JPH1027246A - Picture processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately align positions at the time of aligning an objective picture and a template picture. SOLUTION: Continuous areas d1 to d4 , D1 to D4 for an objective picture (s) and a template picture T are respectively found out and the positions of centers of gravity g1 to g4 , G1 to G4 in respective continuous areas d1 to d4 , D1 to D4 are found out. A regression straight line(RSL) is found out from the array of the centers of gravity g1 to g4 and an angle θ between the found RSL and an RSL obtained from the array of centers of gravity G1 to G4 . The RSL of the template picture T is rotated around the intersection (c) of both the RSLs only by the angle θ so as to be superposed to the RSL of the objective picture (s). Then the representative point (middle point) M of the picture T is moved in parallel so as to be superposed to the the representative point (middle point) m of the picture (s). The angle θ and the moving distances of the representative points become a relative position between the objective picture (s) and the template picture T.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method for obtaining specific information by applying an image processing technique to an inspection area set in an image.

[0002]

2. Description of the Related Art Conventionally, it has been considered that a target image including an inspection target is collated with a template image and a similarity is inspected to classify the inspection target or to judge the quality of the inspection target. I have. By the way, the target image is T
It is often obtained by imaging an object using an image input device such as a V-camera, and is usually accompanied by inclination (rotational displacement) and vertical and horizontal displacement (parallel displacement) with respect to the template image. On the other hand, the target image and the template image cannot be compared as they are. That is,
Processing for correcting the rotational displacement and the parallel displacement between the target image and the template image is required.

As this type of processing, for example, as shown in FIG. 9, a characteristic portion of a target image s (here, a corner of a frame)
It is conceived that the position of the target image s is matched with the template image by setting one or two specific matching regions DM including the following, and performing pattern matching on the matching region DM. By the way, when the target image s and the template image are collated as described above, the collation with the template image is often performed for the inspection area da set in a part of the target image s. Now, if the inspection area da includes the date of manufacture as an inspection target, and a template image is used to inspect a mistake in the date of manufacture or mixing of another date of manufacture, etc., It is necessary to create a template image for each manufacturing date to be inspected and manually input information on the matching area DM. Therefore, when the types of template images increase, the creation of template images becomes very troublesome. Further, the date of manufacture is not always stamped at the same position, and even if the position of the inspection target is shifted with respect to the matching area DM and the inspection target actually matches the contents of the template image. In some cases, the position of the template image is deviated from the position of the template image, and thus it is determined that they do not match.
In addition, the target image is limited to a shape having a characteristic portion in order to match the position with the template image, and a frame surrounding the character is required to make the manufacturing date and the like an inspection target. I do.

On the other hand, as in the method described in Japanese Patent Application Laid-Open No. 1220075/1990, the position of the approximate center of gravity of the inspection object is determined and the overall inclination is determined. There has been proposed a method of correcting the position of the inspection target based on the above. By employing this method, it is not necessary to include the information of the matching area DM in the template image, and the creation of the template image is facilitated.

[0005]

However, since the method described in the above publication determines the center of gravity of the entire inspection target in one process, the inspection area is used to remove noise components other than the inspection target. Needs to be set appropriately, and there is no way to remove noise components in the inspection area, and there is a problem that the reliability of the calculated center of gravity is low. Similarly, there is a problem that the inclination of the inspection target is easily affected by the noise component.

The present invention has been made in view of the above circumstances, and an object of the present invention is to make it possible to align a target image and a template image with each other without being affected by a noise component and to adjust the position with high accuracy. To provide an improved image processing method.

[0007]

According to a first aspect of the present invention, in order to achieve the above object, a target image input from an image input device is collated with a pre-registered template image to match the target image. In the image processing method for inspecting the similarity with the template image, the target image is an image including a plurality of continuous regions that can be treated as a group of pixels and the center of gravity of each continuous region is substantially aligned, The regression line obtained from the position of the center of gravity of each continuous region and the arrangement of these centers of gravity is obtained, and the regression line obtained from the center of gravity of each continuous region of the template image and the angle between the regression line obtained from the target image are obtained. While correcting the rotational displacement, based on the positional relationship between the center of gravity of each continuous region of the template image and the center of gravity of each continuous region obtained from the target image Correcting the line displacement, and sets the inspection area is set as a reference position of the center of gravity of the contiguous range in the template image in the target image.

According to this method, since the center of gravity of each continuous region is obtained for a target image including a plurality of continuous regions, the continuous region is found from within the target image without setting an inspection region in advance, whereby A desired test object can be found. In this case, even if a noise component is included, the noise component can be easily removed based on the size and number of continuous regions. In addition, since the straight line representing the direction of the target image is a regression line determined by the arrangement of the centers of gravity, the direction of the target image can be obtained with good reproducibility. As described above, since the direction of the target image can be obtained with good reproducibility by removing the noise component, the accuracy of the alignment with the template image is improved. Moreover, since the inspection area is set based on the position of the center of gravity of the continuous area in the template image, the inspection area in the target image can be automatically set.
Since the center of gravity, regression line, and inspection area of each continuous area of the template image can be obtained at the same time when the template image is set, it can be registered in advance as an attribute of the template image, and can be registered at the time of inputting the target image. There is no need to ask again. In addition, the size of the inspection area may be determined based on the center of gravity of the continuous area, and the size of the inspection area does not change even if the content of the inspection target changes as in the date of manufacture. With such a target image, an inspection area of the same size can be automatically set for all inspection targets.

According to a second aspect of the present invention, in the first aspect, the target image and the template image are binary images. This method is a desirable embodiment, and the process of obtaining a continuous area and a center of gravity for a binary image is simple.
High-speed processing becomes possible. According to a third aspect of the present invention, the outline of the inspection target is extracted from the binary image by a feature extraction process of generating a binary image by binarizing the target image input from the image input device and extracting a change point of the pixel value. After the extraction, a boundary line separated from the outline by a predetermined pixel width is set outside the inspection target, and the inside of the boundary line is set as an inspection region.

This method is a desirable method for setting the inspection area. In this method, the boundary of the inspection area is set along the outline of the inspection object, and the inside of the boundary is used as the inspection area. It is possible to perform an inspection that is hardly included and hardly includes a noise component. In addition, since the inspection area is automatically set according to the inspection target, the user does not need to manually set the inspection area.

According to a fourth aspect of the present invention, an outline of an inspection target is extracted from the target image by differentiating the target image input from the image input device and tracing an edge at which a change in the absolute value of the differential exceeds a predetermined value. After that, a boundary line separated from the outline by a predetermined pixel width is set outside the inspection object, and the inside of the boundary line is set as the inspection region. The method comprises:
Similarly to the invention of the above, the method for setting the inspection area is preferable, in which the boundary of the inspection area is set along the outline of the inspection object, and the inside of the boundary is the inspection area. An inspection that is not included and hardly includes a noise component becomes possible. In addition, since the inspection area is automatically set according to the inspection target, the user does not need to manually set the inspection area.

According to a fifth aspect of the present invention, there is provided an image processing method for checking a similarity between a target image and a template image by comparing the target image input from an image input device with a template image registered in advance. A density distribution in a predetermined inspection area set so as to surround the inspection object in the grayscale image including the inspection object, and an average value of the density distribution obtained for the template image in the inspection area having the same size as the inspection area The difference between the average value of the density distribution obtained from the target image and the target image is set as an offset value, and the pixel value of the difference image in which the density difference of each pixel between the target image and the template image is set as a pixel value is determined in advance. The range is shifted by the offset value, and when the pixel value of each pixel of the difference image deviates from the shifted determination range, the target image and the template image are shifted. Than is extracted as the matching portion.

This method is an effective method for inspecting a difference image between a target image and a template image in a state including density information, and obtains density values representing the target image and the template image, respectively. Since the difference between the two density values is used as an offset value to correct the determination range for the density determination,
Correction of the density difference between the target image and the template image enables accurate determination.

According to a sixth aspect of the present invention, in the fifth aspect of the present invention, the density distribution of the target image and the template image is generated based on the density obtained for each pixel from the inspection area of each image. This method is a desirable embodiment for obtaining the density distribution between the target image and the template image. Since the pixel density in the inspection area is obtained on a pixel-by-pixel basis,
The concentration distribution can be obtained with high accuracy.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION

(Embodiment 1) In this embodiment, a method for correcting a rotational displacement and a parallel displacement between a target image and a template image will be described. The target image and the template image are binary images and have a plurality of continuous regions in which pixels having the same pixel value can be treated as a group of adjacent pixels.
In addition, it is assumed that the positions of the centers of gravity of the respective continuous regions are substantially aligned. The inspection target s in the target image is actually a character string or the like as shown in FIG. 2, and if the pixels in each character have the same pixel value, each character becomes a continuous area d. A straight line representing the arrangement of the centers of gravity g of the continuous regions d employs a regression line j obtained based on the arrangement of the centers of gravity g.

In the present embodiment, as shown in FIG. 3, basically, a target image obtained by capturing an object by a TV camera as the image input device 1 is compared with a template image. The image captured by
The binarization unit 2 binarizes the density of each pixel with an appropriate threshold and stores the binarized image in the frame memory 3. Here, the continuous area extraction unit 4 extracts an aggregate of one pixel value from the binary image stored in the frame memory 3 as a continuous area. In other words, a pixel having a pixel value different from that of the background is detected, and if there is a pixel having the same pixel value in surrounding pixels (near 8), the pixel is regarded as belonging to the same continuous area, and a pixel is newly added. The same process is performed on the found pixel to determine a continuous area. A so-called labeling process is performed by giving the same attribute value (label) to pixels in one continuous region. In this way, a plurality of island-shaped continuous regions having different attribute values can be obtained in the binary image. Here, noise components other than the inspection target are removed by inspecting the size and the number of the continuous regions.

Next, in the correction unit 5, the template storage unit 6
Is corrected so that the template image stored in the frame memory 3 is superimposed on the target image stored in the frame memory 3. The specific procedure of this processing will be described later. The template image whose position has been corrected is input to the matching unit 7 together with the target image, and a difference image between the two is obtained, and the similarity is inspected. The similarity is checked as follows. That is, the number of pixels having different pixel values between the target image and the template image in the predetermined inspection area is obtained, and the obtained number of pixels is used as the evaluation value of the similarity. In this case, the smaller the number of pixels having different pixel values, the higher the similarity.

Hereinafter, a processing procedure of the correction unit 5 will be described. Now, assume that the target image s has four continuous areas d ₁ , d ₂ , d ₃ , and d _{4 as} shown in FIG. Here, it enters alongside the centroid g ₁ to g ₄ are substantially on a straight line in each successive region d ₁ to d _4, does not necessarily have to be completely identical. The template image T also has four continuous areas D _{1 to} D ₄ like the target image s.
₁ to D centroid G ₁ ~G ₄ of ₄ almost aligned on. Here, in order to superimpose the target image s and the template image T, it is necessary to obtain a relative position between the target image s and the template image T. Since the target image s can be regarded as being obtained by rotating and translating the template image T in a two-dimensional plane, the inclination of the target image s with respect to the template image T is obtained, and then the target image s and the template image T And the distance to it.

The template image T is created from a gray-scale image obtained by capturing a standard object in advance using the image input device 1 and stored in the template storage unit 6. as an attribute value of the template image T, the continuous region D ₁ and the position of the center of gravity G ₁ ~G ₄ of to D _4, continuous region D ₁ to D determined by the arrangement of the center of gravity G ₁ ~G ₄ of ₄ regression line (y = a · x + B) and, from the center of gravity of the two ends contiguous region D _1, D ₄ foot V _1, V ₂ of the perpendicular dropped to the regression line in the extending direction of the regression line, in the both perpendicular foot V _1, V ₂ The point M (Mx, My) is stored in the template storage unit 6. Further, the inspection area DA is stored in the template storage unit 6 as an area surrounded by two straight lines Mx ± X ₀ and two straight lines My ± Y ₀ .
Here, A, B, X ₀ and Y ₀ are constants. Here, the perpendicular feet V ₁ and V ₂ do not necessarily need to be stored in the template storage unit 6 as attribute values of the template image T.

Next, a template image is created for the target image s.
When obtaining a difference image with the image T, first,
Connection area d₁~ D_FourAnd the continuous area d₁~
d_FourCenter of gravity g₁~ G_FourRespectively. Also, continuous territory
Area d₁~ D_FourCenter of gravity g₁~ G _FourRegression determined by sequence
In the longitudinal direction of the straight line (y = a x + b) and the regression line
Continuous area d at both ends₁, D_FourFrom the center of gravity to the regression line
Vertical line v₁, V _TwoAnd the perpendicular foot v₁, V_TwoMidpoint of
m (mx, my). Where a and b are constants
is there.

Since the regression line obtained as described above is determined under the same conditions for the target image s and the template image T, each regression line is a straight line representing the template T and the target image s, respectively. Will be shown. Since the constants a and A represent the inclination of the regression line, the relative inclination angle θ between the target image s and the template image can be obtained from the following equation based on these constants a and A. θ = tan ^{−1 a−} tan ⁻¹ A Further, if both regression lines are regarded as simultaneous equations, the intersection c (cx, cy) of the two regression lines can be obtained by obtaining x and y.
Can be requested.

When the angle θ and the intersection point c are determined as described above, the middle point M (Mx, My) can be rotated by the angle θ so that the two regression lines overlap with the intersection point c as the center of rotation. That is, the midpoint M (Mx, My) obtained from the template image T can be rotated and moved by performing the following equation. Mx ′ = (Mx−cx) cosθ− (My−cy) sinθ + cx My ′ = (Mx−cx) sinθ + (My−cy) cosθ + cy The point M ′ (Mx ′, My ′) thus obtained is a regression line. Since it is a point above y = a × x + b, the deviation (dx, dy) between the point M ′ and the middle point m can be obtained as follows. dx = mx−Mx ′ dy = my−My ′ where dx and dy are distances in the x and y directions, respectively.

Using the angle θ of the regression line and the deviation (dx, dy) between the point M ′ obtained from the template image T and the middle point m obtained from the target image s in the above procedure, each successive regions D ₁ to D ₄ can be superimposed on each successive region d ₁ to d ₄ of the target image s and also corresponds to the inspection region DA which is stored in the template storage section 6 as the attribute value of the template image T The inspection area da to be set can be set for the target image s. Thereafter, a difference image between the inspection area da and the inspection area DA is obtained, and the number of pixels having different pixel values is used as the similarity evaluation value.

(Embodiment 2) This embodiment is a method corresponding to claim 3, wherein the target image s is a character printed on some member as shown in FIG. 4 (a). In the present embodiment, as shown in FIG. 5, an image stored in the frame memory 3 after binarizing the grayscale image captured by the image input device 1 composed of a TV camera by the binarizing unit 2 using an appropriate threshold value. Used. In the target image s, which is a binary image stored in the frame memory 3, a change point of the pixel value is extracted by the feature extraction processing unit 8a. This process is, for example, a process of raster-scanning a rough inspection area set in an image and extracting coordinates of a pixel value change point. In this manner, the outline u of the target image s can be extracted as shown in FIG. However, depending on the lighting conditions and the like at the time of imaging by the image input device 1, the outline of the actual object may exist outside the outline u of the target image s in the image. In step 8a, the outer shape line u obtained as described above is subjected to expansion processing. The expansion process is to set a boundary w that is separated from the outline u by a predetermined pixel width outside the target image s, and it is possible to set the boundary w as shown in FIG. it can.

With the above-described procedure, a relatively complicated
Even if the target image s has a shape, the boundary w along the outline u
Can be set, and the inside of the boundary line w
Extract specific information from the image of the target image s as the inspection area
That is largely unaffected by background information.
Information about the target image s without extracting
You can. In particular, as in the first embodiment, the target image
continuous area d of s₁~ D _FourProcessing to find the center of gravity of the
This processing is performed to obtain a difference image from the rate image T.
By applying morphological processing, the effects of the background can be removed and refined.
A high degree of inspection becomes possible.

(Embodiment 3) This embodiment is basically the same as the embodiment 2, except that a binary image is not used when extracting the outline u of the target image s. Is different. That is, as shown in FIG. 6, a grayscale image obtained by capturing an image with the image input device 1 such as a TV camera is stored in the frame memory 3, and the differentiation unit 8b performs spatial differentiation on the density. Although the method of spatial differentiation is well known, relatively simple and high-speed processing can be expected by using a Sobel operator. The differential processing unit 8b obtains a differential absolute value (that is, a rate of change in density) by performing spatial differentiation within an inspection region roughly set so as to surround the target image s. By mapping the differential absolute value obtained in this way to a three-dimensional space, a three-dimensional figure having peaks and valleys can be obtained. Therefore, edge extraction processing for extracting a ridge portion of a mountain (that is, tracking the maximum value of the differential absolute value) 4), the processing shown in FIG.
As shown in (b), the outline u of the target image s can be obtained.

The processing after obtaining the outline u of the target image s in this manner is the same as in the second embodiment.
In FIG. 4, a dilation process is performed on the outline u, and a boundary w that is separated from the outline u by a predetermined width is set outside the target image s as shown in FIG. That is. (Embodiment 4) This embodiment is characterized by a process of determining the similarity from the difference image after matching the inspection areas ad and AD between the target image s and the template image T as in Embodiment 1. Have However, the target image s and the template image T have density values as pixel values, and the pixel value of each pixel of the difference image is a density difference. Therefore, if the target image s and the template image T completely match with the inspection areas ad and AD matched, all the pixel values of the pixels included in the difference image obtained from the grayscale images of both are all Becomes 0. In short, the difference image means each inspection area a
Since the difference in density between pixels having the same relative position in d and AD is used as the pixel value of each pixel, if there is no density difference, the pixel value of the pixel of the difference image is 0. However, in practice, the lighting conditions when capturing an image with the image input device 1 and the state of the target to be imaged (for example, when printed characters are used as the target, dark or light ink may occur). ), The densities of the target image s and the template image T do not match.

Therefore, a judgment range for each pixel value of each pixel of the difference image is set, and when the pixel value deviates from the judgment range, it is regarded as inconsistent, so that a slight change in the condition can be ignored. Is the current situation. However, if such a determination range is set wide, there may be a case where the target image s and the template image T are erroneously determined to be mismatched even though they substantially match.

In order to avoid this kind of misjudgment, in the present embodiment, the density distribution of each pixel in the inspection area AD of the template image T is determined, and the maximum value (peak value) obtained in the density distribution is obtained. Is obtained, and the density difference is corrected using this average value. That is, if the inspection target has a substantially constant density such as a character string, the maximum values p ₁ and p _{2 of the} frequencies should appear in the density of the inspection target and the density of the background, respectively, as shown in FIG. is there. In FIG. 7, p
₁ the maximum value of the concentration of the test object, p ₂ is the maximum value of the density of the background. Thus, by calculating the average value of the _two local maximum values p ₁ and p ₂ , it is possible to know the overall brightness level in the inspection areas ad and AD of the grayscale image. Examples of this type of inspection target include characters and marks printed on various materials such as paper, synthetic resin, semiconductor, and ceramics, but of course, other inspection targets may be used.

When the similarity is determined by actually obtaining the difference image, the inspection area A set for the template image T is used.
The average value obtained from the density distribution in D is stored in advance in the template storage unit 6 as the attribute value of the template image T, and the inspection area ad set for the target image s
And the template image T obtained from the density distribution in
(A dashed line in FIG. 8 indicates the density distribution of the template image T, and a solid line indicates the density distribution of the target image s). This difference becomes the offset value. Now, it is assumed that the image of the target image s is darker than the template image T (the relationship is opposite to that of FIG. 8). Assuming that the higher the density value, the higher the brightness, the offset value under the above condition becomes negative (offset value = average value of the target image s−average value of the template image T).

By the way, an appropriate judgment range is set for the pixel value of each pixel of the difference image as described above, and when the pixel value is within the judgment range, the target image s corresponding to that pixel and the template The part with the image T is regarded as almost coincident. However, this determination range is set as a reference when the average values obtained from the target image s and the template image T match each other. Therefore, when the offset value is obtained as described above, the determination range of the pixel value of the difference image is shifted by the offset value. That is, if the offset value is negative as described above,
The judgment range is shifted to the negative side. If the pixel values of the difference image are determined using the determination range shifted in this way, it is possible to remove a difference such as an illumination condition at the time of input between the template image T and the target image s. Assuming that the image s and the image s are input under the same condition, the difference between the two can be extracted. Also, if there is a local difference between the target image s and the template image T, it should deviate from the above-described determination range after the shift, and that part is extracted as a part where the target image s and the template image T do not match. can do. In this way, the similarity is determined based on the area of the part extracted as the mismatched part, and the larger the area, the lower the similarity. Since the determination range is shifted by the offset value as described above, the margin of the determination range can be reduced as compared with the related art, and as a result, the similarity based on the difference image is set by narrowing the determination range. It can be obtained with high accuracy.

[0032]

According to the first aspect of the present invention, there is provided an image processing method for checking the similarity between a target image and a template image by comparing the target image input from the image input device with a template image registered in advance. , The target image is an image that includes a plurality of continuous regions that can be treated as a group of pixels, and the center of gravity of each continuous region is substantially aligned, and the position of the center of gravity of each continuous region and the arrangement of these centers of gravity The rotational displacement is corrected based on the angle between the regression line obtained from the center of gravity of each continuous region of the template image and the regression line obtained from the target image. The parallel displacement is corrected based on the positional relationship between the position of the center of gravity of the region and the center of gravity of each continuous region obtained from the target image. Inspection areas set based on the center of gravity of the continuous area are set as the target image.The center of gravity of each continuous area is determined for the target image including a plurality of continuous areas, so that the target area can be set without setting the inspection area in advance. Finding a continuous area from within an image has the advantage that a desired inspection target within the target image can be found. Further, since the straight line representing the direction of the target image is a regression line determined by the arrangement of the centers of gravity, there is an advantage that the direction of the target image can be obtained with good reproducibility.
As a result, the accuracy of alignment with the template image is increased. Moreover, since the inspection area is set based on the position of the center of gravity of the continuous area in the template image, there is an advantage that the inspection area in the target image can be automatically set.

If the target image and the template image are binary images as in the second aspect of the present invention, the processing for obtaining the continuous area and the center of gravity of the binary image is simple, so that high-speed processing is possible. There are advantages. According to a third aspect of the present invention, the outline of the inspection target is extracted from the binary image by a feature extraction process of generating a binary image by binarizing the target image input from the image input device and extracting a change point of the pixel value. After the extraction, a boundary line separated by a predetermined pixel width from the outline is set outside the inspection object, and the inside of the boundary line is set as an inspection region. After extracting the outline of the inspection target from the target image by differentiating the target image and tracing an edge where the change in the differential absolute value is equal to or more than a predetermined value, the predetermined pixel width from the outline is outside the inspection target. Since the separated boundary line is set and the inside of the boundary line is set as the inspection region, the invention of any claim sets the boundary line of the inspection region along the outline of the inspection target. Almost no background, no noise There is an advantage that the amount does not include most of the inspection is possible. In addition, since the inspection area is automatically set according to the inspection target, there is an advantage that the user does not need to manually set the inspection area.

According to a fifth aspect of the present invention, there is provided an image processing method for checking a similarity between a target image and a template image by comparing the target image input from an image input device with a template image registered in advance. A density distribution in a predetermined inspection area set so as to surround the inspection object in the grayscale image including the inspection object, and an average value of the density distribution obtained for the template image in the inspection area having the same size as the inspection area And a difference between the average value of the density distribution obtained from the target image and the average value of the density distribution as an offset value. The range is shifted by the offset value, and when the pixel value of each pixel of the difference image deviates from the shifted determination range, the target image and the template image are shifted. And of being extracted as the matching portion,
This is an effective method for inspecting the difference image between the target image and the template image while including the density information, in which the density values representing the target image and the template image are obtained, and the difference between the two density values is offset. Since the determination range for the density determination is corrected by using the value, there is an advantage that the density difference between the target image and the template image is corrected and the determination can be performed with high accuracy.

If the density distribution of the target image and the template image is generated based on the density obtained for each pixel from the inspection area of each image, the density distribution can be obtained with high accuracy. it can.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating the concept of a first embodiment.

FIG. 2 is a diagram illustrating the concept of the first embodiment.

FIG. 3 is a block diagram of an apparatus corresponding to the first embodiment.

FIG. 4 is a diagram for explaining the concept of Embodiments 2 and 3;

FIG. 5 is a block diagram of an apparatus corresponding to a second embodiment.

FIG. 6 is a block diagram of an apparatus corresponding to a third embodiment.

FIG. 7 is a diagram illustrating the concept of a frequency distribution according to a fourth embodiment.

FIG. 8 is a diagram illustrating the concept of a frequency distribution according to the fourth embodiment.

FIG. 9 is an explanatory diagram of a conventional method.

[Explanation of symbols]

1 The image input device 2 binarization unit 4 contiguous region extraction unit 5 corrector 6 template storage unit 8a feature extraction processing unit 8b differential processing unit c intersection d ₁ to d ₄ contiguous regions D ₁ to D ₄ contiguous region da inspection area DA inspection area g ₁ to g ₄ centroid G ₁ ~G ₄ centroid j regression line p ₁ maximum value p ₂ maxima s target image T template image u outline w border

Claims (6)

[Claims] 1. An image processing method for checking the similarity between a target image and a template image by comparing a target image input from an image input device with a pre-registered template image. It is an image that includes a plurality of continuous regions that can be treated as a pixel group, and the center of gravity of each continuous region is substantially aligned, and the regression line obtained from the position of the center of gravity of each continuous region and the arrangement of these centers of gravity is The rotational displacement is corrected based on the angle between the regression line obtained from the center of gravity of each continuous region of the template image and the regression line obtained from the target image, and the position of the center of gravity of each continuous region of the template image and the target image are corrected. The parallel displacement is corrected based on the positional relationship with the center of gravity of each continuous region obtained from An image processing method comprising: setting an inspection region set on the basis of a target image as a target image. 2. The image processing method according to claim 1, wherein the target image and the template image are binary images. 3. A binary image is generated by binarizing a target image input from an image input device, and an outline of an inspection target is extracted from the binary image by a feature extraction process of extracting a change point of a pixel value. An image processing method, wherein a boundary line separated from the outline by a predetermined pixel width is set outside the inspection object, and the inside of the boundary line is set as an inspection region. 4. An object to be inspected after differentiating an object image inputted from the image input device and extracting an outline of the object to be inspected from the object image by tracing an edge at which a change in a differential absolute value exceeds a predetermined value. An image processing method, wherein a boundary line separated by a predetermined pixel width from the outline is set outside the outline, and the inside of the boundary line is set as an inspection area. 5. An image processing method for checking the similarity between a target image and a template image by comparing a target image input from an image input device with a pre-registered template image. A density distribution in a predetermined inspection area set so as to surround the inspection target in the image is obtained, and the density distribution is obtained from an average value of the density distribution obtained for the template image in the inspection area having the same size as the inspection area and the target image. The difference between the average value of the density distribution and the average value of the density distribution is set as the offset value, and the determination range preset for the pixel value of the difference image in which the density difference between each pixel of the target image and the template image is set as the pixel value is the offset value And when the pixel value of each pixel of the difference image deviates from the shifted determination range, it is extracted as a mismatched portion between the target image and the template image. An image processing method characterized by outputting. 6. The image processing method according to claim 5, wherein the density distribution between the target image and the template image is generated based on the density obtained for each pixel from the inspection area of each image.