1 is input to the printer control unit 2 of the printer main body 1, and the printer control unit 2 converts the original image data G1 into print data prd drawn by the print engine 3. With this print data prd, an image is printed on paper, and a printed matter 5 is produced.

  An image inspection device 6 is connected to the printer main body 1. The image inspection apparatus 6 includes a master image generation unit 7, a print image reading unit 8, a buffer memory 9, and a comparison inspection unit 10. The print image reading unit 8 is provided with a frame memory 8 '. The comparison inspection unit 10 has a SIMD computer as a parallel computer.

  The print data prd is sent to the master image generation unit 7 of the image inspection apparatus 6. The master image generation unit 7 generates a master image G2, reference point information Ref as a set of reference point coordinates described later, and alignment marks. The master image G2 and the reference point information Ref are buffered (temporarily saved).

  The printed matter 5 is scanned by the print image reading unit 8. This scan data is used as the inspection target image G3.

  The comparison inspection unit 10 collates and inspects the inspection target image G3 and the master image G2 that is buffered and synchronized with the page of the printed matter 5. The mismatch data between the master image G2 and the inspection target image G3 is output as an inspection result, for example, as “defect”.

  FIG. 2 shows a rectangular inspection object image G3 printed on the printed matter 5. Alignment marks 11 such as so-called “register marks” are printed at the four corners of the inspection target image G3. Using this alignment mark 11, alignment between the inspection target image G3 and the master image G2 is performed.

  In FIG. 3, the entire image area of the inspection target image G3 is equally divided into 8 in the horizontal direction (main scanning direction) and equally divided into 18 in the vertical direction (sub-scanning direction). The total number of pixels in the main scanning direction of the inspection target image G3 is, for example, 1200, and the total number of pixels in the sub-scanning direction, for example, 1800. Therefore, the size of each block G3 (i, j) is 150 × 100 pixels. Here, the symbol i is a positive integer from 1 to 8, and the symbol j is a positive integer from 1 to 18.

  As shown in FIG. 4, the inspection target image G3 is displaced in position with respect to the master image G2. For example, when the reference mark 11 ′ is the alignment mark of the master image G2, the alignment mark 11 of the inspection target image G3 is inspected even if the alignment mark 11 in the upper left corner is aligned with the alignment mark 11 ′. It gradually shifts due to distortion of the target image G3.

FIG. 4 conceptually shows a state in which the inspection target image G3 is shifted in the extending direction as it goes from the upper left corner to the lower right corner.
It can be assumed that the positional deviation of each block G3 (i, j) with respect to the entire inspection target image G3 is linear. Therefore, the positional shift of the master image G2 with respect to each corresponding block G2 (i, j) is caused by linearly shifting the block G3 (i, j) of the inspection target image G3 corresponding to the block G2 (i, j). Can be corrected. That is, the block G3 (i, j) of the inspection target image G3 can be matched with the corresponding block G2 (i, j) of the master image G2.

  In FIG. 4, symbols G3 (3, 5), G3 (4, 5), G3 (5, 5), and G3 (6, 5) indicate blocks in the central portion of the inspection target image G3. In this example, the position of the master image G2 with respect to each corresponding block G2 (3, 5), G2 (4, 5), G2 (5, 5), G2 (6, 5) as it goes from left to right. It shows that the deviation gradually increases.

  In FIG. 4, reference symbols G2 (1,1), G3 (1,1), G2 (8,1), G3 (8,1), G2 (1,18), G3 (1,18), G2 (8, 18) and G3 (8, 18) indicate four corner blocks having the alignment mark 11.

The displacement of each block G3 (i, j) of the inspection target image G3 with respect to the corresponding block G2 (i, j) of the master image G2 can be calculated by the following equation, for example.
With the horizontal direction as the X axis and the vertical direction as the Y axis, for example, the reference point coordinates (reference point information) of the block G2 (1, 1) of the master image G2 in the X axis direction are (x1, y1), and the block G2 (8 The reference point coordinates of 1) are (x2, y1), the reference point coordinates of block G2 (1, 18) are (x1, y2), and the reference point coordinates of block G2 (8, 18) are (x2, y2). To do.

  On the other hand, for example, the displacement amount in the x direction of the block G3 (1, 1) of the inspection target image G3 is u1, the displacement amount in the x direction of the block G3 (8, 1) is u2, and the block G3 (1, 18). ) In the x-direction and u4 in the x-direction of the block G3 (8, 18).

Here, Δ (x, y) is defined as a positional deviation amount in the x direction at the coordinate position (x, y).
The positional deviation amount Δ (x, y1) at the position x in the first row is assumed to be linear.
Then, the displacement Δ (x, y1) is
Δ (x, y1) = {(u2−u1) / (x2−x1)} * (x−x1) + u1 = f1 (x)
Similarly, the positional deviation amount Δ (x, y2) at the position x of the 18th row is:
Δ (x, y2) = {(u4−u3) / (x2−x1)} * (x−x1) + u3 = f2 (x)

Therefore, the amount of displacement in the x direction at the x position is
Δ (x, y) = [{f2 (x) −f1 (x))} / (y2−y1)] * (y−y1) + f1 (x)
The symbol “*” means multiplication.

  On the other hand, the amount of displacement in the y direction at the y position is similarly determined as follows. The displacement amount in the y direction of the block G3 (1, 1) of the inspection target image G3 is v1, the displacement amount in the y direction of the block G3 (8, 1) is v2, and the displacement amount of the block G3 (1, 18) in the y direction. Is v3, and the shift amount in the y direction of the block G3 (8, 18) is v4.

Assuming that the misalignment amount at the position y in the first row is linear, the misalignment is linear.
Δ (x1, y) = {(v2−v1) / (y2−y1)} * (y−y1) + v1 = f1 (y)
Similarly, the amount of misalignment at the position x of the tenth row is
Δ (x2, y) = {(v4−v3) / (y2−y1)} * (y−y1) + v2 = f2 (y)
Therefore, the amount of displacement in the y direction at the y position is
Δ (x, y) = [{f2 (y) −f1 (y))} / (x2−x1)] * (x−x1) + f1 (y)

This linear interpolation formula is used when determining the amount of positional deviation in the x and y directions of each block of the inspection target image G3 that does not have the alignment mark 11.
That is, the reference point coordinates (x1, y1), (x2, y2), (x3, y3), (x4, y4) at the four corners of the master image G2 are used as the four-corner block G2 (1, 1) having the alignment mark 11. , G2 (8,1), G (1,18), G2 (8,18), and the reference point coordinates of the four corners of the inspection target image G3, which is a printed matter, are the blocks G3 (1,1, ), G3 (8, 1), G3 (1, 18), and G3 (8, 18), the positional deviation amounts u1, u2, u3, u4 in the x direction, and the positional deviation amounts v1, v2 in the y direction. , V3, v4.
Thereafter, using the linear interpolation formula for obtaining the positional deviation amount in the x direction and the linear interpolation formula for obtaining the positional deviation amount in the y direction, the positional deviation in the x direction for each block of the inspection target image G3 not having the alignment mark 11 is used. The amount of displacement in the y direction is obtained.
Since the matching process with the positional deviation correction is the same process repeated for each pixel, it can be efficiently executed using a parallel computer such as a SIMD-structured processor.
FIG. 5 conceptually shows a conventional example of processing for part of the inspection object image G3 and the positional deviation.

The shift amount Δ (x, y) of each pixel can be assumed to be the same for the pixels in the block G3 (i, j) of the same inspection target image G3. This is established by defining a block having such a size that the shift amount can be assumed to be the same. Simultaneous parallel calculation can be performed if the pixel position shifts in the same block G3 (i, j) are equal.

Conventionally, this parallel calculation processing is performed according to the following processing procedure, and image data matching processing is performed.
That is, the addresses of the start point coordinates (x1, y1), (x2, y2), (x3, y3), (x4, y4) of the master image G2 are stored in the simultaneous parallel calculation target space (register file, processor element) of the parallel computer. The pixel data for one block is loaded with the address of the inspection target image G3 corresponding to (1) as the load start address (see S.1 ′ in FIG. 8).

  Next, in the simultaneous parallel calculation target space, as conceptually shown in FIG. 5, blocks G3 (i, j), G3 (i + 1, j), G3 (i + 2, j), G3 ( The pixel data is shifted for each block in the x and y directions by an amount corresponding to the positional deviation amounts (Δx1, Δy1), (Δx2, Δy2), (Δx3, Δy3), (Δx4, Δy4) of i + 3, j). (See S.2 ′ in FIG. 8).

  Next, the data of each pixel of the blocks G3 (i, j), G3 (i + 1, j), G3 (i + 2, j), G3 (i + 3, j) of the inspection target image G3 and the corresponding block G2 ( i, j), G2 (i + 1, j), G2 (i + 2, j), and data of each pixel corresponding to G2 (i + 3, j) are subjected to matching calculation for each block (S3 ′ in FIG. 8).

There are various methods of pixel data matching processing, such as a simple difference method and a correlation calculation method using a small area at the center of the pixel, but the description thereof is omitted here.
The accumulation of the matching results of the pixel data in the block G3 (i, j) is the inspection result for the entire image in the block G3 (i, j). If the difference between the master image G2 and the inspection target image G3 is greater than or equal to a predetermined value, it is determined that the inspection is disqualified.

  In this conventional image data matching processing method, as shown in FIG. 5, for each block G3 (i, j) of the inspection target image G3, a positional shift with respect to each corresponding block G2 (i, j) of the master image G2 The amount is different.

  For example, the start point coordinates of the block G3 (i, j) of the inspection target image G3 with respect to the start point coordinates (x1, y1) of the block G2 (i, j) of the master image G2 are (x1 + Δx1, y1 + Δy1).

For example, the start point coordinates of the block G3 (i + 1, j) of the inspection target image G3 with respect to the start point coordinates (x2, y2) of the adjacent block G2 (i + 1, j) of the master image G2 are (x2 + Δx2, y2 + Δy2).

Similarly, the start point coordinates of the block G3 (i + 2, j) of the inspection target image G3 with respect to the start point coordinates (x3, y3) of the block G2 (i + 2, j) of the master image G2 are (x3 + Δx3, y3 + Δy3).

  The start point coordinates of the block G3 (i + 3, j) of the inspection target image G3 with respect to the start point coordinates (x4, y4) of the block G2 (i + 3, j) of the master image G2 are (x4 + Δx4, y4 + Δy4). That is, the positional deviation amounts (Δx1, Δy1), (Δx2, Δy2), (Δx3, Δy3), (Δx4, Δy4) are different for each block G3 (i, j) of the inspection target image G3.

  Therefore, when the pixel data of the block G3 (i, j) of the inspection target image G3 to be loaded with respect to the number N of elements of the processor element (PE) is smaller than N / 2, the position from the inspection target image G3 Data of pixels of at least two blocks G2 (i, j) having different shift amounts are loaded into the simultaneous parallel calculation target space, and two blocks G3 ( Even if the same calculation process is performed simultaneously on i, j), the calculation process cannot be performed.

  That is, when the pixel data of the block G3 (i, j) of the inspection target image G3 to be loaded with respect to the number N of elements of the processor element (PE) is smaller than N / 2, the simultaneous parallel calculation target space Usage efficiency cannot be achieved.

  On the other hand, in the Example which concerns on this invention, the test object image G3 is imaged by imaging devices CAM, such as a camera, for example. The image data is temporarily stored in the frame memory 8 'for each block G3 (i, j) as shown in FIG.

  In FIG. 6, each block G3 (i, j) is continuously stored in the frame memory 8 ′. However, if the address of each block G3 (i, j) is known, each block G3 (i, j) ) Can be associated, the storage order is not limited to this.

  When managing the inspection target image G3 stored in the frame memory 8 ', each block G3 (i, j) is given a start address (starting point coordinate) corresponding to the block division of the master image G2. Since the size of the frame memory 8 'and the number of vertical and horizontal pixels of the image are determined, the start address can be calculated in advance.

  On the other hand, the positional deviation Δ (x, y) generated for each block G3 (i, j) of the inspection target image G3 is separately calculated in advance for each block G3 (i, j). The calculation method is as described above. From the position shift Δ (xi, xj) of each block G3 (i, j) with respect to the block G2 (i, j) obtained by this calculation, the start address after the position shift of each block G3 (i, j) is calculated. Can be sought.

  When this block G3 (i, j) is loaded into the simultaneous calculation target space of the processor element PE of the SIMD processor during the matching process, the block G3 (i, j) is read and loaded from the start address after this misalignment correction. As a result, if the pixel data is subjected to memory mapping in the simultaneous parallel calculation target space, it can be loaded at the position where the positional deviation of the block G3 (i, j) has been corrected at the end of the memory mapping.

  In this case, if there is a free space for the processor element (PE) that can load another individual block G3 (i, j) of the inspection target image G3 in the SIMD space, another block G3 (i + 1, j) can be loaded.

  When loading another block G3 (i + 1, j), if loading is performed from the load start address after the positional deviation correction, the loading can be completed at the position where the positional deviation has been eliminated for this block G (i + 1, j).

  Thereafter, the pixel data in the block G2 (i, j) of the master image G2 and the pixel data in the block G3 (i, j) of the inspection target image G3 are matched. At the same time, the pixel data in another block G2 (i + 1, j) of the master image G2 and the pixel data in another block G3 (i; 1, j) of the inspection target image G3 are matched.

  Thereby, when the number of elements of the pixel to be processed is N / 2 or less, matching processing can be performed on at least two or more blocks at the same time, so that the utilization efficiency and speed of the parallel computer can be improved.

The processing procedure will be described in detail below.
First, the inspection target image G3 is divided into blocks corresponding to the block division of the master image G2, and the start address (start point coordinates) of each block G2 (i, j) of the master image G2 and the start address of the inspection target image G3 Association with G2 (i, j) is performed.

  FIG. 6 shows a block G3 (n + 1) of the inspection target image G3 corresponding to the start point coordinates (n + 1, m) and (n + 2, m) of the blocks G2 (n + 1, m) and G2 (n + 2, m) of the master image G2. , M), G3 (n + 2, m) start point position coordinates (n + 1, m), (n + 2, m) are shown.

  Next, a positional deviation amount Δ (xi, xj) of each block G3 (i, j) of the inspection target image G3 with respect to the corresponding block G2 (i, j) of the master image G2 is calculated (see S1 in FIG. 9).

  A load start address for loading each block into the simultaneous parallel calculation target space is calculated from the positional deviation amount Δ (xi, xj) of each block G3 (i, j) of the inspection target image G3 obtained in step S1. (See S2 in FIG. 9).

FIG. 7 shows an example of the load start address after the positional deviation correction.
In FIG. 7, the start point coordinates (x1 ′ = x1 + Δ (x1) of the block G3 (i, j) of the inspection target image G3 corresponding to the start point coordinates (x1, y1) of the block G2 (i, j) of the master image G2. , Y1), y1 ′ = y1 + Δ (y1, x1)), the start point coordinates (x2) of the block G3 (i + 1, j) of the inspection target image G3 corresponding to the start point coordinates (x2, y2) of the block G2 (i + 1, j) '= X2 + Δ (x2, y2), y2' = y2 + Δ (y2, x2)), the start point coordinates (x3, y3) of the inspection target image G3 corresponding to the start point coordinates (x3, y3) of the block G (i + 2, j) of the master image G2. x3 ′ = x3 + Δ (x3, y3), y3 ′ = y3 + Δ (y3, x3)), block G3 of inspection target image G3 corresponding to start point coordinates (x4, y4) of block G2 (i + 3, j) of master image G2 (I + , The start point coordinates j) (x4 '= x4 + Δ (x4, y4), y4' is = y4 + Δ (y4, x4)).

Next, one block G (i, j) of the inspection target image G3 is loaded from the load start address into the simultaneous parallel calculation target space and memory-mapped (see S3 in FIG. 9).
Subsequently, it is determined whether or not another block G3 (i + 1, j) can be mapped to the simultaneous parallel calculation target space (see S4 in FIG. 9). When another block G3 (i + 1, j) can be mapped to the simultaneous parallel calculation target space, this mapping is continued (see S5 in FIG. 9).

  After executing step S5, when it is determined in step S4 that another block G3 (i + 1, j) cannot be mapped, the two or more blocks BG (i, j) mapped in the simultaneous parallel calculation target space Matching processing is simultaneously performed on the pixel data (see S6 in FIG. 9).

  In this embodiment, the processor element (number of PEs) of SIMD is 352, whereas the main scanning width of one block G3 (i, j) of the inspection target image G3 is 150, so FIG. As shown in FIG. 2, two blocks G3 (n + 1, m) and G3 (n + 2, m) can be memory-mapped in the simultaneous parallel calculation target space.

  This is because even if the data of the pixels of the two blocks is subjected to memory mapping in the simultaneous parallel calculation target space, the data of 300 pixels is obtained, which is smaller than the number 352 of processor elements PE.

  As described above, in the conventional matching processing, the SIMD calculation processing efficiency is about 43%, but according to this embodiment, two blocks G3 (i, j) of the inspection target image G3 are matched simultaneously. Since the processing can be performed, the processing efficiency is about 85%, and the speed can be effectively increased by parallel calculation processing.

  According to this embodiment, when a plurality of blocks G3 (i, j) of the inspection target image G3 are loaded into the processor element PE of the SIMD processor, the start address of the block G3 (i, j) is set to the block G3. It shifted to the position according to the deviation | shift amount for every (i, j).

For this reason, when the memory mapping is completed, the positional deviation corresponding to the block G2 (i, j) of the master image G2 has been eliminated, and data matching processing can be simultaneously performed for each subsequent pixel. it can.
That is, matching of pixel data for each block can be performed simultaneously for at least two or more blocks of the image to be inspected, and matching processing can be made more efficient and faster.

8 '... frame memory G2 ... master image G3 ... inspection object image G2 (i, j), G3 (i, j) ... block (n + 1, m), (n + 2, m), (n + 1', m '), ( n + 2 ', m') ... start point coordinates

JP 2001-56301 A JP 2001-56302 A Patent No. 3299066

Claims (5)

The entire image area of the master image and the entire image area of the inspection target image stored in the frame memory are equally divided into a plurality of blocks, and the inspection target image is obtained for each of at least two blocks of the inspection target image. A calculation step for calculating a deviation amount between the start point coordinates of each block and the start point coordinates of each block of the master image respectively corresponding to each block;
When the pixel data of each block of the inspection target image is subjected to memory mapping from the frame memory to the simultaneous parallel calculation target space of a parallel computer, the start point coordinates of each block of the inspection target image are shifted by an amount corresponding to the shift amount. The data of the pixels of each block of the inspection target image is loaded from the load start address using the position coordinates as the load start address, and the data of the pixels in at least two blocks of the inspection target image correspond to the master image. And a matching processing step for simultaneously performing matching processing with pixel data in each block.   The matching processing step is an address for calculating a load start address after misalignment correction when loading each block into the simultaneous parallel calculation target space from the misalignment amount of each block of the inspection target image obtained in the calculation step. A calculation step, a memory mapping step of memory mapping one block of the image to be inspected to the simultaneous parallel calculation target space, and subsequently determining whether another block can be memory mapped to the simultaneous parallel calculation target space An execution step of continuously executing the memory mapping when the other block can be memory-mapped to the simultaneous parallel calculation target space, and after executing the address calculation step and the execution step, If the other block cannot be memory mapped Matching an image data processing method according to claim 1, characterized in that simultaneously matching processing for the data of the pixels of each block is already memory mapped to the same time parallel computing target space when it is disconnected.   The image data matching processing method according to claim 2, wherein the parallel computer has a SIMD structure.   A processing apparatus capable of executing the image data matching processing method according to claim 1.   An image inspection apparatus using the image data processing method according to claim 1.