KR102090568B1 - Image processing method and defect inspection method - Google Patents

Abstract

The present invention alleviates the burden of judgment, etc. that requires skill from an operator when selecting an inspection target area from a captured image of an inspected object, and also makes it easy to select an inspection target area by simply setting a threshold. The task is to make it possible.
As a means for solving such a problem, the image processing method divides a monochromatic original image having first and second regions that are line-symmetric with respect to a reference straight line into first and second regions by a reference straight line. , A difference pixel value that is a difference between the pixel values of the two original pixels for each pair of the two original pixels arranged at a line symmetry position with respect to the reference straight line in the first and second regions A difference calculating step for calculating is provided. The image processing method is also a difference image generation process in which a difference image is generated by arranging difference pixels having difference pixel values, and a pair of an original pixel at a first position in a first area and an original pixel at a second position in a second area And a difference image generation process in which a difference pixel having a difference pixel value calculated by using is arranged in first and second positions of the difference image to generate a difference image.

Description

Image processing method and defect inspection method {IMAGE PROCESSING METHOD AND DEFECT INSPECTION METHOD}

The present invention relates to an image processing method for selecting an inspection target area to be subjected to defect inspection on a surface of an object to be inspected using a captured image obtained by imaging the surface of the object to be inspected by an imaging device, and a defect inspection method using the same In particular, it relates to an image processing method and a defect inspection method using the same, which is capable of selecting a region to be inspected with high precision and easily, regardless of the type of defect, for an inspected object having line symmetry.

Defect inspection is widely performed to inspect the surface of an object to be inspected for the presence or absence of a defect by using the captured image obtained by imaging the surface of the object to be inspected by an imaging device. The applicant has already applied for a defect inspection method for detecting linear defects called cracks (Patent Document 1).

On the surface of the inspected object, in addition to the above-mentioned cracks, there are defects covering various aspects such as cracks, scratches, marks, and misses. When detecting these defects, an image processing algorithm (image processing method) that digitally processes the digitized captured image can be used.

As an imaging device for imaging an object to be inspected, a camera equipped with an imaging element such as a charged-coupled devices (CCD) or a complementary metal-oxide-semiconductor (CMOS) is used. When light emitted from the object to be inspected is input to these imaging elements at the time of imaging, the intensity of light is converted into the intensity of electrical signals, digitized, and recorded as a digital image.

(1) Digital image

Here, the digital image will be described. The minimum element constituting the image is called a pixel, and the digital image is composed of pixels arranged in two dimensions. Each pixel individually has a numerical value expressed by a binary method in which 0 and 1 are combined as color information, and the numerical value represents the intensity of light emitted from the inspected object or the color of the surface of the inspected object. The numerical value of each pixel is referred to as a pixel value, and images are divided into, for example, color images and gray scale images.

In a color image, the color of one pixel is determined by the ratio of the components of the three primary colors of R (red), G (green), and B (blue) as components constituting the pixel. Therefore, when expressing the pixel value of one pixel in a color image, 24 bits (= 8 bits x 3 colors) in which RGB elements are represented by 8 bits are often used.

For a color image, an image expressed in black and white shades is called a gray scale image. A gray scale image expresses a pixel value of one pixel in 8 bits, and does not include color information, but only brightness information. Dark pixels have low (small) pixel values, and bright pixels have high (large) pixel values. This number of contrast steps is called gradation. The gradation varies depending on the amount of information allocated to one pixel. Here, the unit of the amount of information is called the number of bits, and the larger the number of bits, the larger the gradation. Specifically, the number of gradations when the number of bits is n is 2 ⁿ . For example, since the gray scale image has 8 bits, the number of gradations is 2 ⁸ = 256. And since the number of gradations is 256, the minimum value of the pixel value in the gray scale image is 0 corresponding to jet black, and the maximum value is 255 corresponding to pure white.

In addition, in a color image, colors are often decomposed into three primary colors of R (red), G (green), and B (blue), and the brightness is often expressed in the same number of gradations for each color. This is equivalent to generating three gray scale images, that is, monochrome images from color images. When the number of gradations of the gray scale image 256 (8 bits) is applied to each color of RGB, as described above, the color image has 8 bits x 3 colors = 24 bits. The number of gradations in this case is 2 ²⁴ = 16777216, and it is possible to express all colors with a digital image. The color that expresses a color image in 24-bit feels extremely natural when viewed with the human eye. For this reason, a color image expressed in 24-bit is called a full color image.

The method for generating a monochromatic image from a color image is not limited to generating three monochromatic images by decomposing the color image into three primary colors of R (red), G (green), and B (blue), as described above. In addition, there is a method of generating a single monochrome image from a color image. As an example, there is a method using an NTSC signal known as a television broadcasting standard. As a component signal used in the step before obtaining the NTSC signal, there is a YIQ signal. The Y component of this YIQ signal is a luminance value. Now, it is considered to generate a luminance value Y as a new pixel value by multiplying and multiplying the pixel values E _R , E _G , and E _B of each of the signals of the color image, that is, the R signal, the G signal, and the B signal. At that time, it is known to apply a weighted average using NTSC coefficients to be closest to the luminance visible to the human eye. Specifically, Y = 0.299E _R + 0.687E _G + 0.114E _B is calculated to generate a luminance value Y.

On the other hand, when detecting a defect in the image as described above, an image processing algorithm that digitally processes the digitized captured image can be used. The use of this image processing algorithm is equivalent to arithmetic on the pixel values. And by studying the calculation method, it is possible to select an area to be subjected to defect inspection existing in the image based on the calculation result. Image processing algorithms that have studied computation methods to have such an elective action are widely used as prior art.

(2) Prior art image processing algorithm

The prior art image processing algorithm will be described with reference to Figs. In addition, in the following description, the object to be inspected for surface defect inspection is mainly described as a work. In addition, the area | region which becomes the object of defect inspection existing in the said image is described as an inspection object area. In addition, an area not subject to defect inspection, that is, an area other than the inspection target area, is described as an excluded area.

When conducting defect inspection of the work surface, a work is imaged using an imaging device to obtain a captured image, and an image processing algorithm is applied to the captured image. Here, the obtained captured image is referred to as a gray scale image. In the imaging method, when the workpiece is a three-dimensional shape such as a hexahedron, for example, the workpiece is placed on a horizontally installed base, and each face is photographed using an imaging device disposed at a position facing each face of the workpiece. Take an image. In addition, when the work is a flat shape made of wood or resin formed of paper or thin board, the work is placed on a horizontally installed base, and the upper surface of the workpiece is imaged using an imaging device disposed on the upper side of the base. .

In the following description, for simplicity, simple shapes such as circles, ellipses, rectangles, and squares are used for the shape of the work in the drawings, the markings on the work surface, and the shape of defects on the work surface. It is shown as a schematic diagram.

32 (a) is a good product image PG1 obtained by imaging one surface of a good work product (good work) WG1 using an imaging device, and the background B1 and the work WG1 are imaged. The work WG1 has a rectangular outer shape, and a circular mark MG1 is marked on the surface. The intersection of the diagonal lines of the work WG1 and the center position of the display MG1 are generally coincident.

33A is a defective image PD1 obtained by imaging one surface of a workpiece having a defect on its surface, that is, a defective workpiece (bad workpiece) WD1 using an imaging device. The difference between the good image PG1 and the bad image PD1 is that the defect D1 exists in the work WD1 of the bad image PD1, and the defect D2 exists in the display MD1.

Here, both the good image PG1 in FIG. 32 (a) and the bad image PD1 in FIG. 33 (a) are gray scale digital images. Then, the color of each area in the good image PG1 and the bad image PD1 is compared as follows. Both the good image PG1 and the bad image PD1, the background B1 is black, the works WG1 and WD1 are dark gray, and the displays MG1 and MD1 are white. Further, in the defective image PD1, the defect D1 of the work WD1 is light gray, and the defect D2 on the display MD1 is also light gray. However, even if it is equally light gray, the defect D2 is slightly darker than the defect D1.

These comparisons are the result of the operator visualizing the defective image PD1 in Fig. 33A, and the actual defective image PD1 is recorded in the imaging device as a digital image as described above. For the recorded digital image, if the relationship between the pixel values of each area is expressed inequality using the name of the area,

Background (B1) <Work (WD1) <Defect (D2) <Defect (D1) <Display (MD1) (1)

It is made. Here, each of the regions is composed of a plurality of pixels. Then, the pixel values of the plurality of pixels constituting the same area have different values individually. However, in the following description, for simplicity, it is assumed that a plurality of pixels constituting the same area have the same pixel value.

When the defective image PD1 shown in Fig. 33 (a) is observed, the defect D1 is brighter (white in color) than the normal part of the work WD1 when the size relationship between the pixel values shown in (1) is observed. It means to be visible. Likewise, it means that the defect D2 is darker (in black color) than the normal portion of the display MD1 on the work WD1. In addition, in the case of the good image PG1 in Fig. 32 (a), the relationship between the pixel values of the region is excluded from (1), and the pixel values associated with the two types of defects are excluded.

Background (B1) <Walk (WG1) <Display (MG1) (2)

It is made. As specific examples of (1) (2) above, the codes assigned to the respective areas in the good image PG1 in Fig. 32A and the defective image PD1 in Fig. 33A are shown. Pixel values are correlated and are shown in tabular form in Figs. 32B and 33B, respectively. Hereinafter, for the sake of simplicity, the object of explanation is limited to the defective image PD1 in Fig. 33A.

Next, an image processing algorithm used for defect inspection of the work surface using the defective image PD1 shown in Fig. 33A will be described. Although the background B1 and the work WD1 are imaged in the defective image PD1 in Fig. 33A, the work WD1 is a target for performing defect inspection. Here, the position of the work in the picked-up image varies variously each time picking up. As an example, the defective image PD2 in which the defective work WD1 and the other defective work WD2 are imaged is shown in FIG. 34 (a) as in FIG. 33 (a). In addition, similarly to that in Fig. 33, the code assigned to each area in the defective image PD2 shown in Fig. 34 (a) and the pixel value of each area are made to correspond to each other in the form of a table in Fig. 34 (b). Is showing.

The position of the work WD2 in the defective image PD2 in Fig. 34A is different from the position of the work WD1 in the defective image PD1 in Fig. 33A. For this reason, it is necessary to accurately extract the work for each captured image regardless of the position of the work. This process is called a work extraction process. As an image processing algorithm that can be used in the work extraction process, a template matching method (TM method) is known in which a specific image pattern is extracted from an image to be irradiated. This is an algorithm executed in the following procedure.

[Procedure 1] As the specific image pattern described above

A predetermined image pattern (template) is prepared.

[Procedure 2] Contrast (match) the image and template to be investigated

Search for the closest match.

[Procedure 3] Matched locations are extracted as the specific image pattern.

For example, when the defective image PD1 in FIG. 33 (a) is an image to be irradiated, the shape of the work WG1 imaged on the good image PG1 in FIG. 32 (a) may be used as a template. . Further, when the template is compared with the defective image PD1, even if the position of the work WD1 is variously changed among the defective images PD1, even when the WD1 is located approximately in the center as shown in FIG. 33 (a), for example. Alternatively, even when the work WD1 is located at the lower right, such as the defective image PD1a shown in FIG. 35 (a), the work WD1 having the shape as shown in FIG. 35 (b) can be extracted. .

Next, for the extracted work WD1, the inspection target area and the exclusion area are selected. This process is called an inspection area selection process. The image processing algorithm as a conventional technique used in the inspection area selection process selects the inspection target area and the exclusion area according to the following [Condition 1].

[Condition 1] The pixel value of the inspection target area is

Is it large or small compared to the pixel value of the excluded area adjacent to the area?

Here, it is necessary to note that there are two cases of the "excluded region adjacent to the region of interest" described in [Condition 1]. Specifically, there are cases where the excluded region is adjacent to the outside of the region to be inspected, and may be adjacent to the inside of the region to be examined as being contained as a subset thereof. In the latter case, when the inspection target area containing the excluded area is referred to as the first inspection target area, the nested excluded area is stratified as a second inspection target area independent of the first inspection target area. ). As a result, a region to be inspected is constructed while combining and including a plurality of regions to be inspected. The inspection area selection process including the stratification of the inspection target area will be described with reference to FIGS. 35 and 36.

First, for the defective image PD1a in Fig. 35A, a work extraction process using a template matching method is executed to extract the work WD1. It is clear that the extracted work WD1 has its position at the lower right of the defective image, and the outermost edge is rectangular. After the work WD1 is extracted in this way, an inspection area selection process is performed.

In the first step of the inspection area selection step, the outermost edge of the work WD1 is disposed by arranging a circumferential edge F1 having a dimension larger by α than the maximum value of the dimension specification of the outermost edge on the outside of the work WD1. Surrounds. The state after executing this first step is shown in Fig. 36 (a). Here, the workpiece WD1 has a deviation in the dimension of the outermost edge due to manufacturing errors. There are standards specified by the maximum and minimum values for the dimensions, and a work WD1 that satisfies the standards is judged to be a good product related to the dimensions. And after picking up the workpiece | work WD1 which is a good product regarding this dimension, the work extraction process is performed using the above-mentioned template matching method. For this reason, the dimension of the outermost edge of the work WD1 obtained as a result of the work extraction process may take all values from the maximum value to the minimum value of the dimensional standard. Therefore, by arranging the circumferential border F1 having a dimension larger than α by the maximum value of the dimensional standard, it becomes possible to arrange the circumferential border F1 further outside the outermost edge for all the works WD1.

In the second step of the inspection area selection step, the inspection target area is selected by focusing on the pixel values of the pixels constituting each area of FIG. 36A. The pixel values of the pixels constituting the area between the circumferential edge F1 and the outermost edge of the work WD1 (hereinafter, the area around the work) WDS1 arranged in the first step in FIG. 36A are backgrounds. It is clear that the pixel values of the pixels constituting (b) are the same. Here, the size of the pixel values of the work surrounding area WDS1 and the work WD1 is focused. 33 (b) and (1), the pixel value of the area around the work WDS1 is 10 equal to the pixel value of the background B1, and the pixel value of the work WD1 is 100. Therefore, within the circumferential frame F1 in Fig. 36A, for example, the pixel value 50 is set as the work selection threshold value TWD1. Then, by selecting an area above the work selection threshold value TWD1, the interior of the work WD1 having a pixel value of 100 can be selected as a target candidate area to be a candidate for the inspection target area.

Next, after the third step of the inspection area selection process, the inspection target area is selected from the candidate candidate areas. The reason why this sorting is necessary is explained below. When the pixel values of the normal area and the defect in the interior of the work WD1 in Fig. 36A are compared, the pixels of the work WD1 (normal area) as shown in Fig. 33B. The value is 100, and the pixel value of the defect D1 of the work is 200. From this, for example, when the pixel value 140 is set as the work defect threshold value TD1 in the area within the work WD1, and the area above the work defect threshold value TD1 is selected, the defect D1 is detected. It seems to be possible. However, the display M1 (normal area) on the work exists in the area within the work WD1, and the pixel value is 250. Therefore, when an area within the work WD1 is selected with a pixel value of 140 or more as the work defect threshold value TD1, the defect D1 having the pixel value 200 and the work area having the pixel value 250 as the normal area The display MD1 is also selected as a defect. It is clear that this is not correct as a result of the defect inspection. In order to prevent this, an area to be inspected may be an area excluding the area in the display MD1 on the work from the area in the work WD1 as the target candidate area. This is after the third step of the inspection area selection process.

Here, the region in the work WD1 selected in the second step of the inspection region selection step is referred to as a first target candidate region. The area in the display MD1 on the work to be excluded from the first target candidate area is referred to as a first exclusion area. Finally, an area excluding the first excluded area from the first candidate area is referred to as a first inspection target area. Specifically, in the third step of the inspection area selection process, only the display MD1 on the work is selected using the template matching method similar to the work extraction process.

Next, in the fourth step, as in the first step, as shown in Fig. 36 (b), the dimension is larger by β than the maximum value of the dimension specification of the outermost edge on the outside of the display MD1 on the work. A circumferential edge F2 is disposed and surrounds the outermost edge of the display MD1 on the work.

In the fifth step, as in the second step, the pixel values of the pixels constituting the area between the circumferential edge F2 and the outermost edge of the display MD1 on the work (hereinafter referred to as the display surrounding area) M1S1 are Pay attention to the same as the pixel values of the pixels constituting the work WD1. 36 (b), the pixel value of the display surrounding area M1S1 is 100 equal to the pixel value of the work WD1, and the pixel value of the display MD1 on the work is 250. Thus, within the circumferential frame F2 in Fig. 36B, for example, the pixel value 200 is set as the display selection threshold TM1. Then, by selecting an area equal to or greater than the display selection threshold TM1, the inside of the display M1 on the work having a pixel value of 250 can be selected as the first exclusion area to be excluded from the first target candidate area.

Next, as a sixth step, the first exclusion area is excluded from the first candidate area, and the first inspection target area is used. Next, the display on the selected work MD1 as the first exclusion area is focused. As shown in Fig. 36B, a defect D2 may be present in the display MD1 on this work. Therefore, it is appropriate to consider the display MD1 on the work as the second target candidate area for detecting the defect D2. That is, after the seventh step, the display MD1 on the work is selected as the second target candidate region, and then, if necessary, the second excluded region is selected from the second target candidate region, and the second target candidate region is selected from the second target candidate region. The area excluding the excluded area is selected as the second inspection target area. According to Fig. 36B, when the display MD1 on the work is used as the second target candidate area, there is no second excluded area in this area. Therefore, the second inspection target area becomes the entire area in the display MD1 on the work. The inspection target area obtained as a result of performing the above process on the defective image PD1a shown in Fig. 35A is the first inspection target area selected in the sixth step and the agent selected in the seventh step. 2 It becomes two areas of the inspection target area. Thus, the inspection area selection process is ended.

When the first inspection target area and the second inspection target area are selected in the inspection area selection process, the process proceeds to a defect threshold setting process that sets a defect threshold value capable of detecting defects in each inspection target area. In this defect threshold setting step, a specific pixel value is set as a defect threshold, where each inspection target region can be selected as a defect and a normal region rather than a defect. At that time, normal areas and defects in the inspection target area are selected according to the following [Condition 2].

[Condition 2] The pixel value of the defect in the inspection target area is:

Is it large or small compared to the pixel value of the normal part in the area?

First, in the first step of the defect threshold setting process, the first defect threshold is set as the defect threshold in the first inspection subject area. As described above, the first inspection target area is an area other than the display MD1 on the work in the work WD1. 33 (b), the pixel value of the first inspection subject area is 100, and the pixel value of the defect D1 of the work is 200. From this, only the defect D1 is selected by, for example, setting the pixel value 150 as the first defect threshold TD1 in the first inspection target area, and selecting a region equal to or greater than the first defect threshold TD1. It becomes possible to do. That is, the first defect threshold TD1 is set to 150.

Next, in the second step of the defect threshold setting step, the second defect threshold is set as the defect threshold in the second inspection target area. The second inspection target area is an area in the display MD1 on the work. 33 (b), the pixel value of the second inspection target area is 250, and the pixel value of the defect D2 in the display MD1 on the work is 180. From this, by setting the pixel value 210 as the second defect threshold TD2 in the second inspection target area, for example, by selecting the area below the second defect threshold TD2, the defect D2 is identified. Screening becomes possible. That is, the second defect threshold TD2 is set to 210. Thus, the defect threshold setting process is ended.

As described above, in the second step of the inspection area selection step, the pixel value 50 is set as the work selection threshold value TWD1. The objective is to select the inside of the work WD1 from the background B1 inside the circumferential border F1 in Fig. 36A, and to select it as the first target candidate area to be the target area for defect inspection. It is for. Further, in the fifth step, the pixel value 200 is set as the display selection threshold value TM1. The purpose is to select the inside of the display MD1 on the work from the inside of the circumferential frame F2 in Fig. 36B as the first exclusion area to be excluded from the first target candidate area.

Similarly, in the first step of the defect threshold value setting process, the pixel value 150 is set as the first defect threshold value TD1. The purpose is to sort defects in the first inspection target area. In addition, in the second step, the pixel value 210 is set as the second defect threshold TD2. The purpose is to sort defects in the second inspection target area.

These various threshold values are optimal values for defect inspection of the work surface in the defective image PD1 in Fig. 33 (a) and the defective image PD1a in Fig. 35 (a). The reason is that the pixel values in each region of these defective images are the values shown in Fig. 33B. Here, the pixel values of each region in the defective image other than the defective image PD1 in Fig. 33A, that is, an image in which another defective work is captured are considered. It is assumed that the pixel values of each region have a variation over a certain range for each work. As long as the defective image is visually observed, the relative brightness or darkness of each area is the same for the captured image of which the work is captured.

However, when considered as a digital image, the difference in pixel values becomes a problem. For example, it is assumed that when the captured images of two different workpieces are visualized, the visual results of the same area in both are all white. However, when these two captured images are recorded as digital images, the pixel value of one white may be 240 and the pixel value of the other white may be 220 in some cases. Similarly, even if the visual results of the same region in both are all dark gray, the pixel value of one dark gray in a digital image may be 100, and the pixel value of the other dark gray may be 80. have. As described above, the pixel values of each region have a certain degree of variation for each work. Therefore, as described above, by using the defective images PD1 in Fig. 33 (a) and using various threshold values set in the inspection area selection process or the defect threshold setting process, all the defective images are reliably used. There is no guarantee that the inspection target area can be selected and defects within the area can be selected. For this reason, it is necessary to perform a threshold value confirmation process to confirm that the various threshold values are appropriate. The procedure of the threshold value confirmation process will be described below.

First, in the first step, B defective images obtained by imaging B defective work pieces (B is a natural number) are prepared. For each of these defective images, the inspection target area is selected and defects are selected using various threshold values set based on the defective image PD1 in Fig. 33A. It is also confirmed that the same inspection target area can be selected for all defective images, and that different defects are clearly selected for each defective image. If an inspection target area or a defective image in which it is not possible to accurately select defects is found, in each region of the defective image PD1 in FIG. 33 (a) used for setting the defective image and the original threshold value The threshold value is corrected by comparing the pixel values of (Fig. 33 (b)). Then, the inspection target area and defects are selected again for the B defective images using the corrected threshold. Confirmation of this sorting and correction of the threshold value are repeated until it is possible to select the same inspection target area for all defective images, and also it is possible to reliably sort out defects.

When it is finished, as the second step, it is possible to select the same inspection target area for all good images when applying various threshold values corrected by checking by using B bad images, as well as Confirm that the defect is not screened. In this case, A good quality images (A is a natural number) are prepared, and the same is done for confirmation of the B bad images described above. If a good product image in which the inspection target area is not accurately screened or a good product in which defects are selected is found, the threshold value is corrected, similarly to the correction corresponding to the confirmation using the B defective images described above. That is, the threshold value is corrected by comparing the pixel value (Fig. 33 (b)) in each region of the defective image PD1 in Fig. 33 (a) used to set the original threshold value. . Then, it is possible to select the same inspection target area for all of the A good quality images, and check and correction are repeated until no defects are screened. Here, the specific values of B and A may be determined by using a method of statistics, taking into account the number of workpieces subject to defect inspection per day or the manufacturing variation of the workpieces during mass production, for example.

In this way, when it is confirmed that various threshold values are appropriate, the threshold value confirmation process is ended. Then, by using these appropriate threshold values, the process proceeds to an inspection execution process in which defect inspection is performed on the captured image of the work as an object to be inspected. In the inspection execution step, the inspection area selection step described above is performed on the captured image of the inspected object to select the inspection target area. Next, the presence or absence of a defect is inspected for the region to be inspected using a defect threshold. If there is no defect, it is judged as a good product, and if there is a defect, it is judged as defective.

Here, an example of the pixel value of the captured image different from the case of the defective image PD1 in Fig. 33A is explained using the defective image PD11 shown in Fig. 37A. In the defective image PD11 in Fig. 37A, the color of each area is visually compared, and it is as follows. The background B11 is white, the work WD11 is light gray, and the display MD11 is black. The defect D11 of the work WD11 is dark gray, and the defect D21 on the display MD11 is light gray. However, even if it is equally light gray, the defect D21 is slightly lighter than the work WD11. When this defective image PD11 is recorded as a digital image, the relationship between the pixel values of the respective regions is large and small.

Display (MD11) <Defect (D11) <Work (WD11) <Defect (D21) <Background (B11) (3)

It is made. When the defective image PD11 in Fig. 37 (a) is observed, the defect D11 is darker (black) than the normal part of the work WD11 when the defective value PD11 in Fig. 37A is observed. ) Means to be visible. Likewise, it means that the defect D21 is brighter (white) than the normal portion of the display MD11 on the work WD11. These pixel values are shown as a table in Fig. 37B.

The relationship between the pixel values of the various threshold values and the areas to be sorted and the threshold values when performing the inspection area selection process and the defect threshold setting process described above with respect to the defective image PD11 in FIG. 37A is as follows. It becomes like this.

First, in the second step of the inspection area selection process, a work selection threshold (TWD11) is set. The purpose is to sort the inside of the work WD11 from the background B11 and to select it as a first target candidate area that becomes the target area for defect inspection. Here, the pixel value of the work WD11 from 130 in FIG. 37B is 130, and the pixel value of the background B11 is 250. When comparing these pixel values, if the work selection threshold (TWD11) is set to 180, for example, and when an area having a pixel value equal to or less than the work selection threshold (TWD11) is selected, selecting the first target candidate area You can see that it is possible.

Next, in the fifth step, the display selection threshold TM11 is set. The purpose is to select the inside of the display MD11 on the work as the first exclusion area to be excluded from the first target candidate area. Here, the pixel value of the work WD11 is 130 from FIG. 37B, and the pixel value of the display MD11 is 40. When these pixel values are compared, the display selection threshold TM11 is set to, for example, 90, and when a region having a pixel value equal to or lower than the display selection threshold TM11 is selected, it is possible to select the first excluded region. Can be seen.

Similarly, the first defect threshold is set in the first step of the defect threshold setting process. The purpose is to sort defects in the first inspection target area. Here, the pixel value of the workpiece WD11 is 130 from Fig. 37B, and the pixel value of the defect D11 is 80. Comparing these pixel values, it is possible to select the defect D11 by setting the first defect threshold TD11 to, for example, 100, and selecting an area having a pixel value equal to or less than the first defect threshold. Able to know.

Also, in the second step, a second defect threshold is set. The purpose is to sort defects in the second inspection target area. Here, the pixel value of the display MD11 from FIG. 37B is 40, and the pixel value of the defect D21 is 170. When comparing these pixel values, it is understood that it is possible to select the defect D21 by setting the second defect threshold TD21 to, for example, 100, and selecting a region having a pixel value equal to or greater than the second defect threshold. You can.

As described above, the setting of various threshold values in the conventional image processing algorithm and the relationship between the pixel values of the areas to be selected and the threshold values are determined based on two pieces of information in the captured image. The first information is the pixel value of the area when the captured image is recorded as a digital image. The second information is a comparison result of brightness (a degree of whiteness) and a darkness (a degree of blackness) of each region, that is, a comparison result of brightness information between each region, which can be obtained by the operator visually viewing the captured image. Of the second information, particularly important is the comparison of the brightness information of the normal portion and the defect in the inspection subject area.

In the description so far, it is assumed that the captured images as defective images and good products are gray scale images. If the captured image is a color image, the color of the color image is decomposed into three primary colors of R (red), G (green), and B (blue), and a gray scale image is generated for each color. Then, one gray-scale image judged to be able to clearly identify the most defective one is selected by the operator seeing three types of gray-scale images generated from the defective images and comparing the brightness information between the respective regions. The image processing algorithm may be applied to the selected gray scale image.

(3) Problems of prior art image processing algorithms

The above-described conventional image processing algorithm has the following problems. That is, when the image processing algorithm is applied to the captured image, the number of visual work by the operator increases. As described above, when setting various threshold values in the inspection area selection process, the operator visually observes the captured image and compares the brightness information between each area. Then, based on the comparison result, it is determined whether or not a defect and an excluded region as a normal region to be identified exist in the region to be examined.

For example, in the description of the inspection area selection process using the defective image PD1 shown in Fig. 33 (a), the pixel value of the work WD1 is 100, as shown in Fig. 33 (b). The pixel value of the defect D1 on the (WD1) is 200. The pixel value of the display MD1 in the work WD1 is 250. For this reason, in order to accurately select the defect D1 on the work WD1, the first inspection target area is excluded by excluding the display MD1 as the first exclusion area from within the work candidate WD1, which is the first target candidate area. Screening. In these tasks, the role that the operator's vision achieves is very large, and skill is required of the operator. As an example, in the case of the defective image PD1 shown in Fig. 33 (a) used in the above description, there were one excluded area and two inspected areas. However, the number of excluded areas and inspection target areas is not limited to this.

It is also conceivable that the number of excluded areas and inspected areas increases further due to the number of parts constituting the work surface, the parts where defects occur, their arrangement, and the pixel values of each part and the defects corresponding to them. have. In addition, when the captured image is a color image as described above, the color of the color image is decomposed into three primary colors of R (red), G (green), and B (blue), and a gray scale image is generated for each color. Then, an operator is added to visually compare the three types of gray scale images. When the exclusion area and the inspection target area increase as described above, or when processing for a color image is required, the time required for visual inspection of the operator increases further, such as an increase in the number of thresholds for selecting the area. . In addition, the item of judgment of the worker increases with it, and the burden on the worker increases. Due to the increase in the time required for the operator's eyesight and the increase in the burden, the inspection speed and inspection accuracy are lowered.

Japanese Patent Application No. 2015-4538

The object of the present invention is to reduce the burden of judgment and the like that requires the skill of the operator by reducing the visual work of the operator when selecting the inspection target area from the picked-up image of the inspected object having linear symmetry, and also thresholding An image processing method that contributes to an increase in the inspection speed and makes it possible to easily select an area to be inspected by simply setting a value, and an image processing method that is not affected by differences in operator skill and defects using the image processing method It is to provide an inspection method.

An image processing method, which is an aspect of the present invention, divides a monochromatic original image having first and second regions that are line-symmetrical to a reference straight line into the first and second regions by the reference straight line, A difference pixel value that is a difference between the pixel values of the two original pixels is calculated for each pair of two original pixels arranged at a line symmetry position with respect to the reference straight line in the first and second regions A difference calculation process to be performed, and a difference image generation process in which a difference pixel having the difference pixel value is arranged to generate a difference image, wherein the original pixel at the first position in the first area and the circle at the second position in the second area And a difference image generating process for arranging difference pixels having the difference pixel values calculated using a pair of pixels at the first and second positions of the difference image to generate the difference images. It is characterized by.

In addition, in the image processing method, the reference straight line includes the first region in the upper half and the second region in the lower half containing the same number of original pixels in the original image. It is characterized in that the first straight line dividing by.

Further, in the image processing method, the reference straight line includes the first region of the left half and the second region of the right half of the original image containing the same number of original pixels. Characterized in that it is a second straight line divided by.

Further, in the image processing method, in the difference image generation process, a first difference image and a second difference image different from the first difference image are generated as the difference image from the same original image, and the image processing is performed. The method is characterized by further comprising an inspection area selection step of selecting an inspection target area in the original image using the first difference image and the second difference image.

A defect inspection method, which is one aspect of the present invention, is a defect inspection method for performing defect inspection on an inspected object by using the image processing method, and is an imaging obtained by imaging a plurality of inspected objects known to be good products. A first step of generating a plurality of quality difference images by using the image processing method from a plurality of first original images generated by using a single-color image generated from the image as a first original image, and a plurality of tests with known defects A second step of generating a plurality of defective difference images by using the image processing method from a plurality of second original images generated by using a single color image generated from the captured image obtained by imaging an object as a second original image. As a threshold setting mode, among the difference pixels of the plurality of defective difference images, a difference pixel value is predetermined from a difference pixel value that is different. It is possible to select the spaced apart pixels arranged at the same position in each defective difference image at the same time as above, and further, the difference pixels arranged at the same position as the spaced apart pixels among the difference pixels of the plurality of good difference images A threshold value setting mode for setting a threshold value of one inspection area capable of not being elected as a separation pixel, and a monochrome image generated from a captured image obtained by imaging a target object to be inspected as a third original image are generated. A difference image is generated from the third original image using the image processing method, and an inspection target area in the third original image is selected using the difference image and the inspection area threshold, and the inspection target area is It is characterized by having a test execution mode for performing a defect inspection for do.

In addition, in the defect inspection method, the captured image is a non-color image, and there is one monochromatic image generated from each captured image.

In the defect inspection method, the captured image is a color image, and there are two or more monochrome images generated from each captured image, and in the first step of the threshold setting mode, each monochrome image is the first. As the original image, the plurality of quality difference images are generated from the generated first original images, and in the second step of the threshold setting mode, each monochrome image is generated as the second original image. The plurality of defective difference images are generated from a plurality of second original images, and in the threshold setting mode, one of the monochrome images having the largest setting range of the inspection area threshold is selected, and the selected monochrome image is An inspection area threshold is set, and in the inspection execution mode, from the captured image obtained by imaging the inspected object as the defect inspection object Among the two or more monochromatic images generated from the data, a monochromatic image of the same kind as the monochromatic image selected in the threshold setting mode is selected as the third original image, and the differential image is generated from the generated third original image. And selecting the inspection target area using the difference image and the inspection area threshold, and performing defect inspection on the inspection target area.

In addition, the defect inspection method is characterized in that it further comprises an object extraction process for extracting the object to be inspected for defects using a template matching method among the third original images.

When the image processing method of the present invention is used when the image of the workpiece to be inspected (the object to be inspected for defects) is linearly symmetrical with respect to a specific straight line, there is almost no process in which an operator makes any judgment by visually viewing the image of the workpiece. , By setting a single threshold value using simple image processing calculation, it is possible to easily select an inspection target area in an image of a work to be inspected. For this reason, there is no case that skill is required for the worker as in the prior art, and the burden on the worker is reduced. In addition, the image processing method can be automated by software to easily carry out defect inspection on an image of a work to be inspected. For this reason, the inspection speed is significantly improved compared to the defect inspection according to the prior art, and at the same time, it is not affected by the difference in the skill level of the operator.

1 (a) (b) is an explanatory diagram of the image processing algorithm of the present invention.
2 is an explanatory diagram of the image processing algorithm of the present invention.
3 (a) (b) (c) (d) (e) is an explanatory diagram of the image processing algorithm of the present invention.
4 (a) (b) (c) is an explanatory diagram of the image processing algorithm of the present invention.
5 (a) (b) (c) (d) (e) are explanatory diagrams of the image processing algorithm of the present invention.
6 (a) (b) (c) is an explanatory diagram of the image processing algorithm of the present invention.
7 (a) (b) is an explanatory diagram of the image processing algorithm of the present invention.
8 (a) (b) (c) (d) (e) is an explanatory diagram of the image processing algorithm of the present invention.
9 (a) (b) (c) is an explanatory diagram of the image processing algorithm of the present invention.
10 (a) (b) (c) (d) (e) are explanatory diagrams of the image processing algorithm of the present invention.
11 (a) (b) (c) is an explanatory diagram of the image processing algorithm of the present invention.
12 is an explanatory diagram of the image processing algorithm of the present invention.
13A and 13B are explanatory diagrams of the image processing algorithm of the present invention.
14 (a) (b) is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
15 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
16 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
17 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
18 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
19 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
20 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
21 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
22 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
23 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
24 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
25 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
26 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
27 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
28 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
29 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
30 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
31 is an explanatory diagram of a defect inspection method using the image processing algorithm of the present invention.
32 (a) (b) is an explanatory diagram of a prior art image processing algorithm.
33 (a) (b) is an explanatory diagram of a prior art image processing algorithm.
34 (a) (b) is an explanatory diagram of a prior art image processing algorithm.
35 (a) (b) is an explanatory diagram of a prior art image processing algorithm.
36 (a) (b) is an explanatory diagram of a prior art image processing algorithm.
37 (a) (b) is an explanatory diagram of a prior art image processing algorithm.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

(1) The image processing algorithm of the present invention

The image processing algorithm which is the basis of the present invention will be described with reference to Figs. Fig. 1 (a) shows the defective work WD1 imaged on the defective image PD1 in Fig. 33 (a) as an original image, that is, a digital image to be subjected to image processing, and this original image It is an explanatory diagram for applying the image processing algorithm of the present invention. Here, the defective work WD1 of FIG. 1A is extracted from the defective image PD1 using a template matching method. Each region of the surface of the defective work WD1 has already been described with reference to FIG. 33A, and thus is omitted here. 1 (b), the pixel values of each region in the original image shown in FIG. 1 (a) are shown in a tabular format. These pixel values are the same as the pixel values of the corresponding area in Fig. 33B. That is, the original image in Fig. 1A is also a gray scale image as in Fig. 33A.

First, in (a) of FIG. 1, attention is paid to the areas of the work WD1 and the display MD1. Defects D1 and D2 are present in these regions, respectively. Then, when the areas except the defects D1 and D2, that is, the same area as the good work WG1 in the good product image PG1 in Fig. 32 (a), it can be seen that the area has line symmetry. The same area as the good product work WG1 is constituted by the defective work WD1 and the display MD1 marked on the surface. In addition, as described above, the defective work WD1 has a rectangular outer shape, and the display MD1 is circular. In addition, the intersection of the diagonal lines of the work WD1 and the center position of the display MD1 generally coincide. Therefore, it can be seen that the region formed of the defective work WD1 and the display MD1 is line symmetrical with respect to the two straight lines L1 and L2 shown in Fig. 1A. Specifically, the first straight line L1 as a reference straight line is symmetrical up and down, and the first straight line L1 divides the original image into the same number of pixels in the upper half and lower half. do. Likewise, the second straight line L2 as the reference straight line is symmetrical to the left and right, and the second straight line L2 divides the original image into the same number of pixels in the left half and the right half.

The image processing algorithm of the present invention can easily select an inspection target region to be subjected to defect inspection for an image having an area that is line-symmetrical with respect to one or more symmetric axes. When explaining the principle of the image processing algorithm of the present invention, for simplicity, it is assumed that 16 pixels arranged in four vertical and four horizontal square shapes as shown in FIG. In Fig. 2, each pixel has a square shape, in which a unique address indicating the position of each pixel is described. This address is given to each pixel by a two-dimensional display combining X in the horizontal direction and Y in the vertical direction in FIG. 2. A specific addressing method will be described below.

First, in order to determine the increasing direction of the address values in the X direction and the Y direction, the origin of X = 0 and Y = 0 is taken at the upper left. Then, the X and Y of the two-dimensional addresses (X, Y) are increased in the arrow X direction and the Y direction of Fig. 2, respectively. That is, the address of the pixel located in the upper left corner becomes (0,0), and the address of the pixel located in the right corner increases the value of the X address from (0,0) by 1 (1,0). do. Similarly, the address of the pixel located below the pixel in the upper left corner becomes (0,1) by increasing the value of the Y address from (0,0) by one. By changing the values of the X address and the Y address in this way, the address of the pixel located in the lower right corner farthest from the origin increases the value of the X address from (0,0) by 3, while increasing the value of the Y address. Increase by 3 to become (3,3). In the following description, the address of the pixel is expressed by this method.

3 to 6 are diagrams for explaining the principle of the image processing algorithm of the present invention. Fig. 3 (a) shows the quality work WG1 in the quality image PG1 of Fig. 32 (a) as 16 pixels arranged in four vertical and four horizontal square shapes as shown in Fig. 2. It was modeled as an original image. In each pixel, pixel values as corresponding to the table in FIG. 32B are described. In addition, although the shape of the good product work WG1 in FIG. 32 (a) is rectangular, in FIG. 3 (a), it is modeled as a square for simplification. In addition, the display MG1 in FIG. 32 (a) is circular, but in FIG. 3 (a), the addresses (1,1), (2,1), and (1) in the center of the pixel composed of all 16 pixels It is modeled as four pixels of (2, 2), (2, 2). In this case, 12 pixels surrounding the four pixels in the central portion are modeled on the good work WG1 in Fig. 32A.

Here, the pixel value of the good work WG1 in FIG. 32B is 100, and the pixel value of the display MG1 is 250. However, in Fig. 3A, several scattered values in the vicinity of 100 are assigned to each of the 12 pixels as pixel values of the good work WG1, and at the same time, around 250 as the pixel values of the display MG1. Several scattered values of are assigned to each of the four pixels. The variation in the pixel values is caused by, for example, noise in an actual image.

In Fig. 3 (a), the brightness (the degree of whiteness) and the darkness (the degree of blackness) of the pixel are visually and relatively expressed in correspondence with each pixel value of 16 pixels. . For example, the pixel of FIG. 3A corresponding to the display MG1 in FIG. 32A is the addresses (1, 1), (2, 1), ( 1,2) and (2,2) are four pixels. The pixel values of these pixels are around 250, and are white when viewed visually. Accordingly, the four pixels in the central portion are in the form of a white-haired square without a shape on the background. On the other hand, the pixel in Fig. 3A corresponding to the good work WG1 in Fig. 32A is 12 pixels surrounding the four pixels in the center as described above. The pixel values of these pixels are around 100 and are dark gray when viewed. Therefore, the 12 pixels have a square shape in which a shape composed of a number of very short horizontal line segments is arranged on a base portion other than a number representing a pixel value. 3 (a) showing these 16 pixels, it becomes possible to visually understand that the center portion is white and the circumference is dark gray as in FIG. 1 (a). Thereafter, in the description of the image processing algorithm of the present invention, the relative brightness or darkness of each pixel is visually expressed in the same manner for a model with 16 pixels other than Fig. 3A.

In addition, in the following description, for the purpose of simplification of a sentence, the pixel arranged at the address (a, b) is referred to as a pixel (a, b). Similarly, the pixel values of the pixels arranged at the addresses (a, b) are denoted by the pixel values of (a, b). Here, in FIG. 3 (a), the first straight line L1 and the second straight line L2 as symmetry axes are overlapped as in FIG. 3 (a), the image processing algorithm of the present invention will be described below in order.

As described above, the image processing algorithm of the present invention is applied to an image having an axis of symmetry as a reference straight line, that is, an area that becomes line-symmetric with respect to the first straight line L1 or the second straight line L2 in Fig. 3A. (One area is called the first area, and the other area is called the second area). Here, in the case where the first straight line L1 is the symmetry axis in Fig. 3A, the symmetry of the pixel values of each pixel is focused. For example, pixels having a pixel value of 100 located at (0,0) and pixels having a pixel value of 103 positioned at (0,3) are arranged at symmetrical positions. Similarly, for example, pixels with a pixel value of 255 located at (2, 1) and pixels with a pixel value of 253 located at (2, 2) are also arranged in a symmetrical position. The pixels constituting the original image are hereinafter referred to as original pixels.

In this case, if the set of 16 original pixels in Fig. 3A is regarded as a square folding species, folding the folding papers with the first straight line L1 as folding marks, thereby folding the original pixels at the symmetrical positions. They can overlap each other. This process is called vertical folding. The image processing algorithm of the present invention first becomes a difference with respect to the pixel values of the two original pixels overlapped by the vertical folding, i.e., the two linear pixels arranged at the line symmetry relative to the first straight line L1. The difference pixel value is calculated for all the original pixels. In other words, the difference pixel values are calculated for each of the entire pair by using two original pixels overlapped by vertical folding as one pair. This process is called a difference calculation process. After the difference calculation step, a difference pixel (first difference image) is generated by disposing a difference pixel having the calculated difference pixel value at the same position as the two original pixels used to calculate the difference pixel value. For example, when a difference pixel value is calculated using a pair of the original pixel at the first position in the first area and the original pixel at the second position in the second area, the difference pixel having the difference pixel value is a difference image Difference images are generated by arranging them in the first position and the second position. This process is called a difference image generation process.

The first difference image generated by executing the difference calculation process and the difference image generation process with respect to the original image in Fig. 3A is shown in Fig. 3B. For example, in the original image of Fig. 3A, when the difference calculation process is performed using the original pixel of (0,0) and the original pixel of (0,3), the difference pixel value obtained is 103- 100 = 3. Therefore, when the difference image generation process is executed using the difference pixel having the difference pixel value 3, the pixel values of (0,0) and (0,3) in the first difference image in FIG. 3B are 3 It becomes. Similarly, if the difference calculation process is performed using the original pixel of (2,1) and the original pixel of (2,2) in the original image, the difference pixel value obtained is 255-253 = 2. Therefore, when the difference image generation process is performed using the difference pixel having the difference pixel value 2, the pixel values of (2, 1) and (2, 2) in the first difference image in FIG. 3B are 2 Becomes

Here, in the first difference image shown in Fig. 3B, the pixel values of all 16 pixels are small values extremely close to 0, which is the minimum pixel value. As described above, these small pixel values are generated using the difference pixels of the two original pixels arranged at symmetric positions with the first straight line L1 as a symmetric axis in FIG. This is due to the fact that it is arranged in the same position as and generates FIG. 3 (b). Since these two original pixels are arranged at symmetrical positions, the pixel values of each are very close. Therefore, the difference pixel values of these two pixels are extremely close to zero. When 16 pixels in Fig. 3B having such pixel values are visually observed, they are black. Corresponding to this, all of the 16 pixels in FIG. 3B are squares in which a shape made up of a number of right-handed diagonal lines is arranged on a base portion other than a number indicating a pixel value.

Next, the image processing algorithm of the present invention pays attention to the second straight line L2, which is a symmetric axis of left and right symmetry in Fig. 3A. This second straight line L2 divides the original image into eight pixels equal to the left half and the right half. The eight original pixels located on the left side of the second straight line L2 and the eight original pixels located on the right side of the second straight line L2 are arranged in symmetrical positions with each other. For example, the original pixel of the pixel value 100 located at (0,0) and the original pixel of the pixel value 110 located at (3,0) are arranged in a symmetrical position. Similarly, for example, an original pixel with a pixel value of 254 located at (1, 1) and an original pixel with a pixel value of 255 located at (2, 1) are also arranged at symmetrical positions.

In this case, as in the case of line symmetry with respect to the first straight line L1 described above, if the set of 16 original pixels in Fig. 3A is considered to be a square folding species, the second straight line L2 is a folded mark. By folding the folding paper, the original pixels in the symmetrical position can overlap each other. This process is called folding left and right. The difference which calculates the difference pixel value which becomes a difference with respect to the pixel values of the two original pixels superimposed by the left and right folds, that is, the two straight pixels L2 arranged at a line symmetry position, with respect to the entire original pixels The calculation process is performed. After the difference calculation step, a difference image generation step of generating a difference image (second difference image) by arranging the difference pixels having the calculated difference pixel values at the same positions as the two original pixels used to calculate the difference pixel values Run

The second difference image generated by executing the difference calculation process and the difference image generation process with respect to the original image in Fig. 3A is shown in Fig. 3C. For example, in the original image of Fig. 3A, when the difference calculation process is performed using the original pixel of (0,0) and the original pixel of (3,0), the difference pixel value obtained is 110- 100 = 10. Therefore, when the difference image generation process is executed using the difference pixel having the difference pixel value 10, the pixel values of (0,0) and (3,0) in the second difference image in FIG. 3C are 10. It becomes. Similarly, if the difference calculation process is performed using the original pixel of (1, 1) and the original pixel of (2, 1) in the original image, the difference pixel value obtained is 255-254 = 1. Therefore, when the difference image generation process is executed using the difference pixel having the difference pixel value 1, the pixel values of (1,1) and (2,1) in the second difference image in FIG. 3C are 1 It becomes.

Here, in the second difference image shown in Fig. 3C, the pixel values of the 16 pixels are all small values extremely close to 0, which is the minimum pixel value. This small pixel value is generated in the same manner as described with reference to FIG. 3 (b), as shown in FIG. 3 (a). This is due to the fact that the difference pixels are arranged at the same positions as these two original pixels to generate Fig. 3 (c). Since these two original pixels are arranged at symmetrical positions, the pixel values of each are very close. Therefore, the difference pixel values of these two pixels are extremely close to zero. When 16 pixels in Fig. 3C having such pixel values are visually observed, they are black. Corresponding to this, in Fig. 3 (c), as in Fig. 3 (b), all 16 pixels are squares in which a portion of the background other than the number representing the pixel value is made up of a number of right-rounded diagonal lines. It is done.

Here, a description is supplemented with respect to a specific algorithm for executing up and down folding and left and right folding in the difference calculation process in software. Addresses are assigned to the 16 pixels in Fig. 3A as shown in Fig. 2, respectively.

Therefore, when performing folding up and down, the software which executes the difference calculation process after determining the address of one end and the other end of the first straight line L1 in FIG. 3 (a) based on FIG. Hereinafter, as the input to the difference calculation software). Accordingly, the difference calculation software recognizes the axis of symmetry and can determine that the fold is up or down by the addresses of the one end and the other end.

Similarly, when performing the folding of the left and right, the addresses of one end and the other end of the second straight line L2 in Fig. 3A are determined based on Fig. 2, and then these addresses are transmitted to the difference calculation software as input. . Accordingly, the difference calculation software recognizes the axis of symmetry and can determine that the fold is left and right by the addresses of the one end and the other end.

The address of one end of the first straight line L1 in Fig. 3A corresponds to a position sandwiched by two pixels (0, 1) and (0, 2). Therefore, the address becomes (0,1.5), which is equal to the median value of the address of the pixels of (0,0) and (0,3) in the direction indicated by arrow Y in FIG. 2.

Similarly, the address of the other end of the first straight line L1 in Fig. 3A corresponds to a position sandwiched between two pixels (3, 1) and (3, 2). Therefore, the address becomes (3,1.5), which is equal to the median value of the address of the pixels (3,0) and (3,3) in the direction indicated by arrow Y in FIG. 2.

If the two addresses (0,1.5) and (3,1.5) determined in this way are passed to the difference calculation software as the input, the difference calculation software has the address of Fig. 2 as a table, so refer to this table and refer to the symmetric axis It is possible to generate the first straight line L1 as. And since the axis of symmetry is horizontal, the difference calculation software judges to perform up and down folding, for example, the pixel values of (0,0) and (0,3) in (a) of Fig. 3 above. The difference is calculated for the pixel values, and similarly, the difference is calculated for the pixel values of (2, 1) and the pixel values of (2, 2).

Similarly for the second straight line L2 in Fig. 3A, (0,0) and (1.5,0), which are the median values of (3,0) as a set of addresses, are transmitted to the difference calculation software as input. At the same time, (0,3) and (1.5,3), which are the median values of (0,3) and (3,3), can be transferred to the difference calculation software as inputs. The difference calculation software having received these addresses can generate the second straight line L2 as a symmetric axis by referring to the table of the address in FIG. 2, similarly to the case of the first straight line L1. Then, from the vertical axis of symmetry, it is determined that the difference calculation software performs left and right folding, for example, the pixel values of (0,0) and (3,0) in (a) of FIG. 3 above. The difference is calculated with respect to the pixel value, and similarly, the difference is calculated with respect to the pixel value of (1,1) and the pixel value of (2,1).

When the first difference image shown in Fig. 3 (b) and the second difference image shown in Fig. 3 (c) are generated from the original image in Fig. 3 (a) by the above steps, each difference image is next generated. Note the pixel values of the 16 pixels that constitute. Specifically, among these 16 pixels, a pixel (hereinafter referred to as an inter-pixel) having a pixel value (hereinafter referred to as an inter-pixel value) that is significantly separated from the majority of the pixel values is selected. The procedure will be described below using Figs. 3D to 4C.

First, pixel values of 16 difference pixels shown in FIG. 3B are arranged in descending order. The results are shown immediately below FIG. 3 (b) as FIG. 3 (d). 3 (d) is composed of three stages arranged at the top and bottom. Numerical values are written in each column, and the meaning of the numerical values of the column is written in the left column. A downward arrow is described between the numerical values of each stage, thereby helping to understand the description of the process of generating the numerical value of the lower stage of the arrow from the numerical value of the upper stage of the arrow.

The heading at the top of FIG. 3D is the descending order of the difference pixel values, and is a result of listing the pixel values of the 16 differential pixels shown in FIG. 3B in the descending order. There are four types of pixel values: 7, 3, 2, and 1, with 7 being the maximum value at the left end, and small pixel values sequentially arranged therefrom. At the right end, 1, which is the minimum value, is placed.

Next, the difference between adjacent difference pixel values at the top is calculated and arranged in the middle. The title of the middle is the difference value of the adjacent difference pixel value, and the difference value is written. The numerical value at the left end is 4, and the value of 4 is the difference between 7 at the left end at the top and 3 adjacent to the right. To show this, the upper pixel values 7 and 3 used for the calculation of the difference value 4 and the difference value 4 of the middle are respectively associated with the downward arrows. The same may be said of other numerical values in interruption.

When the above process is completed, the maximum value in the numerical value of the interruption is selected next. This election is a process for selecting the above-described inter-pixel. In the case of Fig. 3 (d), the numerical value 4 at the left end in the middle is the maximum value. In addition, the two pixel values at the top that generated the difference value that becomes the maximum value are 7 at the left end and 3 adjacent to it. The interval of 7 and 3 is a section in which the difference value of the middle is the maximum, and when the median of 7 and 3 is set as a threshold value in this section, the above-described inter pixel can be selected. In fact, when looking at the four values written at the top of FIG. 3 (d), the remaining three values are 3, 2, and 1 with respect to the value 7 at the left end, and the difference between these three values is relatively small. On the other hand, 70,000 at the left end are largely separated, and are considered to be separated pixels.

In this way, after the section where the difference value of the middle is maximized is selected, the median value of the difference pixel value at both ends of the section is calculated, and is used as a threshold value for selecting the separated pixel. The result of this threshold calculation is shown at the bottom of Fig. 3D. The heading at the bottom is the threshold value in the maximum section of the difference value, and the threshold value of 5 (rounding up to the decimal point) is 5 as the median value of 7 and 3, which are the upper pixel values that calculated the maximum difference value 4 selected from the number of interruptions. It is described as a value.

When the threshold value is set through the above steps, a pixel value that is greater than this threshold value is selected from the upper pixel values. This elected pixel value is a separation pixel value, and a pixel having a separation pixel value is a separation pixel. It is clear from the above description that the pixel value to be selected is 7.

As a result, the pixel value "7" of the separated pixel is separated from the other pixel values "3, 2, 1" by a predetermined value or more. An example of the predetermined value is the maximum difference value 4. In this case, the pixel value "7" of the separated pixel is separated by 4 from the pixel value "3", separated by 5 from the pixel value "2", and separated by 6 from the pixel value "1", They are separated from other pixel values by a predetermined value or more (ie, 4 or more).

Similar to the process of Fig. 3 (d) for the first difference image shown in Fig. 3 (b), the second difference image shown in Fig. 3 (c) is also used in the process of Fig. 3 (e). By setting the threshold value, the inter-pixel value 10 is selected from the pixels in Fig. 3C. The relationship between FIG. 3 (c) and FIG. 3 (e) is the same as the relationship between FIG. 3 (b) and FIG. 3 (d). In addition, the meanings of the arrows at the top, middle, and bottom of FIG. 3 (e) and each stage are the same as in FIG. 3 (d). Therefore, detailed description of FIG. 3E is omitted.

3 (b) and 3 (c), which were selected through the above steps, to surround the pixel values with double borders in FIGS. 4 (a) and 4 (b), respectively. Display. After the separation pixels are selected in the two difference images, the common portion of the separation pixels in the two difference images is selected as the next step. Then, the selected common portion is an area where defects may be present in the original image, that is, an area to be inspected for defect inspection.

Here, in FIG. 4A and 4B, a common portion of the inter-pixel is not present. The reason is that the original image in Fig. 3 (a) is a model of a good work WG1 shown in Fig. 32 (a), that is, a work having no defects, and the first straight line L1 as two symmetric axes And the line symmetry with respect to the second straight line L2. Therefore, in the original image in Fig. 3 (a), there is no inspection target area for defect inspection.

In the original image of Fig. 3A, the original pixel arranged at the same position as the common portion of the spaced apart pixels shown by enclosing the pixel values in Figs. 4A and 4B with double borders. Fig. 4 (c) shows an image shown by enclosing the pixel values with double borders. As described above, the spaced pixels in the two difference images shown by the double frame in Figs. 4A and 4B do not have a common portion. For this reason, there is no pixel in which the pixel value is surrounded by a double frame in FIG. 4C, and the image is the same as in FIG. 3A.

From the above description using the model of Fig. 3 (a), the original image of the good work WG1 having a line-symmetrical region as shown in the good product image PG1 in Fig. 32 (a) of the present invention When the image processing algorithm is applied, it can be seen that the inspection target area to which defect inspection is to be performed is not selected.

(2) Application example of the image processing algorithm of the present invention

Next, the case where the image processing algorithm of the present invention is applied to an image of a defective work WD1 having a line-symmetrical region as shown in the defective image PD1 in Fig. 33A is described. 5 (a) is a model of the defective work WD1 in the defective image PD1 in FIG. 33 (a) as an original image using 16 pixels as in FIG. 3 (a) above. . In each pixel, pixel values as corresponding to the table in Fig. 33B are described.

Here, the difference between the points (a) of FIG. 5 and (a) of FIG. 3, that is, the expression of the defects in FIG. 5 (a) will be described. The difference between FIG. 5 (a) and FIG. 3 (a) lies in the pixel values of the pixels arranged at the addresses (0,0) and (2,2). The pixel value of the address (0,0) in Fig. 3A is 100, and this pixel value corresponds to the pixel value of the good work WG1 shown in Fig. 32B. On the other hand, the pixel value of the address (0,0) in Fig. 5A is 200, and this pixel value corresponds to the pixel value of the defect D1 on the defective work WD1 shown in Fig. 33B. Is responding. Similarly, the pixel value of the address (2, 2) in Fig. 3A is 100, and this pixel value corresponds to the pixel value of the display MG1 shown in Fig. 32B. On the other hand, the pixel value of the address (2,2) in Fig. 5A is 180, and this pixel value corresponds to the pixel value of the defect D2 on the display MD1 shown in Fig. 33B. Doing.

Also in FIG. 5 (a), as in FIG. 3 (a), four pixels of addresses (1,1), (2,1), (1,2), and (2,2) in the center Twelve pixels surrounding the defective work WD1. In addition, the pixel value of the defective work WD1 in Fig. 5A is the same as that in Fig. 3A, except for the above (0,0) indicating a defect. Similarly, the pixel value of the display MD1 is set to the same value as that in Fig. 3 (a) except for the above (2, 2) indicating a defect. In addition, the relationship between the pixel value of each pixel shown in FIG. 5 and the square ground shape showing the pixel is the same as in FIG. 3.

If the pixel values in Fig. 5A are compared with the pixel values in Fig. 3A as described above, (0,0) and (2,2) in Fig. 5A. The pixel values of are 200 and 180, which are different from those in Fig. 3A. When these pixel values are visually observed, they are all light gray as shown in Fig. 33B. In this visual, the pixels shown in light gray do not exist in Fig. 32 (b) showing the correspondence of each region of Fig. 32 (a), which is a good image, and as a result, Fig. 3 (a) is modeled. It does not exist in (a). Therefore, since it corresponds to this light gray, the pixels of (0,0) and (2,2) in Fig. 5A are formed of a number of dots on the background portion other than the number representing the pixel value. It becomes a square in which the shape made is arrange | positioned.

The first difference image generated by executing the above-described difference calculation step and the difference image generation step with respect to the original image in Fig. 5A is shown in Fig. 5B. 5C shows the second difference image generated by performing the same process. Here, Fig. 5 (b) and Fig. 5 (c) are compared with Figs. 3 (b) and 3 (c), respectively. First, when FIG. 5 (b) and FIG. 3 (b) are compared, it can be seen that there is a difference between the two. All of the pixel values of the 16 pixels in FIG. 3B are extremely close to 0 as described above. On the other hand, of the 16 pixel values in Fig. 5B, the pixel values of (0,0) and (0,3) are 97, and the pixel values of (2,2) and (2,1) are 65. These are pixel values that are significantly larger than 0, that is, pixel values separated from 0, while pixel values of other pixels are extremely close to zero.

The reason why the pixel values of these four pixels are separated from 0 will be described below. In the original image in Fig. 5A, the pixel value of (0,0) corresponding to the defect is 200. In addition, the pixel arranged at the symmetric position with the pixel of (0,0) with the first straight line L1 as the axis of symmetry is (0,3) that does not correspond to the defect, and the pixel value is 103. Differences between these two pixel values Pixel values are arranged at positions (0,0) and (0,3) of the original image, which is the first difference image in Fig. 5B. For this reason, due to the large difference between the pixel value of (0,0) corresponding to the defect and the pixel value of (0,3) not corresponding to the defect in the original image, (0) in FIG. The pixel values of, 0) and (0,3) become large, and are separated from zero. In addition, in the original image in Fig. 5A, the pixel value of (2, 2) corresponding to the defect is 180. And the pixel arranged in the symmetric position with the pixel of (2,2) with the first straight line L1 as the axis of symmetry is (2,1) which does not correspond to the defect, and the pixel value is 255. Differences between these two pixel values Pixel values are arranged at positions (2, 2) and (2, 1) of the original image, which is the first difference image in Fig. 5B. For this reason, due to the large difference between the pixel value of (2,2) corresponding to the defect and the pixel value of (2,1) not corresponding to the defect in the original image, (2) of FIG. The pixel values of, 2) and (2,1) become large, and are separated from zero.

Next, when comparing FIG. 5 (c) and FIG. 3 (c), it can be seen that there is a difference between the two. All of the pixel values of the 16 pixels in FIG. 3C are extremely close to 0 as described above. On the other hand, among the 16 pixel values in Fig. 5C, the pixel values of (0,0) and (3,0) are 90, and the pixel values of (1,2) and (2,2) are 71. These are pixel values that are significantly larger than 0, that is, pixel values separated from 0, while pixel values of other pixels are extremely close to zero.

The reason why the pixel values of these four pixels are separated from 0 will be described below. In the original image in Fig. 5A, the pixel value of (0,0) corresponding to the defect is 200. In addition, the pixel arranged at the symmetric position with the pixel of (0,0) with the second straight line L2 as the axis of symmetry is (3,0) that does not correspond to the defect, and the pixel value is 110. The difference between these two pixel values The pixel values are arranged at positions (0, 0) and (3, 0) of the original image, which is the second difference image in Fig. 5 (c). For this reason, due to the large difference between the pixel value of (0,0) corresponding to the defect and the pixel value of (3,0) not corresponding to the defect in the original image, (0 of FIG. 5C) The pixel values of, 0) and (3,0) become large, and are separated from zero. In addition, in the original image in Fig. 5A, the pixel value of (2, 2) corresponding to the defect is 180. And the pixel arranged in the symmetric position with the pixel of (2,2) with the second straight line L2 as the axis of symmetry is (1,2) that does not correspond to the defect, and the pixel value is 251. Differences between these two pixel values Pixel values are arranged at positions (2, 2) and (1, 2) of the original image, which is the second difference image in Fig. 5 (c). For this reason, due to the large difference between the pixel value of (2,2) corresponding to the defect and the pixel value of (1,2) not corresponding to the defect in the original image, (2) of FIG. The pixel values of, 2) and (1,2) become large, and are separated from zero.

Due to the above, when the difference calculation process and the difference image generation process are executed, the difference pixel value generated from the pixel value of the area where the defect may exist is the difference pixel value generated from the pixel value of the area where the defect does not exist. It can be seen that the separation is significantly larger than that. By using this, if an interpolated pixel having the inter-pixel value is selected in the difference images in FIGS. 5B and 5C, the original pixel generating the inter-pixel in the original image is defective. It can be judged that there is a possibility. In the process of selecting the inter-pixel values, the threshold value is set in a section in which the difference value between adjacent difference pixel values becomes the maximum in the description using FIGS. 3D and 3E above. Corresponds to the process.

5 (d) and 5 (c) of FIG. 5 (d) and FIG. It is shown in (e). The description method of FIG. 5 (d) and FIG. 5 (e) is the same as that of FIG. 3 (d) and FIG. 3 (e), respectively. Therefore, detailed description of FIGS. 5D and 5E is omitted. When the inter-pixel values are selected from the pixels in FIG. 5B using the threshold values described at the bottom of FIG. 5D, 65 and 97 are obtained. Similarly, if the inter-pixel values are selected from the pixels in Fig. 5C using the threshold values described at the bottom of Fig. 5E, they become 71 and 90.

5B and 5C, which were selected through the above steps, enclose the pixel values in double borders in FIGS. 6A and 6B, respectively. Is indicated by. After the separation pixels are selected in the two difference images in this way, a common portion of the separation pixels in the two difference images is selected as the next step. The common parts of the spaced-out pixels in FIGS. 6A and 6B are (0,0) and (2,2).

Fig. 6 (c) shows an image in which the original pixels arranged in the same position as the common portion in the original image in Fig. 5A are surrounded by double borders of pixel values. The pixel of this double frame is the same as the pixel corresponding to the defect in Fig. 5A. That is, as shown in the defective image PD1 in Fig. 33 (a), the image processing algorithm of the present invention is applied as compared to the original image of the defective work WD1 in which areas other than the defects D1 and D2 become line symmetry. In this case, it can be seen that the inspection target area to which defect inspection is to be performed can be easily selected. It can be seen that the number of processes is greatly reduced when this process is compared with the process of selecting the inspection target region in the above-mentioned conventional technology. In addition, it can be seen that the process of making a judgment by visually inspecting the original image is significantly reduced. This is an effect produced by using the image processing algorithm of the present invention when the original image has a region that is line symmetric with respect to the axis of symmetry.

In the above description, a model of a work having a symmetrical shape that is vertically symmetrical and vertically symmetrical, having a small square display around the center of the square, as shown in Fig. 3 (a), is exemplified as the original image. In this case, as described above, there are two symmetrical axes, and the difference image generated from the original image becomes two corresponding to the number of symmetrical axes. However, in the present invention, the number of symmetrical axes is not limited to two. When C is a natural number, if the number of symmetric axes is C, the number of difference images generated through the difference calculation process and the difference image generation process is C. In this case, the spaced-out pixels are selected for the C difference images, and the common portion thereof is used as the inspection target area for defect inspection in the original image.

(3) Work conveyance and image processing of the present invention: using the first straight line L1 and the second straight line L2

On the other hand, in order to obtain captured images of a large number of workpieces, it is necessary to continuously convey the workpieces by means of conveying to a position where the imaging device can image the workpieces. At this time, even if the work has the above-mentioned symmetry axis, depending on the outer shape of the work and the arrangement of parts on the surface, due to the conveying method of the work in the conveying means, in a plurality of captured images In some cases, the direction of the axis of symmetry of the work may not be determined as one. The example will be described with reference to FIG. 7. In this description, for simplicity, a good-quality work piece having one symmetric axis is taken as an example.

Fig. 7 (a) is a view showing the shape and dimensions of the original image generated by imaging the good work WG2 (hereinafter, the work WG2) and then extracting the work WG2 using the above template matching method. to be. The work WG2 is a square in which one side is a, and the circular mark MG2 is marked in contact with the side at the midpoint P0 of the side. There is a relationship of d < a / 2 between the length a / 2 from the edge of the work WG2 serving as one end of the side to the midpoint P0 and the diameter d of the display MG2. Overlapping this original image, the first straight line L1 and the second straight line L2 are described as in Fig. 1A. In Fig. 7 (a), the original image is clearly not line symmetry with respect to the first straight line L1, but line symmetry with respect to the second straight line L2. That is, the axis of symmetry is the second straight line L2.

Here, the conveying means for conveying the work WG2 to a position at which imaging by the imaging means becomes possible will be described using Fig. 7B. Fig. 7B shows a view of the work WG2 conveyed by the conveying means T1 from the upper side of the conveying means T1. The conveying means T1 is provided horizontally, and is equipped with elongated straight feeders F1 having edges E1 and E2 substantially parallel to each other on both sides for conveying while the work WG2 is mounted. . A mechanism may be added to the edges E1 and E2 to prevent the placed work WG2 from jumping out of the feeder F1. The work WG2 is mounted on the upper surface of the feeder F1 so that the surface shown in Fig. 7 (a) is on the upper side and is substantially parallel to the two opposite edges E1 and E2 of the surface. And the work WG2 is conveyed by a linear path by the feeder F1 moving to the direction of arrow X1 shown in FIG. 7B by the action of the drive mechanism not shown. The imaging means (not shown) is provided on the upper side of the conveying means T1, and when the work WG2 is transported to a position immediately below it, an image of the top surface of the work WG2 shown in Fig. 7A is obtained. It is possible to take images.

7 (b), the work WG2 placed on the conveying means T1 is shown in four directions W1 to W4. As described above, the work WG2 is square. Then, when the work WG2 is mounted, the conveying means T1 is such that the surface shown in Fig. 7 (a) becomes the upper side as described above, and the two opposite edge edges E1 and E2 facing the surface are approximately It is positioned to be parallel. That is, when the work WG2 is conveyed by the conveying means T1, the display MG2 has four positional relationships with respect to the conveying direction (arrow X1). In Fig. 7 (b), these four positional relations are the first work W1 to be conveyed, the second work W2 from the first, the third work W3, and the last work W4. Is represented as.

In Fig. 7B, the above-described first straight line L1 and second straight line L2, which can be symmetrical axes in the captured image, are superimposed on each work W1 to W4. As is evident from the first straight line L1 and the second straight line L2 in the work W1 to W4, the captured image of W1 and W3 is the second straight line L2 as the axis of symmetry, and the imaging of W2 and W4 In the image, the first straight line L1 is a symmetric axis. That is, depending on the direction in which the work WG2 is placed on the conveying means, the direction of the axis of symmetry of the original image generated by extracting the work from the captured image is not determined as one.

In such a case, it is very troublesome for the operator to visually view the original image individually and to determine at any time which one of the first straight line L1 and the second straight line L2 becomes the axis of symmetry. However, in the image processing algorithm of the present invention, even if it is necessary to select one of two symmetrical axes for each original image, as in the description using FIGS. 3 and 4 or FIGS. 5 and 6 described above, the two By generating differential images on both sides of the axis of symmetry, it is possible to accurately select an inspection target region. The process will be described with verification using FIGS. 8 to 11.

Fig. 8 (a) is an original image obtained by modeling the good work WG2 shown in Fig. 7 (a) by 16 pixels as in Fig. 3 (a). The display MG2 in Fig. 7A is modeled as two pixels (1,0) and (2,0) in Fig. 8A. Their pixel values are similar to those of the four pixels (1, 1), (2, 1), (1, 2), and (2, 2) arranged in the center in Fig. 3A. In addition, 14 pixels other than the two pixels (1, 0) and (2, 0) in FIG. 8 (a), a non-defective work (WG2) excluding the display (MG2) in FIG. 7 (a). Is modeled. Their pixel values are similar to those of the 12 pixels surrounding (1,1), (2,1), (1,2), and (2,2) arranged at the center in Fig. 3A. . 8 (a), the first straight line L1 and the second straight line L2 are overlapped as in FIG. 3 (a).

In addition, in the 16 squares shown in Figs. 8 (b), 8 (c) and 9 (a) to 9 (c) used in the following description, each pixel value The shape of the inside of the square other than the number denoting is the same as that of FIG. 3. In addition, the notation method of FIG. 8 (d) and FIG. 8 (e) is the same as that of FIG. 3 (d) and FIG. 3 (e). Therefore, detailed description of these is omitted.

The first difference image generated by executing the difference calculation process and the difference image generation process with respect to the original image in Fig. 8A is shown in Fig. 8B. Also, the second difference image generated in the same manner is shown in Fig. 8C. Here, the original image shown in Fig. 8A does not have symmetry with respect to the first straight line L1. Specifically, the pixels of (1,0) and (2,0) and the pixels of (1,3) and (2,3) arranged at symmetrical positions with respect to the first straight line L1 have significantly different pixel values. . This is clear from FIG. 8 (a) being an image modeling the good work WG2 of FIG. 7 (a). The pixels (1,0) and (2,0) in Fig. 8A correspond to the display MG2 in Fig. 7A. On the other hand, the pixels (1,3) and (2,3) in Fig. 8A correspond to the good work WG2 in Fig. 7A.

7 (a), it is clear that the first straight line L1 has no symmetry. For this reason, in the first difference image in Fig. 8 (b) generated by folding up and down on the original image, (1,0) and (1, which are arranged at positions where the original image does not have the symmetry) The pixel value of 3) is 145. Similarly, the pixel values of (2,0) and (2,3) in which the original image is disposed at a position not having the symmetry are 142. These pixel values are separated from each other by 12 pixel values. On the other hand, these pixel values around 150 first appeared in Fig. 8B. In order to visually cope with this, all of these four pixels have a square shape in which a base portion other than a number representing a pixel value is formed of a plurality of right and left diagonal lines.

On the other hand, the original image in Fig. 8 (a) has symmetry with respect to the second straight line L2. For this reason, in the second difference image in Fig. 8 (c) generated by folding the left and right sides of the original image, all the pixels are pixel values extremely close to zero. Next, the results obtained when the same process as in Figs. 3 (d) and 3 (e) is performed with respect to Figs. 8 (b) and 8 (c) are shown in Figs. 8 (d) and 8 (c). It is shown in FIG. 8 (e). When the inter-pixel values are selected from the pixels in FIG. 8B using the threshold values described at the bottom of FIG. 8D, 142 and 145 are obtained. Similarly, using the threshold value described at the bottom of FIG. 8 (e), if an inter-pixel value is selected from the pixels in FIG. 8 (c), it becomes 9 and 10.

8 (b) and 8 (c), which were selected through the above steps, to surround the pixel values with double borders in FIGS. 9 (a) and 9 (b), respectively. Display. In this way, after the separation pixels are selected in the two difference images, a common portion of the separation pixels in the two difference images is selected as the next step. There is no common portion of the separation pixels in Figs. 9A and 9B. The reason is the same as the reason that the common portion does not exist in the inter-pixel in FIGS. 4A and 4B described above. For this reason, when the image shown by enclosing the pixel value of the original pixel arranged in the same position as this common part in the original image in Fig. 8A is surrounded by a double frame, as shown in Fig. 9C. Therefore, a pixel value surrounded by a double border does not exist. As described above, the image processing algorithm of the present invention can be applied to the original image of a good-quality work product having symmetry with respect to only one of the two symmetrical axes as in the case of the original image having two symmetrical axes.

Next, the defective work in which a defect is added to the good work WG2 of Fig. 7 (a) is modeled as in Fig. 8 (a), and the same as in Figs. 8 (b) to 9 (c). To carry out the process. Fig. 10A is an original image of the defective work WD2 in which two defects are added to the original image shown in Fig. 8A. The pixels showing defects are (0, 1) and (2, 3).

Here, looking at the original image corresponding to the good work WG2 of Fig. 8 (a), that is, Fig. 7 (a), the following can be seen. In (a) of FIG. 8, (1,3) and (2,3) arranged in a symmetrical position with these two pixels with respect to the pixels (1,0) and (2,0) and the first straight line (L1) When the pixel of) is removed, the original image becomes a shape similar to the letter H of the alphabet. And from the 12 pixel values constituting the original H-shaped image, the original H-shaped pixel has symmetry with respect to either of the first straight line L1 and the second straight line L2. The region formed by the four removed pixels has symmetry only with respect to the second straight line L2.

In addition, the pixel of (0,1) added as a defective pixel in Fig. 10A is symmetrical with respect to any of the H-shaped original pixels, that is, the first straight line L1 and the second straight line L2. It belongs to the area having. In addition, in (a) of FIG. 10, pixels (2, 3) added as defective pixels belong to a region having symmetry only with respect to the removed original pixel, that is, the second straight line L2. 10 (a) is a model in which pixels showing defects are arranged in two places having different properties from the viewpoint of symmetry in the original image. In addition, all of the pixels of (0, 1) and (2, 3) are arranged at positions corresponding to defects on the work other than the display MG2 in the good work WG2 shown in Fig. 7A. have. In addition, in (a) of FIG. 10, the pixel value of (0, 1) is 22, and the pixel value on the work other than the display MG2 is considerably small as compared to 100, and when viewed, the defect looks black compared to the normal part. to be. In addition, in (a) of FIG. 10, the pixel value of (2, 3) is 170, and the pixel value on the work other than the display MG2 is considerably larger than that of 100, and visually, a defect that appears white compared to the normal portion to be.

The first difference image generated by executing the above-described difference calculation process and the difference image generation process with respect to the original image of the defective work shown in Fig. 10A is shown in Fig. 10B. 10 (c) shows a second difference image generated by performing the same process. 10 (d) and 10 (c), the same results as those of FIGS. 8 (d) and 8 (e) were performed. It is shown in (e). Using the threshold value described at the bottom of FIG. 10D, if an inter-pixel value is selected from the pixels in FIG. 10B, it becomes 73, 83, 142, and 145. Similarly, if the inter-pixel values are selected from the pixels in FIG. 10C using the threshold values described at the bottom of FIG. 10E, it becomes 64 and 87.

10B and 10C, which were selected through the above steps, to surround the pixel values in double borders in FIGS. 11A and 11B, respectively. Display. In this way, after the separation pixels are selected in the two difference images, a common portion of the separation pixels in the two difference images is selected as the next step. The common portions of the spaced-out pixels in Figs. 11A and 11B are (0,1), (1,3), and (2,3). In the original image of FIG. 10 (a), surrounding the pixel values of the original pixels arranged at the same position as the common portion of the spaced pixels shown in FIGS. 11 (a) and 11 (b) is surrounded by a double border. The image shown by is shown in Fig. 11C. The double-rimmed pixel in Fig. 11C is an inspection target area to be inspected for defects, and includes pixels corresponding to the defects in Fig. 10A. In this way, the image processing algorithm of the present invention can be applied to the original image of a defective work having symmetry only with respect to either of the two symmetrical axes, as in the case of the original image having two symmetrical axes, and can be applied easily. The area to be inspected can be selected.

(4) Work conveyance and image processing of the present invention: using the third straight line L3

In the description of the image processing algorithm of the present invention using the above Figs. 3 to 6 and the like, the first straight line or the second straight line serving as the symmetric axis of the original image was horizontal or vertical. However, the axis of symmetry in the algorithm of the present invention is not limited to these. As an example of another axis of symmetry, the inclined axis of symmetry will be described with reference to FIGS. 12 and 13.

12 is a view showing the shape and dimensions of an original image generated by imaging a good work WG3 (hereinafter, the work WG3) and then extracting the work WG3 using the above template matching method. The work WG3 is a square with one side a, and at one point P30 of one side and one point P31 of the side adjacent to the side, a circular display MG31 in contact with each side has a diameter d. It is marked. Here, the values of a and d are the same as the work WG2 shown in Fig. 7A. In addition, the circular display MG32 having the same dimensions as the display MG31 so that the circular display MG31 is not in contact with the work WG3 at each point P32 and P33 of the remaining two sides. ) Is marked. There is a relationship of d < a / 2 between the length a / 2 from the edge of the work WG3 serving as one end of each side to the midpoint P0 and the diameter d of the marks MG31 and MG32. The work WG3 of FIG. 12 is line symmetrical with respect to the third straight line L3 which is the right lowering connecting the upper left corner and the lower right corner.

The original image in which the work WG3 shown in Fig. 12 is modeled similarly to Fig. 8A is shown in Fig. 13A. In Fig. 13A, three pixels (2,0) (3,0) (3,1) correspond to the display MG31 in Fig. 12. Similarly, three pixels of (0,2) (0,3) (1,3) in Fig. 13A correspond to the display MG32 in Fig. 12. These six pixel values are values similar to the pixels of (1,0) and (2,0) in Fig. 8A. The 10 pixels other than the 6 in Fig. 13A are similar values to the 14 pixels other than (1,0) and (2,0) in Fig. 8A. Here, it can be seen that the arrangement of the pixels in FIG. 13A is line symmetry in which the third straight line L3, which is the right lowering connecting the upper left and lower right of the original pixel, is a symmetric axis. In other words, the folding in this case is the upper, lower, left and right foldings if the same expressions as in the vertical folding or left and right folding in FIG. 3 are made.

When the difference calculation process and the difference image generation process are executed on the original pixel of Fig. 13A as in Fig. 3, the addresses of one end and the other end of the third straight line L3, which are also symmetrical axes, are different as described above. Pass to the output software. On the other hand, in the case of the third straight line L3, unlike in the case of the first straight line L1 in FIG. 3 (a), for example, the address of one end and the other end is the address of the pixel arranged in the original pixel. Overlap. Specifically, the address of one end of the third straight line L3 in Fig. 13A is (0,0), and the address of the other end is (3,3). 13 (a), the third straight line L3 overlaps the pixels of (1, 1) and (2, 2) not only in the one end and the other end, but also in other passages. The difference calculation process and the difference image generation process in the case where the address where the axis of symmetry passes and the address of the pixel overlap are described below.

First, when the third straight line L3 in Fig. 13A is a symmetric axis, it is considered how the original image is divided into the same number of pixels in the upper right and lower left by the third straight line. As described above, the pixels of (0,0), (1,1), (2,2), and (3,3) are on the third straight line L3. For this reason, these four pixels cannot be divided into the upper right and lower left by the third straight line as the axis of symmetry. Therefore, it is assumed that these four pixels are not targeted in the difference calculation process. In a specific algorithm, the difference pixel values of these four pixels are set to zero. The meaning of the difference pixel value 0 may be considered as a result of either excluding a pixel from the object of difference calculation or calculating a difference with the pixel itself.

In Fig. 13A, for 12 pixels other than the 4 pixels overlapping with the third straight line L3, the pixel values of the pixels arranged symmetrically at the upper right and lower left of the third straight line Calculate the difference. For example, a pixel value of (2,0) may calculate a difference from a pixel value of (0,2), and a pixel value of (2,1) may calculate a difference from a pixel value of (1,2). The difference image thus generated is shown in Fig. 13B. In Fig. 13B, the pixel values of (0,0), (1,1), (2,2), and (3,3) are all 0, as described above, and 12 other pixel values Since the is arranged symmetrically with respect to the third straight line L3, the pixel value is extremely close to zero.

(5) Defect inspection method of the present invention

The defect inspection method of the work using the image processing algorithm of the present invention described so far will be described below with reference to Figs.

14 is a schematic flowchart of a defect inspection method of a work using the image processing algorithm of the present invention. This schematic flowchart is composed of two modes. 14A is a threshold setting mode, and a differential image is generated using the image processing algorithm of the present invention from an original image of a plurality of defective workpieces, and an inspection target area of the work is selected using the generated differential image. This is the process of setting a threshold for doing so. 14B is an inspection execution mode, and an inspection target area is selected from the original image of the work to be inspected using the threshold set in the threshold setting mode, and then defect inspection is performed on the area. Process. The threshold setting mode and the test execution mode are connected by a combiner of the number 101 shown at the bottom of FIG. 14 (a) and the top of FIG. 14 (b).

15 to 31 are detailed flowcharts of a defect inspection method of a work using the image processing algorithm of the present invention. Specifically, the detailed steps of the definition completion process of step numbers S1 to S14 shown in FIG. 14 are written. In Figs. 15 to 31, the process name and step number of the definition completion process described in Fig. 14 are written at the top, and a terminal labeled IN indicating the entrance of the definition completion process is disposed below it. Next, following this terminal, detailed steps of the definition completion process are shown. And when all the steps are completed, the terminal labeled OUT, which indicates an exit for exiting from the definition completion process and proceeding to the next definition completion process shown in Fig. 14, is placed at the bottom.

In addition, the step numbers in FIGS. 15 to 31 are assigned the step numbers in FIG. 14 to the upper 1 digit or the upper 2 digits in order to indicate the correspondence with FIG. 14, and the details are assigned to the lower 2 digits that follow. Step numbers in the flowchart are assigned in ascending order from 01.

In addition, as will be described later, in the defect inspection method of a work using the image processing algorithm of the present invention, common processing executed in some defined completion processing shown in Fig. 14 is subrouted. The subroutine consists of Sub1 and Sub2, and FIG. 20 shows the detailed flowchart of Sub1, and FIG. 23 shows the detailed flowchart of Sub2. The step number in these detailed flowcharts is set to 21 in the upper 2 digits for Sub1 in FIG. 20 and 22 in the Sub2 in FIG. 23. For the two digits below the step number, the granting methods are different in Figs. In Sub1 in Fig. 20, the second digit from the bottom is set to 5 or 6 as a number indicating the correspondence between the step and S6 and S6 shown in Fig. 14 (a). Then, the lowest digit is assigned as a step number within the step S5 or S6, and numbers are assigned in ascending order from 1. On the other hand, in Sub2 in Fig. 23, since Sub2 corresponds only to Step S7 shown in Fig. 14A, the second digit from the bottom is fixed to the number 0. Then, the lowest digit is assigned as a step number in step S7 in the order from 1 to ascending order.

In addition, in the steps shown in FIGS. 15 to 31 and corresponding descriptions, the notation that an image is stored in a register is used several times. The notation of storing this image is organized into one array (for example, a three-dimensional array) by associating the X address and Y address of the pixel defined by FIG. 2 with the pixel values of the pixel, and constituting the image. This means that an array group consisting of an array number corresponding to the number of pixels is formed and stored in a register.

First, the threshold setting mode will be described with reference to FIGS. 14A and 15 to 26.

(5.1) Step S1

In Fig. 14A, step S1 following the terminal START is a known work imaging process. This is a process of preparing a plurality of workpieces each of which is known as a good product or a defective product, and imaging them. Fig. 15 shows a detailed flowchart of the known work imaging step (S1).

In step S101 of Fig. 15, A good-quality work product is imaged to obtain a good-quality image. It is assumed that these A pieces are good products. Next, it progresses to step S102, and the numbers from 1 to A are assigned to the quality goods image picked up in step S101. Subsequently, in step S103, B defective works are imaged to obtain a defective image. It is assumed that these B works are defective. Next, the flow proceeds to step S104, and the numbers 1 to B are assigned to the defective image captured in step S103. By the above steps S101 to S104, A good image and B bad images, each of which are assigned a unique number, are obtained.

After step S105, preparations are made for the execution of steps S2 to S7 following step S1 in Fig. 14A for B defective images. First, in step S105, 0 is stored in the register J that stores the number of the defective images being processed in steps S2 to S7, and the register J is initialized. Next, in step S106, 1 is added to the value of J. Then, in step S107, an image of the number designated by the value of J is taken out of the defective images in which numbers from 1 to B are assigned. Here, J = 1, and the taken out image becomes the defective image 1. Then, step S1 is finished and the process proceeds to step S2 shown in Fig. 14A.

(5.2) Step S2

In Fig. 14A, step S2 is a work extraction process to be inspected. This is a step of extracting the work from the captured image using the template matching method described above. 16 shows a detailed flowchart of the work extraction process (S2) to be inspected.

In step S201 of FIG. 16, using the template matching method, the outermost edge of the work is searched for and determined. Next, in step S202, the inside of the outermost edge of the work is taken as an image of the work to be inspected. Then, step S2 ends and the process proceeds to step S3 shown in Fig. 14A.

(5.3) Step S3

In (a) of FIG. 14, step S3 is a single color image generation process, and is a process of generating a single color image for the object to be examined extracted in step S2. 17 shows a detailed flowchart of the monochrome image generation process (S3).

In step S301 of Fig. 17, it is determined whether or not the image of the work to be inspected is a color. If the determination result is Yes, that is, in the case of a color image, the process proceeds to step S302. In step S302, K monochromatic images are generated from images of the inspection target work, which are color images. Here, K is a natural number. For example, as described above, if three color images are generated by decomposing a color image into three primary colors of R (red), G (green), and B (blue), K = 3. In addition, when the Y component of the YIQ signal, which is the component signal used in the step before obtaining the NTSC signal, is generated as described above, K = 1. After generating K monochromatic images from the color images in step S302, the process proceeds to the next step S303.

In step S303, numbers from 1 to K are assigned to the K monochromatic images generated in step S302 according to a predetermined criterion. Here, as an example of a predetermined criterion, a case where three monochromatic images are generated by decomposing a color image into three primary colors of R (red), G (green), and B (blue) will be described. In this case, 1 is given to the monochrome image generated based on R (red), 2 is given to the monochrome image generated based on G (green), and 3 to the monochrome image generated based on B (blue). The criterion for granting is considered. After assigning a number to the monochrome image in step S303, the process proceeds to step S304.

After step S304, preparations for executing steps S4 to S7 following step S3 in Fig. 14A are performed for the defective image 1. First, in step S304, 0 is stored in register N that stores the number of the monochrome image being processed in steps S4 to S7, and the register N is initialized. Next, in step S305, 1 is added to the value of N. In step S306, a monochromatic image numbered from 1 to K (a first monochromatic image in this case) of a number designated by the value of N is stored in a register named original image. Then, step S3 ends and the process proceeds to step S4 shown in Fig. 14A.

Steps S302 to S306 up to this point are executed when the image of the work to be inspected is judged to be a color image in step S301. If the image of the work to be inspected is not a color image, and the determination in Step S301 is No, then the procedure proceeds to Step S307 following Step S301. Here, it is a monochrome image that the image of the work to be inspected is not a color image, and the monochrome image generated from the work to be inspected is the image itself of the work to be inspected. This is equivalent to the number K = 1 of monochromatic images. Therefore, in step S307, the image of the work to be inspected is the first monochromatic image. Then, in step 308, 1 is stored in the register K indicating the number of monochrome images, and the process proceeds to step S309. If it progresses to step S309, 1 is stored in the register N which stores the number of the monochrome images processed by steps S4 to S7 following step S3 in FIG. 14A. Then, the process proceeds to step S306, where the Nth monochrome image, that is, the first monochrome image, is stored in a register named original image. Then, step S3 ends and the process proceeds to step S4 shown in Fig. 14A.

(5.4) Step S4

In Fig. 14A, step S4 is a symmetry determination process, and the symmetry of the monochrome image generated in step S3 is judged. 18 shows a detailed flowchart of the symmetry determination step (S4).

In step S401 of Fig. 18, in step S306 of the monochromatic image generation step (S3) shown in Fig. 17, the Nth monochromatic image stored in the register named original image is taken out, and it is determined whether it has a symmetric axis. If it does not have a symmetry axis, the judgment is No and proceeds to the terminal END, and the defect inspection is finished. The reason why the defect inspection is finished when the original image does not have a symmetric axis will be described below. The defect inspection method showing the schematic flowchart in Fig. 14 uses the image processing algorithm of the present invention shown in Figs. In addition, as described above, the image processing algorithm of the present invention is directed to an original image having a region that is linearly symmetric with respect to one or more symmetric axes. For this reason, the image processing algorithm of the present invention cannot be used for an original image that does not have a symmetric axis, and therefore the defect inspection is terminated at this point.

On the other hand, in step S401, when the Nth monochromatic image has a symmetry axis, the determination is Yes, and the procedure goes to step S402. In step S402, the number of symmetric axes of the original image is stored in the register C.

After step S403, preparation for executing steps S5 to S7 following step S4 in Fig. 14A is performed on the defective image 1. First, in step S403, 0 is stored in register M that stores the number of symmetry axes in the monochrome image that is being processed in steps S5 to S7, and register M is initialized. Next, in step S404, 1 is added to the value of M. Next, in step S405, a set of coordinates of the Mth axis of symmetry is stored in a register named ONEM. Then, in step S406, the coordinates of the other end of the M-th axis of symmetry are stored in a register named OTEM. These ONEM and OTEM are factors representing the coordinates of one end and the other end of the symmetry axis in the image processing algorithm of the present invention described using Figs. 3A, 5A, and 13A. In step S5 which will be described later, it is delivered to the subroutine Sub1. Thus, step S4 is ended and the process proceeds to step S5 shown in Fig. 14A.

(5.5) Step S5 and S6

In Fig. 14A, step S5 is a difference calculation process. And step S6 following it is a differential image generation process. For these processes, the image processing algorithm of the present invention shown in Figs. 3 to 6 can be used. Since this image processing algorithm is used multiple times in subsequent processes, it is defined by the subroutine Sub1 that summarizes these two processes into one in a detailed flowchart.

19 shows the difference calculation process (S5) and the difference image generation process (S6). As described above, these two processes are constituted by the subroutine Sub1 defined as step S21. As shown in step S21, the factors of the subroutine Sub1 are the original image, the coordinates of one end of the M target axis (ONEM), and the coordinates of the other end of the M symmetric axis (OTEM).

The original image as the first argument is the original image set in step S3 of FIG. 14A in the main program, specifically, the step N in FIG. 17, that is, the Nth monochrome image. For the original image set in step S306, that is, the N-th monochrome image, as described above, in step S4 of Fig. 14A, in detail, symmetry is determined in step S401 shown in Fig. 18. If the determination is Yes, steps S402 to S406 are executed as described above. As described above, ONEM as the second argument is a register that stores a set of coordinates of the Mth axis of symmetry in step S405. Likewise, as described above, the OTEM as the third argument is a register that stores the coordinates of the other end of the M-th axis of symmetry in step S406.

The subroutine Sub1 that received the original image, ONEM, and OTEM as the three arguments from the main program is as described above in step S5 (difference calculation process) in Fig. 14A and the step S6 (differential image) following it. Creation process).

Fig. 20 shows a detailed flowchart of the subroutine Sub1. In Fig. 20, steps S2151 to S2153 following the terminal Sub1 (original image, ONEM, OTEM) are steps S5 (difference calculation process) in Fig. 14A.

First, in step S2151, ONEM and OTEM are connected by a straight line to generate a symmetric axis. Generating the axis of symmetry by this step is the first straight line L1 in the description of the image processing algorithm of the present invention, which has already been used in Figs. 3A, 5A and 13A. , It is described as a process of connecting the coordinates of one end and the other end of the second straight line L2 and the third straight line L3 in a straight line. After generating the axis of symmetry in step S2151 of FIG. 20, the process proceeds to step S2152.

In step S2152, the original image is divided into two with the same number of pixels on both sides of the axis of symmetry. In this step, for example, the 16 original pixels shown in Fig. 5 (a) are divided into 8 original pixels located on the upper side of the first straight line L1 as the axis of symmetry, and 8 original pixels located on the lower side. Corresponds to doing. After dividing the original pixel into two at step S2152 of FIG. 20, it progresses to step S2153.

In step S2153, the difference pixel values BAB, which are the differences between the pixel values BA and BB, are calculated for the original pixels PA and PB arranged at positions of line symmetry on both sides of the symmetry axis. In this step, for example, in FIG. 5 (a), the original pixel (pixel value 200) located at (0,0) is referred to as the original pixel PA, and the original pixel (pixel value) located at (0,3). When 103) is referred to as the original pixel PB, the difference between each pixel value 200 and 103 is calculated as 200-103 = 97, and this value corresponds to the difference pixel value BAB. If the difference pixel value is calculated in step S2153 of Fig. 20, step S5 (difference calculation process) ends, and the process proceeds to step S6 (difference image generation process) in Fig. 14A in subroutine Sub1. . The difference image generation process is composed of step S2161 as shown in FIG. 20.

In step S2161, the difference pixel PAB having the difference pixel value BAB is arranged at the positions of the original pixels PA and PB in the original image to generate a difference image. In this step, for example, in FIG. 5 (a), the difference pixel value BAB of 97 is obtained by the original pixel PA positioned at (0,0) and the original pixel PB positioned at (0,3). When generated, it corresponds to disposing a difference pixel having a pixel value of 97 at the positions (0, 0) and (0, 3) in Fig. 5B. 20, when step S2161 ends, step S6 (differential image generation process) ends. Then, at the same time, the subroutine Sub1 as step S21 in Fig. 19 ends and returns to the main program. At that time, the difference image generated in step S2161 in Fig. 20 is transferred to the main program as a return value. In Fig. 20, this is indicated as Return (differential image) in the terminal following Step S2161. The return destination to the main program is the input of the threshold range setting process described as step S7 in Fig. 14A.

(5.6) Step S7

In Fig. 14A, step S7 is a threshold value range setting process. This is the difference image generated in step S6 in the difference image, such as (d) (e) of FIG. 3 or (d) (e) of FIG. 5 used in the description of the image processing algorithm of the present invention described above. It is a process of arranging the difference pixel values in descending order to calculate difference values between adjacent difference pixels, and selecting a section in which the difference value becomes the maximum.

The detailed flowchart of the threshold value range setting process (S7) is shown in FIGS. 21-23. In the first step S701 in FIG. 21, the difference image generated in the subroutine Sub1 shown in FIG. 20 is stored in a register named difference image NJ. Here, J and N as two numbers used in the register name of the difference image NJ will be described.

J in this name is a number corresponding to J of the defective image J taken out in step S107 of the known work imaging step S1 shown in FIG. 15. Then, as described above, first, in the work object extraction step (S2) shown in FIG. 16, the image of the work object to be inspected is extracted from the defective image J using the template matching method, and then the solid color shown in FIG. In the image generation step (S3), one or more monochrome images are generated from the image of the work to be inspected. The monochromatic images are numbered from 1 to K (K is 1 or more) (steps S302 and S307), and in step S306, the Nth monochromatic image is stored in a register named original image. Here, as described above, N is a value of the register N that stores the number of monochrome images being processed in steps S4 to S7 shown in Fig. 14A. And this N coincides with N in the register name called differential image NJ. In step S306, the original image in which the Nth monochrome image is stored is, as described above, in the symmetry determination step (S4) shown in FIG. 18, the difference calculating step (S5) shown in FIG. 19, and the difference image generating step (S6). By processing sequentially, a difference image is generated. In summary, in the register called differential image NJ, a differential image generated from the Nth monochrome image generated based on the defective image J is stored.

As described above, after the difference image is stored in the difference image NJ in step S701, the process proceeds to step S22. Step S22 is defined as subroutine Sub2 because it is a common process executed in several steps shown in Fig. 14 as described above. The subroutine Sub2 takes the difference image as an argument. Here, in the subroutine Sub1 shown in Fig. 20, the difference image generated in step S2161 is returned to the main program as a return value. The subroutine Sub2 receives the difference image as an argument from the main program. Then, by using the received difference image, the difference pixel values in the difference image are arranged in descending order as shown in (d) (e) of FIG. 5 used in the description of the image processing algorithm of the present invention described above. The process of calculating the difference value and selecting the section where the difference value becomes the maximum is executed. The detailed flowchart of the subroutine Sub2 is shown in Fig. 23.

Step S2201 in FIG. 23 corresponds to a process in which the pixel values are written in descending order as the heading in FIG. 5 (d) (e). That is, in step S2201, numbers 1 to X are first assigned to the X pixels constituting the difference image. Next, the pixel values PV (1) to PV (X) of the pixels of each number are stored in X array APs 1 to AP (X) in descending order from the maximum value to the minimum value. The X pixel values PV (1) to PV (X) stored in X arrays from the AP (1) to the AP (X) are sequentially recorded from left to right, as shown in FIG. 5 (d) (e). Is the number placed at the top. As described above, the number in Fig. 5 (d) corresponds to Fig. 5 (b) which is a difference image generated by using the first straight line L1 as the symmetry axis in Fig. 5 (a), The number in FIG. 5 (e) corresponds to FIG. 5 (c) which is a differential image generated by using the second straight line L2 as the axis of symmetry in FIG. 5 (a). After executing step S2201 in this way, the process proceeds to step S2202.

In step S2202, 0 is stored and initialized in the register S that stores the number of the pixel being processed in step S2204 described later. Next, in step S2203, 1 is added to the numerical value stored in the register S. In this step, the numerical value of the register S is 1, indicating that the pixel assigned to No. 1 in S2201 is processed. When the value of the register S is confirmed, the process proceeds to step S2204.

Step S2204 corresponds to a process in which a difference value of adjacent difference pixel values is written as a heading in the middle in FIG. 5 (d) (e). In step S2204, a difference pixel value (NP) that becomes a difference between pixel values stored in two adjacent arrays, from the maximum pixel value stored in the array AP 1 to the minimum pixel value stored in the array AP (X) ) Is sequentially calculated. That is, if expressed using the value of the register S determined in step S2203, AP (S + 1) -AP (S) is calculated and the result is stored in a register named NP (S). Here, since the total number of pixels is X, the value of S is sequentially added one by one from 1 to (X-1) and processed in step S2204. The processing of the difference calculation and storage is described as NP (S) ← AP (S + 1) -AP (S) in step S2204 according to the notation of the software. In step S2204, the difference pixel values stored in the register NP (S) are sequentially displayed in correspondence with numbers and arrows arranged at the top in FIG. 5 (d) (e). . When step S2204 ends, the process proceeds to step S2205.

In step S2205, it is determined whether or not the number S of pixels processed in step S2204 is equal to (X-1), which is a value smaller than the total number of pixels by one. That is, it is judged whether or not the processing of step S2204 has been completed for all X pixels constituting the difference image. If the determination result is No, the flow returns to step S2203, and 1 is added to the value of the register S. Accordingly, the number of pixels to be processed in step S2204 is increased by one. Then, in step S2204, a process is performed for the pixel whose number is increased by one. If the process in step S2204 is completed for all X pixels repeatedly, that is, S = X-1, the determination result in step S2205 is Yes. In that case, the process proceeds to step S2206.

Step S2206 corresponds to a process of selecting the maximum section among the processes written as a threshold in the maximum section of the difference value as a heading at the bottom in FIG. 5 (d) (e). The threshold is set in the main program after the subroutine Sub2 is finished, and the high and low limit values (high lower limit values) and low upper limit values (hereinafter referred to as lower limit values) described later are returned to the main program as return values. In step S2206, among the (X-1) difference pixel values from NP (1) to NP (X-1) generated in step S2204, the maximum difference value MXNP is selected. This selects a maximum value of 58 among the difference values of adjacent difference pixel values written in the middle in Fig. 5B, or the maximum value among the difference values of adjacent difference pixel values written in the middle in Fig. 5C. Corresponds to the election of 66. In this way, if the maximum difference value MXNP is selected in step S2206, the process proceeds to step S2207.

Step S2207 is a preparation for executing the setting of the threshold value in the main program during the process written as a threshold value in the maximum section of the difference value as a heading at the bottom in FIG. 5 (d) (e). It is fair. In step S2207, the pixel values PV (MX + 1) and PV (MX) stored in two adjacent arrays AP (MX + 1) and AP (MX) used for calculating the maximum difference value (MXNP) selected in step S2206. ) Is defined as a high-limit value (HBP) and a low-limit value (LTP), respectively. Specifically, PV (MX + 1) is stored in a register named HBP, and PV (MX) is stored in a register named LTP at the same time. This storage processing is described in HBP ← PV (MX + 1) and LTP ← PV (MX) in step S2207 according to the notation of the software.

Here, as shown in Fig. 5 (d) (e), a problem caused by using the threshold value generated from only one difference image as it is, and a process for solving the problem will be described.

As described above, in FIG. 5 (d) (e), the difference value of the adjacent difference pixel value is indicated at the middle. In addition, a threshold value calculated in the maximum section of the difference value is shown at the bottom. The threshold at the lower end is for selecting an inter-pixel value from the difference pixel values described at the upper end, and the median value of the differential pixel value (upper) at both ends of the maximum section is calculated to be a threshold value. Specifically, in the case of Fig. 5D, since 58 is selected as the section having the maximum value in the middle, the median values of 65 and 7, which are differential pixel values at both ends of the section, are calculated at the lower end. The median is 36, so the threshold is 36. Similarly to Fig. 5E, 66, which is the section having the maximum value at the middle, is selected, and the median values of 71 and 5, which are differential pixel values at both ends of the section, are calculated at the lower end. The median is 38, so the threshold is 38.

On the other hand, Fig. 5 (d) (e) is an explanatory diagram of the image processing algorithm of the present invention, and the original image is only one shown in Fig. 5 (a). However, since the difference image processed by the subroutine Sub2 shown in Fig. 23 is generated through the following steps, there are multiple. In the process of generating a differential image, as described above, one or more single-color images based on the defective image J as one of the B defective images (step S103 in FIG. 15) (step S107 in FIG. 15) are initially described. It is created (step S302 or S307 in Fig. 17). Next, among the one or more (K) monochrome images, the Nth monochrome image is used as the original image (step S306 in FIG. 17), and a differential image is generated from the original image (steps S2151 to S2161 in FIG. 20).

The difference image thus generated is described as an argument in the terminal Sub2 located at the top of the detailed flowchart of the subroutine Sub2 shown in FIG. 23. That is, in the case of two or more single-color images generated based on the same defective image (J), the sub-routine Sub2 includes all the difference images generated from each of the plurality of single-color images from the first monochrome image to the K-th monochrome image. Process. The number of defective images J for generating these K monochromatic images is not one, but plural (B).

As is clear from these points, the subroutine Sub2 sequentially processes a plurality of difference images. As described above, these plural difference images are respectively generated based on the B individual defect images. For this reason, the pixel value of the difference image processed by the subroutine Sub2 also deviates in response to the deviation of the pixel value in the B defective images. That is, in (d) (e) of FIG. 5, the difference pixel value shown at the upper end varies for each difference image. Due to this, a difference between the difference values of adjacent difference pixel values indicated in the middle also occurs for each difference image. Therefore, there arises a problem that the threshold value shown at the lower end is deviated for each difference image and is not determined as one. This is the "first problem." In order to solve this first problem, it is necessary to solve the difference in the difference values for each difference image, so that only one threshold value can be set for a plurality of difference images generated from all (B) defective images. Do.

In addition, as described above, the difference image processed by the subroutine Sub2 is generated from each of a plurality of (K) monochromatic images generated from one defective image J. For example, as described above, if three defective monochrome images are generated by dividing a defective image (J), which is a color image, into three primary colors of R (red), G (green), and B (blue), three ( A monochrome image of K = 3) is generated. In addition, as described above, these three monochromatic images represent the pixel values of each component in a gray scale corresponding to the ratio of each component of the three primary colors in the color image.

Then, all three kinds of difference images generated from each single color image are sequentially processed by subroutine Sub2. In that case, there is no guarantee that the maximum section of the difference value of the adjacent difference pixel values shown in the middle of Fig. 5D (e) will be the same corresponding to each difference image. This is because, as described above, the component ratios of the three primary colors of the R (red), G (green), and B (blue) in each pixel of the defective image J are different for all pixels.

Due to this, the pixel values of the pixels arranged at the same position in the K difference images are different depending on the component ratio. Therefore, for each of the K difference images generated from the same defective image J, the difference pixel values shown at the top of Fig. 5 (d) (e) are different. Therefore, even when the difference value of the adjacent difference pixel value indicated in the middle is calculated, the result of the calculation can be a completely different value for each of the K difference images. That is, the maximum period of the difference value of the adjacent difference pixel values corresponding to the difference image generated from each of the plurality of monochrome images may be a completely separate section corresponding to the monochrome image. In that case, when setting the threshold value using the maximum section of the difference value, a problem arises that a criterion for selecting a monochrome image for threshold setting is required. This is the "second problem." In order to solve this second problem, it is necessary to clarify the selection criteria and select a monochrome image that is optimal for setting a threshold among a plurality of monochrome images generated based on one defective image.

In the threshold value range setting step (S7) shown in Figs. 21 to 23, in order to solve these problems, the difference image is processed by the following steps. First, in step S2207 shown in Fig. 23, as described above, the pixel values PV (MX + 1) and PV (MX) are defined as the high-low limit value HBP and the low-high limit value LTP, respectively. Then, as described in the terminal return at the bottom of FIG. 23, the high-low limit value HBP and the low-high limit value LTP are transmitted as a return value to the main program of the threshold range setting process S7 shown in FIG.

In Fig. 21, the main program receives the return value from the subroutine Sub2 in step S22, and executes the process in step S702 onwards described later. By this method, it is possible to set only one threshold value for a plurality of differential images generated from all (B) defective images. Then, among the K difference images corresponding to each of all the monochrome images (K for each defective image) generated from all (B) defective images, a difference image that is optimal for setting the threshold value based on a standard is selected. . The selection of the optimal difference image is the same as the selection of the optimal monochrome image.

21 and 22, the process of enabling a single threshold value to be set for a plurality of differential images generated from all defective images, and corresponding to each of all monochrome images generated from all defective images A process of selecting a difference image that is optimal for setting the threshold among the difference images to be described will be described.

In step S702 of the threshold value range setting step (S7) shown in FIG. 21, from step S22, i.e., from the subroutine Sub2 shown in FIG. 23, the lower limit value HBP and the lower limit value LTP as the return value (step S2207 in FIG. 23) And the bottom terminal Return). Then, the lower limit value HBP of these return values is stored in a register named the lower limit value HBPMNJ. Similarly, the low limit value (LTP) among the return values is stored in a register named low limit value (LTPMNJ). In step S702, the storage processing is described as a high-limit value (HBPMNJ) ← a high-limit value and a low-limit value (LTPMNJ) ← low-limit value in accordance with the software notation.

Here, J, N, and M as three numbers used in the register names of the high-low limit value (HBPMNJ) and the low-high limit value (LTPMNJ) will be described. The 2nd and 3rd numbers N and J in this name are the same as N and J in the register name called the difference image NJ described in step S701. Since these N and J have already been described, a detailed description is omitted. If only the conclusion is described, it means that the lower limit value (HBP) and the lower limit value (LTP) of the difference image generated from the Nth monochrome image generated based on the defective image J are.

Next, the first number M in this register name will be described below. As described above, in step S401 in the symmetry determination step S4 shown in Fig. 18, it is determined whether the Nth monochrome image (original image) has a symmetry axis. When the determination result is Yes, the number of symmetric axes of the original image in step S402 is stored in the register C. Then, in step S403, 0 is stored in register M and initialized. Next, in step S404, 1 is added to the register M. Here, the register M stores the number C of the axis of symmetry of the original image to be processed by steps S5 (difference calculation process) to step S7 (threshold range setting process) of Fig. 14 (a) which becomes a post process. Is a register to do. Then, in steps S405 and S406, the coordinates of one end and the other end of the Mth symmetric axis corresponding to the register M are stored in the registers ONEM and OTEM.

Then, as is clear from Figs. 21 to 23 as a detailed flowchart of step S7 (threshold range setting process) shown in Fig. 14A, the register M is the symmetry axis processed by the subroutine Sub2 shown in Fig. 23. Number. Summarizing the above, the high-limit value (HBPMNJ) and the low-limit value (LTPMNJ) are the high-limit values of the differential images generated on the basis of the M symmetry axis from the N-th monochrome image generated based on the defective image J, respectively. HBP) and Low Upper Limit (LTP).

As described above, in step S702 shown in Fig. 21, after setting the contents of each register of the high-low limit value HBPMNJ and the low-high limit value LTPMNJ, the process proceeds to step S703. In step S703, it is determined whether the number of symmetric axes stored in the register M has reached C, the number of symmetric axes. The register M is added one by one in step S404 of the symmetry determination step (S4) shown in FIG. The original image corresponding to the added register M value is processed by FIGS. 19 to 23 as described above. In step S702 of FIG. 21, the contents of each register of the high-low limit value HBPMNJ and the low-high limit value LTPMNJ are set.

The determination in step S703 of Fig. 21 is to determine whether or not the contents of each register of the high-low limit value (HBPMNJ) and low-high limit value (LTPMNJ) are set corresponding to all of the C symmetry axes of the Nth monochrome image. It is no different. If the result of this judgment is No, since the contents of these registers corresponding to all the axes of symmetry are not set, Fig. 18 (symmetry determination step (S4)) via a joiner in which the jump destination number 204 is described. Jump to step S404. In this step S404, 1 is added to the value of the register M that stores the number of symmetry axes. Then, similarly to the description so far, steps S405 and S406 in Fig. 18 and steps shown in Figs. 19 to 23 are executed again in a state where the value of M is increased by one. If the addition of 1 to the value of M is repeated, the determination result in step S703 of Fig. 21 (threshold range setting process (S7)) becomes Yes. In that case, the process proceeds to step S704.

In step S704, it is judged whether or not the number of monochrome images stored in the register N has reached K, which is the number of monochrome images. The register N is added one by one in step S305 of the monochrome image generation process (S3) shown in FIG. The original image corresponding to the added register N value is processed by FIGS. 18 to 23 as described above. In step S702 of FIG. 21, the contents of each register of the high-low limit value HBPMNJ and the low-high limit value LTPMNJ are set. In step S704 of FIG. 21, the contents of each register of the lower limit value HBPMNJ and the lower limit value LTPMNJ correspond to all of the K single-color images generated based on the J-th defective image. It is no different from whether or not it has been set. If the result of this judgment is No, since the contents of these registers corresponding to all the monochrome images are not set, step S305 in FIG. 17 (monochrome image generation step (S3)) via a combiner in which the jump destination number 203 is described. To jump. In this step S305, 1 is added to the value of the register N that stores the number of monochrome images. Then, as in the description so far, the steps S306 in Fig. 17 and the processes described in Figs. 18 to 23 are again performed in a state where the value of N is increased by one. If the addition of 1 to the value of N is repeated, the determination result in step S704 in FIG. 21 (threshold range setting step (S7)) becomes Yes. In that case, the process proceeds to step S705.

In step S705, it is judged whether or not the number of defective images stored in the register J has reached B, the number of defective images. The register J is added one by one in step S106 of the known work imaging step (S1) shown in FIG. 15. The defective image corresponding to the value of the added register J is processed by FIGS. 16 to 23 as described above. In step S702 of FIG. 21, the contents of each register of the high-low limit value HBPMNJ and the low-high limit value LTPMNJ are set. The determination in step S705 of FIG. 21 is equivalent to the determination as to whether or not the contents of each register of the high and low limit values HBPMNJ and LTPMNJ are set corresponding to all of the B defective images. If the result of this judgment is No, since the contents of these registers corresponding to all the defective images are not set, step S106 in Fig. 15 (base work imaging step (S1)) via a combiner in which the jump destination number 202 is described. To jump. In this step S106, 1 is added to the value of the register J that stores the number of defective images. Then, similarly to the description so far, in the state where the value of J is increased by 1, the steps described in step S107 of FIG. 15 and the steps of FIGS. 16 to 23 are executed again. If the addition of 1 to the value of J is repeated, the determination result of step S705 in Fig. 21 (threshold range setting step (S7)) becomes Yes. In that case, the process proceeds to step S706.

As is clear from the above description, the three kinds of judgments in steps S703 to S705 in Fig. 21 are the following three judgments. First, in step S703, it is determined whether or not the lower limit value HBP and the lower limit value LTP as the return value from the subroutine Sub2 are generated for all symmetry axes of the monochrome image. Next, in step S704, it is determined whether or not the return value from subroutine Sub2 is generated for all the monochrome images generated based on the defective image. Then, in step S705, it is determined whether or not the return value from subroutine Sub2 is generated for all the defective images.

When all of these three judgment results are Yes, all three numbers are used for J, N, and M as three numbers used in the names of the register high-low limit value (HBPMNJ) and low-high limit value (LTPMNJ). The processing is completed. That is, the lower limit of the register (HBPMNJ) and the lower limit of the limit (LTPMNJ) by the steps up to this point, for all the monochrome images generated based on the defective image, the lower limit of the difference image related to all symmetry axes of the monochrome image (HBP) and a low upper limit value (LTP) are generated and stored for the entire defective image.

Next, in steps S706 and S707 of Fig. 21, the register M for storing the number of symmetry axes is initialized again, and 1 is added to the value. Then, it jumps to step S708 in Fig. 22 (threshold range setting process (S7)) via a combiner in which the jump destination number 206 is described.

The threshold value range setting process (S7) shown in FIG. 22 processes the plurality of high and low limit values stored in the register high and low limit values (HBPMNJ) and the low and high limit values (LTPMNJ) in step S702 in FIG. It is a process to be executed. As described above, this process has two purposes. The first object is to enable a single threshold value to be set for a plurality of differential images generated from all (B) defective images. The second object is to select a difference image that is optimal for setting the threshold among K difference images corresponding to each of all monochrome images (K for each bad image) generated from all (B) bad images. .

In the first step S708 in Fig. 22, the difference image generated based on the Mth axis of symmetry is used from each of the K monochromatic images corresponding to each defective image from the defective image 1 to the defective image B. The minimum lower limit value (MNBMN), which is the minimum among (B × C) calculated high limit values (HBPMNJ) (1≤M≤C, 1≤J≤B for each of N where 1≤N≤K), is calculated. Elect.

Next, the process proceeds to step S709, and the difference images generated based on the M symmetry axis are calculated from each of the K monochrome images corresponding to each defective image from the defective image 1 to the defective image B. A maximum upper limit value MXTMN that becomes the maximum value is selected from among the B lower limit values LTPMNJ (1 ≤ M ≤ C and 1 ≤ J ≤ B for each of N having 1 ≤ N ≤ K).

In this way, in step S708, a minimum value is selected from among (B × C) high-limit values (HBPMNJ) (1≤M≤C, 1≤J≤B for each of N, where 1≤N≤K), and then step The meaning of selecting the maximum value among the B low upper limit values LTPMNJ (1 ≤ M ≤ C and 1 ≤ J ≤ B for each of N having 1 ≤ N ≤ K) in S709 will be described below.

The lower limit value HBP is generated in steps S2206 and S2207 of the subroutine Sub2 shown in FIG. 23. Then, as described above, step S2206 corresponds to a process of selecting the maximum section among the processes written as a threshold in the maximum section of the difference value as a title at the bottom in FIG. 5 (d) (e). This maximum section is, for example, in the case of Fig. 5D, a section in which 58 as the maximum difference value is generated in the middle of interruption, that is, a section in which difference pixel values 65 and 7 are used as adjacent difference pixel values at the upper end. Then, in step S2207 in Fig. 23, two adjacent arrays AP (MX + 1) and the pixel values PV (MX + 1) and PV (MX) stored in the two adjacent arrays AP (MX + 1) used to calculate the maximum difference value (MXNP). ) Are respectively set as the high-limit value (HBP) and the low-limit value (LTP). When the process of step S2207 is correlated with Fig. 5D, 65 of the difference pixel values described at the top of Fig. 5D are the lower limit value HBP and 7 is the lower limit value LTP.

As described above, there are variations in the pixel values of each of the B defective images. For this reason, variations exist in the pixel values of the K monochromatic images generated from each defective image corresponding to the defective image. Due to this, when a difference image is generated from each monochromatic image, a variation exists in the pixel value of the difference image in response to the variation in the pixel value of the defective image that is the basis of each difference image. When a plurality of difference images in which these deviations are generated is processed by the subroutine Sub2 in Fig. 23, deviations are generated between the lower limit value HBP and the lower limit value LTP generated from each difference image. That is, if the original image shown in Fig. 5 (a), that is, the Nth monochrome image (step S306 in Fig. 17) is changed corresponding to the B defective images, the value of each stage in Fig. 5 (d) (e) There are deviations. As a result, the high-low limit value HBP and the low-high limit value LTP generated in step S2207 in Fig. 23 are in correspondence with the deviation of the pixel values of the B defective images.

Due to variations in the high and low limit values HBP and LTP for each difference image, if the threshold value is set as described at the bottom of FIG. 5 (d) (e), the threshold value is Differences occur for each difference image.

Therefore, even if it is an N-th monochrome image generated from any of the B defective images, a process for making it possible to set the same threshold value for the difference image generated from the N-th monochrome image is necessary. As the process, steps S708 and S709 in Fig. 22 are performed.

The lower limit value HBPMNJ processed in the process of step S708 is located on the left of the upper difference pixel value corresponding to the maximum section of the difference value of the adjacent difference pixel values described in the middle of Fig. 5D. It corresponds to a large pixel value. This pixel value is hereinafter referred to as a maximum section large pixel value.

Similarly, the low upper limit value LTPMNJ processed in the process of step S709 is located on the right side of the upper difference pixel value corresponding to the maximum section of the difference value of the adjacent difference pixel values described in the middle of Fig. 5D. Corresponds to a small pixel value. This pixel value is hereinafter referred to as a maximum section small pixel value.

That is, in step S708, the minimum lower limit value (MNBMN) that becomes the minimum value among the maximum interval conversation element values is selected. The minimum lower limit value (MNBMN) is calculated from each of the K monochrome images corresponding to the B defective images, using the difference image generated based on the Mth axis of symmetry, among the (B × C) maximum interval dialog values, It becomes the value closest to the corresponding maximum section digester value.

In addition, in step S709, the maximum upper limit value MXTMN that becomes the maximum value among the maximum section digest factor values is selected. The maximum upper limit value MXTMN is calculated from each of the K monochrome images corresponding to the B defective images, using the difference image generated based on the Mth axis of symmetry, among the (B x C) maximum section digest factor values, It becomes the value closest to the corresponding maximum interval conversation element value.

That is, the combination of the minimum lower limit value MNBMN and the maximum upper limit value MXTMN makes the difference value in the maximum section of the difference value of the adjacent difference pixel value indicated in the middle in FIG. 5 (d) (e) the minimum value. It's like a combination. In this way, the meaning of the difference value being the minimum value is that the range in which the threshold value is set for each Nth monochrome image is common to all the differential images generated from the Nth monochrome image. Therefore, the section constituted by the combination of the minimum lower limit value MNBMN and the maximum upper limit value MXTMN is a section in which it is possible to reliably set a threshold value for all of the (B × C) difference images. In this way, after selecting the minimum lower limit value MNBMN and the maximum upper limit value MXTMN, the process proceeds to step S710.

In step S710, the difference range (RN) = minimum lower limit value (MNBMN)-maximum upper limit value (MXTMN) is calculated for the minimum lower limit value (MNBMN) and the maximum upper limit value (MXTMN) corresponding to each of the K monochrome images. This range (RN) is as described above of (B × C) difference images calculated from each of the K monochrome images corresponding to the B defective images using the difference image generated based on the Mth axis of symmetry. For all, the difference value of the section in which it is possible to reliably set the threshold value in the maximum section of the difference value of the adjacent difference pixel value (for example, the section having the difference value 58 in the middle of Fig. 5D) to be. If the range RN is calculated in step S710, step S7 ends, and the process proceeds to step S8 shown in Fig. 14A.

(5.7) Step S8

In Fig. 14A, step S8 is a threshold image generation process, and an optimal threshold image is selected for setting a threshold for selecting an inspection subject area, and a threshold is set. 24 shows a detailed flowchart of the threshold image generation process (S8).

In step S801 of Fig. 24, it is determined whether or not K, which is the number of monochrome images generated from the defective images, is 2 or more. This is the same content as in step S301 shown in Fig. 17 (single color image generation step S3), in which it is determined whether or not the image of the work to be inspected is a color image. That is, in step S801 of Fig. 24, if the image of the workpiece to be inspected is a color image, the number K of monochrome images is 2 or more, and if the image of the workpiece to be inspected is not a color image, the number K of monochrome images is 1, which is less than 2 It becomes. If the determination result in step S801 is Yes, that is, the number of monochrome images generated based on the defective image is 2 or more, the process proceeds to step S802.

In step S802, a maximum value is selected from among K ranges (RN) (1≤N≤K) calculated from all monochromatic images, and a difference image having a range (RN) that becomes the maximum value is the Mth threshold image. As elected. As described above, the number M of the Mth threshold image is the number M of the Mth axis of symmetry of the original image, that is, the Nth monochrome image. That is, a threshold image corresponding to the number of symmetry axes is selected.

Here, the meaning of selecting the difference image having the range RN to be the maximum value as the Mth threshold image is explained. As described above, as described above, it is possible to reliably set a threshold value for all ((B x C)) differential images generated from the Nth monochrome image generated based on the B defective images. It is the difference value of the decreasing section. This corresponds to the means for solving the above-described first problem, that is, the threshold value is not determined by one because of the variation in each difference image. In addition, selecting the maximum value from the range RN calculated in this way compares the range RN as the minimum difference value, and selects the maximum value from which it is possible to easily set the most critical value among them. It is no different. The selection of the difference image having the range RN which becomes the maximum value corresponds to the reference setting for solving the above-described second problem, that is, a problem that requires a criterion for selecting a monochrome image for threshold setting.

Summarizing the above, in order to solve the above-mentioned first problem that the threshold value shown at the bottom of FIG. 5 (d) (e) is deviated for each difference image and is not determined as one, the range in step S710 of FIG. 22 ( RN). Accordingly, the difference in the difference value for each difference image is resolved, and only one threshold is set for all ((B × C)) difference images generated from the Nth monochrome images generated based on the B defective images. The value can be set. In addition, in order to solve the above-mentioned second problem that a criterion for selecting a monochrome image for threshold setting is necessary, in step S802 of FIG. 24, the maximum of K ranges (RN) corresponding to K, which is the number of monochrome images, A value is selected, and a difference image having a range RN that becomes the maximum value is selected as the Mth threshold image. Accordingly, the selection criteria are clarified, and among the plurality of monochromatic images generated based on one defective image, a monochromatic image optimal for setting a threshold can be selected in a subsequent step S803.

On the other hand, when the determination result of step S801 is No, that is, the number of monochrome images generated based on the defective image is 1, the process proceeds to step S804. In step S804, a difference image generated from one single color image is taken as the Mth threshold image. After selecting the Mth threshold image in this way, the process proceeds to step S803.

In step S803, a monochrome image generating the Mth threshold image is selected as a monochrome image for the Mth threshold. This monochromatic image for the M threshold value is a monochromatic image used to calculate the threshold value for selecting the separated pixel values shown at the bottom of Fig. 5D (e) with respect to the Mth axis of symmetry of the original image. After selecting the monochrome image for the Mth threshold value, the process proceeds to step S805.

In step S805, in the Mth threshold image, the median value of the range RN = (minimum lower limit value (MNBMN) + maximum upper limit value (MXTMN)) / 2 is calculated, and the value is used as the threshold value for selecting an inspection target area Let it be the inspection threshold. The Mth inspection threshold value calculated in step S805 corresponds to the threshold value in the maximum section of the difference value shown at the bottom of Fig. 5D (e). After setting the Mth inspection threshold in this way, the process proceeds to step S806.

In step S806, it is determined whether the number of symmetric axes stored in the register M has reached C, the number of symmetric axes. The register M is added one by one in step S707 of the threshold value range setting process (S7) shown in FIG. The original image corresponding to the added register M value is processed by Figs. 22 and 24 as described above. Then, in step S805 of Fig. 24, the Mth inspection threshold is set. The determination in step S806 of Fig. 24 is nothing more than a determination as to whether or not the Mth inspection threshold is set corresponding to all of the C symmetry axes of the Nth monochrome image. When the determination result is No, since the Mth inspection threshold value corresponding to all the symmetry axes is not set, the step of FIG. 21 (threshold range setting process (S7)) via the combiner in which the jump destination number 205 is described. Jump to S707. In this step S707, 1 is added to the value of the register M that stores the number of symmetry axes. Then, as in the description so far, the processes of Figs. 22 and 24 are executed again with the value of M being increased by 10,000. If the addition of 1 to the value of M is repeated, the determination result in step S806 of Fig. 24 (threshold image generation process (S8)) becomes Yes. In that case, step S8 is terminated and the process proceeds to step S9 shown in Fig. 14A.

(5.8) Step S9

In Fig. 14A, step S9 is a threshold value confirmation process. This indicates that the pixel values of all the pixels of the difference image generated from the good image are smaller than the Mth inspection threshold, that is, even if the Mth inspection threshold is applied to the good image, it is represented by a double frame in Fig. 6C. It is a process of confirming that the inspection target area to which defect inspection as described above cannot be selected. 25 and 26 show detailed flowcharts of the threshold checking process (S9).

In FIG. 25, steps S901 and S902 are preparations for performing a threshold value confirmation process with respect to the A quality goods images obtained in step S101 shown in the known work imaging process S1 shown in FIG. First, in step S901, 0 is stored in the register I that stores the number of a good product image that is being processed in the threshold checking process, and the register I is initialized. Next, in step S902, 1 is added to the value of I.

Then, in step S903, an image of a number designated by the value of I is taken out of a quality product image numbered from 1 to A. Here, I = 1. Then, the process proceeds to step S904.

Steps S904 and S905 are the same as steps S201 and S202 of the work object extraction step (S2) shown in Fig. 16, respectively. That is, in steps S904 and S905, the image of the work to be inspected is generated from the good product image 1 using the template matching method. After the image of the work to be inspected is thus generated, the process proceeds to step S906.

Steps S906 and S907 are preparations for executing a subsequent step in the threshold value confirmation process for the difference image generated from the monochrome image generated based on the good product image 1. First, in step S906, 0 is stored in the register M that stores the number of symmetry axes in the monochrome image that is being processed in a subsequent step in the threshold checking process, and the register M is initialized. Next, in step S907, 1 is added to the value of M.

Next, the process proceeds to step S908, and it is determined whether or not the image of the work to be inspected is a color. If the judgment is Yes, that is, in the case of a color image, the process proceeds to step S909. In step S909, K monochromatic images are generated from the image of the work to be inspected, which is a color image. This step S909 is the same as step S302 in the monochrome image generation process S3 shown in FIG. 17. After generating K monochromatic images, the process proceeds to step S910.

In step S910, among the generated monochrome images, a monochrome image corresponding to the monochrome image that generated the Mth threshold image is taken as a monochrome image for the Mth threshold. As described above, it is step S803 of the threshold image generation process (S8) shown in FIG. 24 that the monochrome image for generating the Mth threshold image is selected as the monochrome image for the Mth threshold value. That is, the monochromatic image for the M threshold value selected in step S803 and the monochromatic image for the M threshold value selected in step S910 are correlated. The purpose of this correspondence is that, even if the Mth inspection threshold set using the monochrome image for the Mth threshold is applied to a good product image, the inspection subject as shown by the double border in FIG. This is to confirm that an area cannot be elected.

On the other hand, when the determination in step S908 is No, that is, when the image of the work to be inspected is a monochrome image, the process proceeds to step S911. In step 911, the monochromatic image, that is, the image of the work to be inspected is regarded as a monochromatic image for the Mth threshold. In this way, if the image for the Mth threshold value is selected in step S910 or S911, the process jumps to step S912 in Fig. 26 via a combiner in which the destination number 209 is described.

In step S912 of Fig. 26, a monochrome image for the Mth threshold value is stored in a register named original image. Next, the process proceeds to step S913, and a set of coordinates of the Mth axis of symmetry is stored in a register named ONEM. This step S913 is the same step as step S405 of the symmetry determination step S4 shown in FIG. 18. Next, the process proceeds to step S914, and the coordinates of the other end of the M-th axis of symmetry are stored in a register named OTEM. This step S914 is the same step as step S406 of the symmetry determination step S4 shown in FIG. 18. Then, the process proceeds to step S21, and the subroutine Sub1 shown in Fig. 20 is executed. Since the subroutine Sub1 has already been described, a detailed description is omitted here.

By executing steps S912 to S21 (subroutine Sub1) in FIG. 26, the image for the Mth threshold value selected in step S910 or S911 in FIG. 25 and the coordinates of one end and the other end of the symmetry axis are subroutines from the main program. It is passed as an argument to Sub1, and the main program receives the difference image generated by the subroutine Sub1 from this Mth threshold image as a return value.

After executing step S21, step S915 is executed for the difference image as the return value. In step S915, it is determined whether or not the pixel values of all the pixels of the difference image are smaller than the Mth inspection threshold. When the determination result is Yes, it is confirmed that the inspection target region cannot be selected as indicated by the double border in FIG. 6C as described above for this difference image. In that case, the process proceeds to step S916.

In step S916, it is determined whether or not the number of symmetric axes stored in the register M has reached C, the number of symmetric axes. Register M is added one by one in step S907 of FIG. The original image corresponding to the added value of the register M is processed in steps S912 to S21 as described above to generate a difference image. In step S915, it is determined whether or not the pixel values of all the pixels of the difference image are smaller than the Mth inspection threshold. The determination in step S916 is equivalent to the determination as to whether or not the determination in step S915 has been made corresponding to all of the C symmetric axes of the image of the work to be inspected. When the determination result of Step S916 is No, since the determination of Step S915 is not made on the difference images corresponding to all the symmetry axes, the process jumps to Step S907 of Fig. 25 via the coupler described in the jump destination number 208. In this step S907, 1 is added to the value of the register M that stores the number of symmetry axes. Then, as in the description so far, the steps S908 to 26 of FIG. 25 are executed again while the value of M is increased by one. If the addition of 1 to the value of M is repeated, the determination result in step S916 is Yes. In that case, the process proceeds to step S917.

In step S917, it is judged whether or not the number of the good product images stored in the register I has reached A, which is the number of good product images. Register I is added one by one in step S902 of FIG. A good product image corresponding to the added register I value is processed by step S903 in Fig. 25 to step S916 in Fig. 26 as described above. Then, it is judged whether or not the pixel values of all the pixels of the difference image corresponding to all the symmetry axes of the good product image are smaller than the Mth inspection threshold. The determination in step S917 is nothing more than a determination as to whether or not all of the A good-quality images have been confirmed to have the pixel values of all the pixels of the difference image corresponding to all symmetry axes smaller than the Mth inspection threshold. When the determination result of Step S917 is No, it is not confirmed that the pixel values of all the pixels of the difference image corresponding to all the symmetry axes are smaller than the Mth inspection threshold value for all of the A good quality images. For this reason, it jumps to step S902 of FIG. 25 via the combiner in which the jump destination number 207 is described. In this step S902, 1 is added to the value of the register I that stores the number of good products images.

Then, in the state that the value of I is increased by 1, as in the previous description, the steps of Step S903 in FIG. 25 to Step S916 in FIG. 26 are executed again. If the addition of 1 to the value of I is repeated, the determination result of step S917 in Fig. 26 is Yes. In that case, step S9 is ended, and jump is made to step S10 in Fig. 14B via a combiner in which the jump destination number 101 shown at the bottom of Fig. 14A is described. Meanwhile, as described above, FIG. 14 (a) is a threshold setting mode, and FIG. 14 (b) is a test execution mode. Therefore, the threshold setting mode is ended in the step where step S9 is ended, and the inspection execution mode is started.

On the other hand, when the determination result in step S915 is No in Fig. 26, it means that the inspection target region is selected when the Mth inspection threshold is applied to the good image. Therefore, the process proceeds to step S918, and the cause of the selection of the region to be inspected is investigated, and countermeasures to eliminate the cause are taken. After taking countermeasures, the process jumps to step S103 of the known work imaging step (S1) of FIG. 15 via a combiner in which the destination number 201 is described. Then, the threshold setting mode shown in Fig. 14A is executed again. In this execution again, the Mth threshold image and the monochrome image for the Mth threshold value are selected in steps S802 to S805 in the threshold image generation process (S8) of FIG. 24 to set the Mth inspection threshold. In step S915 of FIG. 26 again, it is determined whether or not the pixel values of all the pixels of the difference image are smaller than the Mth inspection threshold. Until the determination result is Yes, the cause investigation and countermeasure in step S918 and the execution of the threshold setting mode are repeated again. If the determination result is Yes, as described above, until the determination of step S916 and the determination of step S917 together become Yes, the threshold checking process (S9) is executed. If the determination result of S917 is Yes, as described above, step S9 ends and the threshold setting mode ends. Then, the process jumps to step S10 in the test execution mode shown in Fig. 14B.

Next, the test execution mode will be described with reference to Figs. 14B and 27 to 31.

(5.9) Step S10

In FIG. 14B, step S10 is an imaging process of the inspection object, and imaging the inspection object. 27 shows a detailed flowchart of the inspection work imaging step (S10).

In step S1001 shown in Fig. 27, the work to be inspected is imaged. When step S1001 ends, step S10 ends, and the process proceeds to step S2 shown in Fig. 14B. In FIG. 14 (b), step S2 is a work extraction process to be inspected, and is the same process as step S2 shown in FIG. 14 (a). That is, the outermost edge of the work is searched and determined from the image of the work to be inspected using a template matching method, and an image of the work to be inspected is generated. When step S2 ends, the process proceeds to step S11 shown in Fig. 14B.

(5.10) Step S11

In FIG. 14B, step S11 is a process of generating a single object image to be inspected, and a single image is generated from the image of the work to be inspected generated in step S2. 28 shows a detailed flowchart of the inspection subject monochrome image generation process (S11).

In step S1101 shown in Fig. 28, 0 is stored in the register M, and the register M is initialized. Here, the register M is a register that stores the number of symmetry axes in the monochrome image generating process to be inspected after step S1102 and subsequent steps S12 and S13 shown in Fig. 14B. to be. The number of the axis of symmetry is set in step S4 (symmetry determining step) in the threshold setting mode of Fig. 14A, which is the previous step, for example. Specifically, in step S402 of the symmetry determination step (S4) shown in Fig. 18, the number of symmetry axes is stored in the register C. The value of this register C is the same value after being set in step S402.

Next, in step S1102, for all the pixels of the image of the work to be inspected, the image in which the pixel value is replaced with the maximum pixel value is stored in a register named the previous image. The operation of the image before this register will be described below.

The image processing algorithm of the present invention has already been described with reference to FIG. 5. In the description, the first straight line L1 and the second straight line L2 as two symmetrical axes shown in Fig. 5A are defined. The differential images corresponding to these symmetry axes are shown in Figs. 5 (b) and 5 (c). Then, the separated pixels selected from all the pixels in Figs. 5 (b) and 5 (c) which are the difference images, and the pixel values are doubled in Figs. 6 (a) and 6 (b). It is indicated by surrounding. After the separation pixels are selected, a common portion of the separation pixels shown in Figs. 6A and 6B is selected, and this common portion is used as an inspection target region. The concept of selecting the common portion of the interstitial pixels in FIG. 6 is to generate differential images corresponding to a plurality of symmetry axes at the same time, select the interstitial pixels from each differential image, and then select a common portion of each interstitial pixel. It is made. However, it is not practical to execute processing corresponding to a plurality of symmetry axes simultaneously, ie in parallel, using software. The reason is that in order to perform such parallel processing, steps to secure a large storage area for storing all pixel values of a plurality of difference images and to write and read pixel values to the storage area are necessary. to be. For this reason, in the defect inspection method using the image processing algorithm of the present invention, it is assumed that the following method is taken in order to execute the steps of Figs. 5 (b) to 6 (c) by software.

First, to the subroutine Sub1 shown in Fig. 20, the coordinates of the original image (a monochrome image generated from the image of the workpiece to be inspected) and one end (ONEM) and the other end (OTEM) of the first axis of symmetry are transmitted as arguments from the main program. The subroutine Sub1 generates a difference image corresponding to the first symmetry axis based on the original image, and returns it as a return value to the main program. The main program selects the separated pixels using the Mth inspection threshold set in the threshold setting mode for the received difference image. The position where the spaced pixels are arranged becomes an inspection target area corresponding to the first axis of symmetry.

Next, let M = 2 by adding 1 to the register M storing the number of symmetry axes. Accordingly, the argument that is then passed from the main program to the subroutine Sub1 is related to the second axis of symmetry. And similarly, in subroutine Sub1, a differential image corresponding to the second axis of symmetry is generated based on the original image. In the same way, in the main program, an inspection target area corresponding to the second axis of symmetry is selected.

In addition, a common part of the inspection target area corresponding to the second symmetric axis and the inspection target area corresponding to the first symmetric axis previously selected is taken as a new inspection target area. That is, the common portion of the two inspection target regions corresponding to the (M + 1) symmetry axis and the M symmetry axis is taken as a new inspection target region. Considering such a process, it is necessary to store the inspection target area corresponding to the M-th symmetry axis in a register while selecting an inspection target area corresponding to the (M + 1) -th axis of symmetry. It is said step S1102 that this register is secured in advance under the name of the previous image, and the image of the work to be inspected is stored therein as an initial value. Here, in step S1102, an image in which the pixel value is replaced with the maximum pixel value is stored for all the pixels of the image of the work to be inspected. The reason is described below.

As described above, the inspection target area is an area where defects may be present in an image of a work to be inspected, that is, an area where defect inspection is to be performed. For this reason, it is necessary to generate an image capable of clearly discriminating the inspection target area and other areas as a reference image when performing defect inspection on the image of the work to be inspected. Therefore, in the present invention, it is assumed that an inspection image is generated as the reference image. The inspection image is generated as an image in which pixels having a maximum pixel value are arranged in an inspection target area, and pixels having a minimum pixel value are arranged in an area other than the inspection target area. To cope with this, an image in which all areas in the image of the work to be inspected becomes the inspection subject area is first generated as an initial value of the previous image (hereinafter, the initial previous image). This is stored in the image before the register in step S1102 in Fig. 28. Then, as described above, in the subsequent step, an intersecting pixel is selected as a region to be inspected corresponding to the first axis of symmetry among the axis of symmetry (C) of the image of the workpiece to be inspected.

Next, a common portion of the inspection target area corresponding to the first symmetry axis and the inspection target area in the initial previous image is selected. Then, a new previous image having this common portion as an inspection target area is generated. Hereinafter, the creation of this new previous image is called updating of the previous image. Here, when updating the previous image corresponding to the first axis of symmetry, it is necessary to be able to select a common portion for the inspection target area corresponding to the first axis of symmetry and the inspection target area of the initial previous image. The condition is that all areas of the initial previous image become the inspection target area. For this reason, in step S1102, the initial previous image stored in the previous image of the register is set to the maximum pixel value so that all regions are the inspection target region. Then, after updating the previous image corresponding to the first axis of symmetry, the same process is performed by adding 1 to the register M storing the number of the axis of symmetry. This is repeated until the register M reaches the number C of said symmetry axes. Then, at the point of reaching C, the previous image that was updated last may be used as the inspection image.

In this way, after initializing the image before the register in step S1102 in Fig. 28, the process proceeds to step S1103. In step S1103, 1 is added to the value of the register M storing the number of symmetry axes. Next, in steps S1104 to S1108, a monochrome image to be inspected is generated and stored in the original register image. Among these steps, steps S1104 to S1107 are the same as steps S908 to S911 in the threshold value confirmation process (S9) shown in FIG. 25, respectively. Therefore, detailed description is omitted. In step S1108, if the monochromatic image to be inspected is stored in the register original image, step S11 ends, and the process proceeds to step S5 shown in Fig. 14B.

In FIG. 14B, step S5 is a difference calculation process, and step S6 following it is a difference image generation process. These steps use the subroutine Sub1 in Figs. 19 and 20 as described above to generate a difference image from the original image set in step S1108 in Fig. 28 above. Therefore, detailed description is omitted. When step S5 and step S6 are completed in Fig. 14B, the process proceeds to step S12.

(5.11) Step S12

In FIG. 14B, step S12 is an inspection area selection process, and using the difference image generated in step S6, an area candidate image having an area to be a candidate for the inspection target area is generated. 29 shows a detailed flowchart of the inspection region selection process (S12).

In step S1201 of FIG. 29, a pixel having a pixel value greater than the Mth inspection threshold value among all pixels of the difference image is selected as the inspection area pixel. This step is the same as selecting the distant pixels shown in Fig. 5 (d) (e) among all the pixels of the difference image. However, since this step S1201 is a process as preparation for selecting an inspection target area, the name of the selected pixel is referred to as an inspection area pixel. After selecting the inspection area pixel in step S1201, the process proceeds to step S1202.

Step S1202 is a previous process for generating an image that clearly identifies the above-described inspection target region and other regions, that is, an inspection image. First, the difference image is divided into pixels other than the inspection area pixel and the inspection area pixel. Next, an area designating pixel having a maximum pixel value is placed at the position of the inspection area pixel, and an non-exam pixel having a minimum pixel value is placed at a position other than the inspection area pixel, and an area candidate image is generated. The region candidate image generated by this process can clearly identify the region to be inspected and other regions by contrasting the pixel values of the inspection region pixels with the pixel values of the non-inspection pixels. If an area candidate image is generated in step S1202, step S12 ends, and the process proceeds to step S13 shown in Fig. 15B.

(5.12) Step S13

Step S13 is an inspection image generation process, and an inspection image is generated based on the area candidate image generated in step S12. 30 shows a detailed flowchart of the inspection image generation process (S13).

In step S1301 of FIG. 30, a common designated pixel that becomes a common part of the area designated pixel is selected for the previous image and the area candidate image. As described above, this process is for selecting a common portion of the generated inspection target region and the previous image whenever the inspection target region is generated corresponding to the axis of symmetry, and updating the previous image. After selecting a common designated pixel, that is, a common portion in step S1301, the process proceeds to step S1302.

In step S1302, a common designated pixel having a maximum pixel value is placed at a position of the common designated pixel, and an update candidate image is generated by setting a pixel value of a pixel other than the common designated pixel as a minimum pixel value. In this step S1302, the pixel value of a pixel other than the common designated pixel selected in step S1301 is set as the minimum pixel value, regardless of the position where the pixel having the maximum pixel value is disposed in the previous image. That is, even if the pixel having the maximum pixel value is disposed in the previous image, when the common designated pixel selected in step S1301 is not disposed at the location, the pixel value of the pixel disposed at the location is the minimum pixel value. Is updated. In this way, the update of the previous image is completed, and the image after the update becomes an update candidate image. After the update candidate image is generated in step S1302, the process proceeds to step S1303.

In step S1303, the update candidate image is stored in the image before the register. Next, in step S1304, it is determined whether the number of symmetric axes stored in the register M has reached C, the number of symmetric axes. The register M is added one by one in step S1103 of the inspection subject monochrome image generation process (S11) shown in FIG. The original image corresponding to the added register M value is processed by steps S1104 to S1108 in Fig. 28 and steps S1301 to S1303 in Figs. 29 and 30 as described above. And in step S1303 of Fig. 30, the previous image is updated. The determination in step S1304 is equivalent to the determination as to whether or not the previous image has been updated corresponding to all of the C symmetric axes of the monochromatic image to be inspected. If the result of this judgment is No, since the previous image corresponding to all the axes of symmetry is not updated, step S1103 in Fig. 28 (monochromatic monochrome image generation process (S11)) via the combiner to which the jump destination number 210 is described. To jump.

In this step S1103, 1 is added to the value of the register M that stores the number of symmetry axes. Then, similarly to the description so far, the steps described in Steps S1104 to S1108 in Fig. 28 and Steps S1301 to S1303 in Figs. 29 and 30 are again executed in a state where the value of M is increased by one. If the addition of 1 to the value of M is repeated, the judgment result of Step S1304 in Fig. 30 (inspection image generation step S13) is Yes. In that case, the process proceeds to step S1305.

In step S1305, the update candidate image is stored in a register named inspection image. Thus, generation of the inspection image is completed. Then, step S13 ends, and the process proceeds to step S14 shown in Fig. 14B.

(5.13) Step S14

In FIG. 14B, step S14 is an inspection execution step, and based on the inspection image generated in step S13, defect inspection is performed on the image of the work to be inspected. The detailed flowchart of the inspection execution process (S14) is shown in FIG. 31.

In step S1401 in FIG. 31, an image of a work to be inspected is taken out, and a defect inspection is performed on the area while referring to the inspection target area of the inspection image. Specifically, an area in which the maximum pixel value is arranged in the inspection image is an inspection target area in the inspection target work. Then, defect inspection is performed on this area. The method applied in the defect inspection may be any method known to the operator.

The defect inspection method using the image processing algorithm in the present invention described above is superior to the defect inspection method according to the prior art in the following points. First, there are very few items for the operator to visually judge the captured image, and the judgment criteria do not assume the skill of the operator.

In the defect inspection method according to the prior art, an operator sees an image of a work subject to defect inspection. Then, in the image, an inspection target area that is a target for defect inspection and an excluded area that is not target for defect inspection are identified. Then, the mutual relationship is checked with respect to the arrangement of each region, and the magnitude of the pixel values of the pixels constituting each region are confirmed. Next, in consideration of the interrelationship of the arrangement and the relationship between the pixel values and the size of the pixels, the inspection target area is selected by setting a threshold for selecting the inspection target area and the exclusion area. In addition, when a plurality of monochrome images are generated from color images, the operator visually observes each monochrome image. Then, by comparing the brightness between the respective areas, one single color image determined to be able to clearly identify the most defective is selected. Then, for the selected monochrome image, the threshold value is set and the region to be inspected is selected.

On the other hand, the image processing algorithm of the present invention and the defect inspection method using the same are used when the image of a work becomes line symmetric with respect to a specific straight line. First, with respect to the axis of symmetry, the difference between the pixel values of the pixels arranged at the target position is taken to generate a difference image. Next, among the pixel values of the pixels constituting the difference image, pixels having a pixel value larger than a preset threshold value are selected, and a location where the pixel is disposed is selected as an inspection target area. When there are a plurality of axes of symmetry, the common portion of the region to be examined corresponding to each axis of symmetry is the final region to be examined. In addition, when a plurality of monochrome images are generated from color images, a predetermined operation is performed on the pixel values for the pixels of the difference image generated from each monochrome image. Then, the calculation results of each difference image are compared, and a single monochrome image optimal for setting a threshold value is selected. Using the monochromatic image, the preset threshold value is set.

In this algorithm and the defect inspection method using the same, there is almost no process in which the operator makes any judgment by visually viewing the image of the work. For this reason, as in the defect inspection method according to the prior art, there is no need for skill for the operator, and the burden on the worker is reduced. In addition, since it is easy to perform defect inspection by automating the algorithm by software, the inspection speed is significantly improved compared to defect inspection according to the prior art, and it is not affected by the difference in operator skill.

WD1, WD2 ... bad work
MD1, MD2 ... display
D1, D2, Da, Db ... defect
WG1 ...
MG1 ... indication
B1 ... Background
F1, F2 ... circumference
L1 ... 1st straight line
L2 ... 2nd straight line
L3 ... 3rd straight line

Claims (8)

A monochromatic circular image having first and second regions that are line-symmetric with respect to the reference straight line is divided into the first and second regions by the reference straight line, and the first and second regions are divided into the first and second regions. A difference calculation step of calculating a difference pixel value that is a difference between the pixel values of the two original pixels for each pair of two original pixels arranged at a line symmetry position with respect to a reference straight line,
A difference image generation process in which a difference pixel having the difference pixel value is arranged to generate a difference image, using a pair of an original pixel at a first position in the first area and an original pixel at a second position in the second area A difference image generation process for generating the difference image by arranging difference pixels having the calculated difference pixel values at the first and second positions of the difference image.
A defect inspection method for performing defect inspection on an inspected object using an image processing method comprising:
A monochromatic image generated from a captured image obtained by imaging a plurality of inspected objects, which are known to be good products, is used as a first original image, and the image processing method is used from the plurality of first original images generated. A first step of generating a plurality of quality difference images,
A plurality of defective difference images are generated by using the image processing method from a plurality of generated second original images, using a monochromatic image generated from a captured image obtained by imaging a plurality of inspected objects having known defects as the second original image. The second process to do
As a threshold setting mode having a,
Among the difference pixels of the plurality of defective difference images, it is possible to select the difference pixels arranged at the same position in each defective difference image while the difference pixel values are separated from other difference pixel values by a predetermined value or more. A threshold setting mode for setting a threshold value of one inspection area capable of preventing the difference pixels arranged at the same position as the separation pixels from among the difference pixels of the plurality of quality difference images;
A monochrome image generated from a captured image obtained by imaging an object to be inspected as a defect inspection is used as a third original image, and a differential image is generated from the generated third original image using the image processing method, and the differential image and the inspection are performed. An inspection execution mode in which an inspection target area in the third original image is selected using an area threshold and a defect inspection is performed on the inspection target area
Defect inspection method comprising a. According to claim 1,
The reference straight line is a first straight line dividing the original image into the first region in the upper half and the second region in the lower half containing the same number of original pixels. Defect inspection method. According to claim 1,
The reference straight line is a second straight line dividing the original image into the first region in the left half and the second region in the right half containing the same number of original pixels. Defect inspection method. According to claim 1,
In the difference image generation step, a first difference image and a second difference image different from the first difference image are generated from the same original image as the difference image,
And an inspection region selection step of selecting an inspection target region in the original image using the first difference image and the second difference image. According to claim 1,
The captured image is a non-color image,
A defect inspection method characterized in that there is only one monochrome image generated from each captured image. According to claim 1,
The captured image is a color image,
Two or more monochrome images generated from each captured image,
In the first step of the threshold setting mode, each of the monochrome images is the first original image, and the plurality of good difference images are generated from the plurality of first original images generated,
In the second step of the threshold setting mode, each of the monochromatic images is used as the second original image, and the plurality of defective difference images are generated from the plurality of generated second original images,
In the threshold setting mode, a single monochrome image having the largest settable range of the inspection region threshold is selected, and the inspection region threshold is set for the selected monochrome image,
In the inspection execution mode, among two or more monochrome images generated from the captured images obtained by imaging the inspected object to be inspected for defects, a monochrome image of the same kind as the monochrome image selected in the threshold setting mode is selected to obtain the second image. With the three original images, the difference image is generated from the generated third original image, the inspection target area is selected using the difference image and the inspection area threshold, and defect inspection is performed on the inspection target area Running
Defect inspection method characterized in that. According to claim 1,
Defect inspection method characterized in that it further comprises a test object extraction process for extracting the object to be inspected for defects using a template matching method from the third original image.
delete

Images