JP2002183733A - Pattern collation method and device by image processing - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce false recognition caused by magnitude unevenness of an inspection target when collating a binarized image including the inspection target with a master image. SOLUTION: In this device, a comparison decision means 5 collates the master image stored in a master image memory 4 with the binarized image obtaining by picking up an image of a space including the inspection target by a camera 1 and binarizing the image, generates a difference image between the binarized image and the master image, calculates the number of blobs including the difference image and an area and an oblateness in each the blob, eliminates the blobs each having the oblateness not less than a prescribed threshold value from the inspection target, and decides that the inspection target is not coincident with the master image when at least one of the areas and the number of the residual blobs is not less than a prescribed threshold.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention mainly relates to a pattern such as a character or a figure formed on an article by stamping, printing, or engraving in a production line, and checks the pattern with a master image using an image processing technique. The present invention relates to a pattern matching method by image processing for judging coincidence / non-coincidence and a device therefor.

[0002]

2. Description of the Related Art Conventionally, in production lines, stamping,
A grayscale image obtained by imaging a space including a pattern such as a character or a figure formed on an article by printing, engraving or the like as an inspection target with an image input unit such as a TV camera is binarized with respect to a density value. Various image processing techniques have been proposed to generate a binarized image, and to perform a collation process for checking whether or not the pattern is chipped or dirty based on the binarized image, and a recognition process for specifying the pattern. Have been.

In this type of collation processing and recognition processing,
The pixel group to be inspected in the binarized image is compared with the template of the master image to determine the degree of coincidence, and when the degree of coincidence is sufficiently high, it is recognized that the inspected object and the template are equal. . Therefore, when the pattern to be inspected is already known, a template having the same contents as the inspection object is prepared, so that it is possible to perform the collation for extracting dirt or chipping of the inspection object from the degree of coincidence. Also,
When the pattern to be inspected is unknown, the target to be inspected can be recognized from the degree of coincidence by comparing a large number of candidates with the target to be inspected.

As described above, the pixel group to be inspected in the binarized image must be collated with the template. Therefore, in order to match the position with the template, the position of the center of gravity of the inspected object is determined, the principal axis angle is determined, and the template is determined. It is necessary to make the center of gravity of the inspection target coincide with the position of the center of gravity of the inspection target, and to match the principal axis of the inspection target (a straight line in the longitudinal direction passing through the center of gravity) with the principal axis of the template. In addition, if the inspection target is a known inspection target, the conditions for imaging the inspection target are constant, so it is not necessary to match the size of the inspection target and the template, but it is also necessary to match the size of the unknown inspection target. is there.

After the conditions for matching the inspection object with the template are set as described above, a difference image between the binary image and the master image is generated. That is, the difference between the inspection target and the template is obtained. Here, the width of the inspection target often differs from that of the template depending on binarization conditions (illumination conditions and threshold values for binarization) and the like. Even if it can be considered that there is a difference, a difference often occurs in the difference image.

In view of this, a set including a specified number of pixels or more (hereinafter, such a set of pixels is referred to as a blob) among a set of pixels having a difference in the difference image is labeled, and the number of blobs subjected to the labeling is determined. And the area (number of pixels) for each blob. If the number of blobs is equal to or greater than the specified threshold or if the area of the blob is equal to or greater than the specified threshold, it is estimated that the inspection target and the template do not match.

That is, if a shape as shown in FIG. 3A is required as a template of the master image M, and if a shape as shown in FIG. A difference image D between the template of M and the inspection target X is as shown in FIG. Here, since the inspection target X is missing from the template of the master image M, when the difference image D as shown in FIG. It must be determined that they do not match. Therefore, by setting the threshold value with a smaller number of pixels than the number of pixels in the difference image D as shown in FIG. 3C as a threshold value for the area of the blob, FIG.
When the difference image D as shown in (c) occurs, it can be determined that the inspection image X and the template of the master image M do not match.

[0008]

As described above, since the size of the inspection object varies depending on the binarization condition and the like, a difference is likely to occur near the contour of the inspection object in the difference image. Therefore, the area value of the blob in the difference image may be relatively large, and even if the inspection image can be considered to be actually coincident with the master image, the area of the blob in the difference image is increased and it is determined that the area does not match. Is more likely to occur. In addition, the variation in the size of the inspection target is not only caused by the binarization condition, but also caused by the variation in the dimension of the pattern to be inspected. May lead to a shift in the position of the center of gravity or the principal axis angle of the inspection target, and as a result, the difference from the template increases, and the probability of erroneously identifying a match as a mismatch increases.

The present invention has been made in view of the above circumstances, and has as its object to reduce false recognition due to the size of an inspection target when comparing a binary image including the inspection target with a master image. An object of the present invention is to provide a pattern matching method and an apparatus thereof by image processing.

[0010]

According to a first aspect of the present invention, a binary image obtained by imaging a space including an object to be inspected by an image input means and binarizing the same with a master image stored in a master image memory is compared. In a pattern matching method based on image processing, a difference image between a binarized image and a master image is generated by a comparison and determination unit, and then the number of blobs included in the difference image and the area and flatness of each blob are generated by the comparison and determination unit. The blob whose flattening rate is equal to or greater than a specified threshold is excluded from the inspection target, and the inspection target does not match the master image if at least one of the number and area of the remaining blob is equal to or greater than the specified threshold. Is determined. According to this method, the oblate ratio for each blob formed in the difference image is obtained, and the blobs whose oblate ratio is equal to or greater than a specified threshold are excluded from the inspection target, and the inspection target is determined by the number of blobs and the area of each blob. Since the match / mismatch with the master image is determined, it is possible to almost eliminate the blob variation element due to the variation in the size of the test object, and to avoid the misrecognition that the test object which can be considered to match the master image is determined to be mismatch. It is greatly reduced.

According to a second aspect of the present invention, there is provided a pattern matching by image processing for comparing a binary image obtained by taking an image of a space including an object to be inspected by image input means and binarizing the image and a master image stored in a master image memory. In the method, a mask area to be excluded from the inspection target among the binarized images is set based on the master image, and a difference image between the binarized image excluding the mask area and the master image is generated by the comparison determination unit. The comparison determination means obtains the number of blobs included in the difference image and the area of each blob, and determines that the inspection target does not match the master image if at least one of the number of blobs and the area is equal to or greater than a specified threshold. It is characterized by doing. According to this method, the mask region generated from the master image is excluded from the inspection target, and the matching between the inspection target and the template of the master image is performed using the difference image between the binary image excluding the mask region and the master image. Since the mismatch is determined, it is possible to substantially eliminate the fluctuation element of the blob due to the variation in the size of the inspection target, and the erroneous recognition of determining that the inspection target that can be considered to match the master image as the mismatch is greatly reduced.

According to a third aspect of the present invention, there is provided an image input means for capturing an image of a space including an object to be inspected, a master image memory storing a master image, and two images captured by the image input means.
A comparison / determination unit for generating a difference image between the binarized image and the master image, wherein the comparison / determination unit obtains the number of blobs included in the difference image and the area and flattening factor for each blob, Blobs whose rate is equal to or more than a specified threshold are excluded from the inspection target, and when at least one of the number and the area of the remaining blob is equal to or more than the specified threshold, the inspection object is determined to be inconsistent with the master image. . According to this configuration, the oblate ratio for each blob formed in the difference image is obtained, and the blobs whose oblate ratio is equal to or more than the specified threshold are excluded from the inspection target, and then the inspection target is determined by the number of blobs and the area of each blob. Since the match / mismatch with the master image is determined, it is possible to almost eliminate the blob fluctuation element due to the variation in the size of the test object, and to avoid the misrecognition such that the test object which can be considered to match the master image is determined to be mismatch. It is greatly reduced.

According to a fourth aspect of the present invention, there is provided an image input device for capturing an image of a space including an object to be inspected, a master image memory storing a master image, and two images captured by the image input device.
A mask area setting unit that sets a mask area to be excluded from the inspection target in the binarized image based on the master image, and a comparison that generates a difference image between the binary image excluding the mask area and the master image Determination means, and the comparison determination means obtains the number of blobs and the area of each blob included in the difference image, and when at least one of the number and the area of the blob is equal to or more than a specified threshold, the inspection object is determined to be a master image. It is determined that they do not match. According to this configuration, the mask region generated from the master image is excluded from the inspection target, and the matching between the inspection target and the master image template is performed using the difference image between the binary image excluding the mask region and the master image. Since the mismatch is determined, it is possible to substantially eliminate the fluctuation element of the blob due to the variation in the size of the inspection target, and the erroneous recognition of determining that the inspection target that can be considered to match the master image as the mismatch is greatly reduced.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (First Embodiment) In this embodiment, a pattern (character or graphic) to be inspected is collated using the apparatus shown in FIG. That is, a camera (TV camera) 1 is provided as an image input unit for capturing an image of a space including a pattern to be inspected, and an image signal output from the camera 1 is converted into a digital signal by an A / D converter 2. The output of the A / D converter 2 is stored in the image memory 3 as an original image, and is compared with the master image stored in the master image memory 4 in advance by the comparison determination means 5. The comparison determination means 5 generates a binary image by binarizing the original image stored in the image memory 3, compares and compares the binary image with the master image, and outputs the comparison result to the external output means. Output through 6. The original image stored in the image memory 3 is converted into an analog signal by a D / A converter 7 and can be displayed on a display device 8 such as a CRT. Although not shown, the display device 8 can also display a master image and images such as a binary image and a difference image generated by the comparison / determination unit 5 as necessary.

In this embodiment, the comparison / judgment means 5 generates a binarized image, obtains the position of the center of gravity and the principal axis angle of the inspection object in the binarized image, and performs alignment with the template in the master image. The steps up to the step of generating a difference image between the inspection target and the template are the same as in the conventional method. Here, according to the conventional method, the degree of coincidence between the inspection target and the template is determined based on two pieces of information of the number of blobs (sets of connected regions of pixels) formed in the difference image and the area of each blob. On the other hand, the present embodiment is different in that the flatness of each blob is obtained, and the flattening rate is also used as a determination condition in addition to the two information used in the conventional method.

The flattening rate in the present embodiment is defined so that a blob having a large area in the contour portion of the pattern has a larger flattening ratio than a blob generated in a portion other than the contour portion. For example, the blob B1 generated at the contour of the pattern has a shape as shown in FIG.
The shape is as shown in FIG. Therefore, the flattening ratio is defined so that the flattening ratio of the blob B1 is higher than that of the blob B2.

The flattening rate is determined by the area Sf of the rectangular area having the same fillet width as the blob and the area Sf of each blob.
Or the ratio of the maximum width in the horizontal direction (x fillet) and the maximum width in the vertical direction (y fillet) of the above-described rectangular area divided by the smaller one. Or can be specified. The flattening ratio according to the former rule corresponds to the reciprocal of the occupying ratio of the blob in the rectangular area. The latter flatness is based on a more general definition. In any case, the flattening ratio may be defined so that the flattening ratio of the blob B1 shown in FIG. 2 is larger than the flattening ratio of the blob B2.

Incidentally, the blob of the contour of the pattern is considered to be caused by a dimensional error between the master and the inspection target due to the binarization condition of the inspection target and the accuracy of the pattern itself. Even if the blob is excluded, there is little possibility of erroneous recognition of the pattern. Therefore, the comparison and determination means 5 compares the flattening rate with a prescribed threshold, excludes the blob whose flattening rate is equal to or more than the threshold from the inspection target, and sets only the remaining blobs to the number of blobs in the same manner as the conventional method. The area of each blob is compared with a threshold.

Since the blob generated near the contour of the pattern can be excluded by the above-described processing, the probability that the inspection target matches the master image is increased, and the pattern is good even though the pattern is free from chipping or contamination. It is possible to reduce the possibility of being determined to be inconsistent. In addition, the blob to be excluded is only in the vicinity of the contour of the pattern, and the possibility of erroneously recognizing a defective product (mismatch pattern) as a good product (matching pattern) hardly increases as compared with the conventional method. Other configurations and procedures are the same as those of the conventional method, and thus description thereof is omitted.

(Second Embodiment) In this embodiment, a mask area generated based on a master image is used in order to remove blobs generated near the contour of a pattern. The mask area is set by extracting an edge from the master image and moving the edge appropriately. For example, a mask area is set by extending an edge outward. Here, the edge is extracted by differentiating the master image and tracking the edge using the differential value and direction value, extracting the edge from the binary image of the master image, and extracting the edge by gray search of the master image. A method of extracting, a method of extracting an edge by Hough-transforming the master image, a method of displaying the master image on the display device 8 and manually specifying an edge surrounding the mask region, or the like can be adopted.

When the mask area is set based on the master image as described above, the mask area is stored in the master image memory 4.
Registered in. In the comparison determination means 5, the master image memory 4
Then, a difference image from the master image is generated by excluding the mask region from the binarized image obtained by binarizing the original image stored in the image memory 3 using the mask region registered in the master image. In the difference image generated in this way, since the blob in the contour portion of the inspection target no longer appears, it is only necessary to determine the presence or absence of the difference only for the blob that is the actual difference between the inspection target and the template. It is possible to enhance the stability of the pattern matching result. In other words, since there is no blob near the contour of the pattern, the probability that the inspection target matches the master image is high, and the possibility that the pattern is judged to be non-coincidence even though it is a non-defective and clean product is reduced. can do. Moreover, the range excluded by the mask region is only the vicinity of the contour of the pattern, and the possibility of erroneously recognizing a defective product (unmatched pattern) as a good product (matching pattern) hardly increases compared to the conventional method. Other configurations and procedures are the same as those of the conventional method, and thus the description is omitted.

[0022]

According to the first aspect of the present invention, there is provided a binarized image obtained by taking an image of a space including an inspection object by image input means and binarizing the image.
In a pattern matching method based on image processing for matching a master image stored in a master image memory, a difference image between a binarized image and a master image is generated by a comparison determination unit, and then, the comparison determination unit Determine the number of blob included, the area and the flatness of each blob,
Blobs whose flatness is equal to or greater than the specified threshold are excluded from the inspection target, and it is determined that the inspection target does not match the master image if at least one of the number and area of the remaining blob is equal to or higher than the specified threshold. Obtain the flatness of each blob formed in the difference image, and remove the blob whose flatness is equal to or more than the specified threshold from the inspection target, and then match the inspection target with the master image by the number of blobs and the area of each blob. -Since the mismatch is determined, it is possible to substantially eliminate the blob fluctuation element due to the variation in the size of the inspection target, and the erroneous recognition such as determining the inspection target that can be regarded as matching the master image as the mismatch is greatly reduced. There is an advantage.

According to a second aspect of the present invention, there is provided a pattern matching by image processing for comparing a binarized image obtained by taking an image of a space including an inspection object by image input means and binarizing the image with a master image stored in a master image memory. In the method, a mask area to be excluded from the inspection target among the binarized images is set based on the master image, and a difference image between the binarized image excluding the mask area and the master image is generated by the comparison determination unit. The comparison determination means obtains the number of blobs included in the difference image and the area of each blob, and determines that the inspection target does not match the master image if at least one of the number of blobs and the area is equal to or greater than a specified threshold. The mask area generated from the master image is excluded from the inspection target,
Since the match / mismatch between the inspection target and the template of the master image is determined using the difference image between the binarized image excluding the mask region and the master image, the fluctuation element of the blob due to the variation in the size of the inspection target is substantially reduced. There is an advantage that misrecognition that can be eliminated and that the inspection target that can be regarded as matching with the master image is determined as mismatching is greatly reduced.

According to a third aspect of the present invention, there is provided an image input device for capturing an image of a space including an object to be inspected, a master image memory storing a master image, and two images captured by the image input device.
A comparison / determination unit for generating a difference image between the binarized image and the master image, wherein the comparison / determination unit obtains the number of blobs included in the difference image and the area and flattening factor for each blob, Blobs whose rate is equal to or greater than the specified threshold are excluded from the inspection target, and if at least one of the number and area of the remaining blob is equal to or greater than the specified threshold, the inspection object is determined to be inconsistent with the master image. Obtain the flatness of each blob formed in the difference image, remove the blob whose flatness is equal to or more than the specified threshold from the inspection target, and then match the inspection target with the master image by the number of blobs and the area of each blob. -Since the mismatch is determined, it is possible to almost eliminate the blob variation element due to the variation in the size of the inspection object,
There is an advantage that erroneous recognition such as determining that the inspection target that can be regarded as matching the master image is not matched is greatly reduced.

According to a fourth aspect of the present invention, there is provided an image input means for picking up an image of a space including an object to be inspected, a master image memory storing a master image, and an image picked up by the image input means.
A mask area setting unit that sets a mask area to be excluded from the inspection target in the binarized image based on the master image, and a comparison that generates a difference image between the binary image excluding the mask area and the master image Determination means, and the comparison determination means obtains the number of blobs and the area of each blob included in the difference image, and when at least one of the number and the area of the blob is equal to or more than a specified threshold, the inspection object is determined to be a master image. A mask area generated from the master image is excluded from the inspection target, and a template of the inspection target and the master image is determined using a difference image between the binarized image excluding the mask area and the master image. Is determined, it is possible to remove almost all of the blob fluctuation factors due to the variation in the size of the inspection object, and the Then the test object discrepancies erroneous as to determine recognition considered an advantage that is greatly reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a diagram illustrating an example of a blob generated in the first embodiment of the present invention.

FIG. 3 is a diagram illustrating a conventional method.

[Explanation of symbols]

 Reference Signs List 1 camera 2 A / D converter 3 image memory 4 master image memory 5 comparison / determination means 6 external output means 7 D / A converter 8 display device

 ────────────────────────────────────────────────── ─── Continued on the front page F term (reference) 5B057 AA01 BA02 CA02 CA12 CA16 DA03 DB02 DB08 DC04 DC09 DC33 5L096 AA07 BA03 CA02 FA53 FA59 GA08 HA07

Claims (4)

[Claims] 1. A pattern matching method by image processing for matching a binarized image obtained by imaging a space including an inspection object by image input means and binarizing the image and a master image stored in a master image memory. The difference image between the structured image and the master image is generated by the comparison and determination means, and then the comparison and determination means obtains the number of blobs included in the difference image, the area and the flattening rate for each blob, The blob that has been described above is excluded from the inspection target, and the image processing is characterized by determining that the inspection target does not match the master image when at least one of the number and the area of the remaining blobs is equal to or greater than a specified threshold. Pattern matching method. 2. A pattern matching method by image processing for matching a binarized image obtained by imaging a space including an inspection target by image input means and binarizing the image with a master image stored in a master image memory. A mask area to be excluded from the inspection target in the binarized image is set based on the above, a difference image between the binarized image excluding the mask area and the master image is generated by the comparison determination means, and then the comparison determination means Determine the number of blobs included in the difference image and the area of each blob, and determine that the inspection target does not match the master image if at least one of the number of blobs and the area is equal to or greater than a specified threshold. Pattern matching method by image processing. 3. An image input unit for imaging a space including an inspection target, a master image memory storing a master image, and a binary image obtained by binarizing an image captured by the image input unit and a master image. A comparison determination unit that generates a difference image, wherein the comparison determination unit obtains the number of blobs included in the difference image, and the area and flatness of each blob,
Blobs whose flatness is equal to or more than a specified threshold are excluded from the inspection target, and when at least one of the number and area of the remaining blob is equal to or more than the specified threshold, it is determined that the inspection target does not match the master image. A pattern matching device using image processing as a feature. 4. An image input unit for capturing an image of a space including an inspection target, a master image memory storing a master image, and a binary image obtained by binarizing an image captured by the image input unit. A mask region setting unit configured to set a mask region to be excluded based on the master image; and a comparison determination unit configured to generate a difference image between the binarized image excluding the mask region and the master image. An image characterized by determining the number of blobs included in the image and the area of each blob, and determining that the inspection target does not match the master image when at least one of the number and the area of the blobs is equal to or greater than a specified threshold. Pattern matching device by processing.