JPH03175343A - Method for extracting flaw by inspection appearance - Google Patents

Abstract

PURPOSE:To accurately judge whether a flaw candidate is proper by setting a flaw candidate region and binarizing the density of each pixel in the region. CONSTITUTION:The image of an object to be inspected is taken by an image input apparatus 1 and the densities of respective images are inputted to a predetermined processing part 3 through an A/D conversion part 2 and the original image, a differentiated absolute image, a differential direction value image and an edge image are obtained to be respectively stored in an original image memory 4, a differentiated absolute value image memory 5, a differential direction value image memory 6 and an edge image memory 7. Next, a judging part 8 sets an inspection region A in the contour line l0 of the object to be inspected with respect to the edge image and limits the object to be inspected as the flaw candidate region B of the closed region surrounding the contour line l11 of a flaw candidate X11 within the region A. Standard density is set on the basis of the density of the inside of the contour line l0 of the object to be inspected outside the vicinity of the region B and a threshold value is set on the basis of the standard density and the density of each pixel in the region B is binarized on the basis of the magnitude relation to the threshold value. By this method, the pixel in the flaw candidate is certainly separated from other pixel to judge the quality of the flaw candidate.

Description

[Detailed description of the invention] [Industrial application field]

The present invention relates to a method for extracting defects by visual inspection, which uses image processing to inspect external defects such as chips, cracks, stains, and foreign objects on the surface of an object to be inspected.

[Conventional technology]

Conventionally, a spatial region including the object to be inspected is imaged using an image input means such as a TV camera, the density of each pixel of the original image is binarized using an appropriate threshold value, and defective parts are identified based on this binary image. Detection methods are known (for example, JP-A-62
(Refer to Japanese Patent Application No. 88946), that is, if there is a region whose value is inverted inside a region that can be considered as an inspection object in a binary image, it is regarded as a defective region.

[Problem to be solved by the invention]

In the conventional method described above, since a binary image is obtained for the entire screen of the original image, the density of the defective part and the density around the defective part may be included in the same area with respect to the threshold value (for example, (Both concentrations may exceed the threshold)
However, there was a problem in that the defective part could not be identified accurately. The present invention aims to solve the above-mentioned problems, and by setting a defect candidate area so as to surround the outline of the defect candidate to be determined as a defective part, the target for determination is limited,
Furthermore, by binarizing the density of each pixel in the defective five-complement area based on the density inside the contour line of the inspection object and separating pixels in the defect candidate from other pixels, it is possible to determine whether the defect candidate is good or not. The present invention aims to provide a method for extracting defects by visual inspection that enables accurate determination.

[Means to solve yA problem]

In order to achieve the above object, the methods of claims 1 and 2 provide a defect candidate area within a closed area surrounding the outline of the defect candidate to be determined as a defective part inside the outline of the object to be inspected. The reference density is set based on the density outside the vicinity of the defect candidate area and inside the outline of the object to be inspected, and the density of each pixel in the defect candidate area is set based on the above reference density. By performing binarization based on the magnitude relationship with the threshold value, pixels within the defect candidate are separated from other pixels. In the method of claim 3, the defect candidate area is defined as a six-sided area surrounded by a pair of sides running in the vertical and horizontal directions of the screen, and four sides intersecting each of the sides at an angle of 45°. It is set to a square shape. In the method of claim 4, reference value calculation lines are set outside the vicinity of the plurality of defect candidate areas and inside the outline of the object to be inspected, and the reference value calculation lines corresponding to different defect candidate areas intersect with each other. If not, calculate the reference density based on the density of each pixel on each reference value calculation line,
When reference value calculation lines corresponding to different defect candidate areas intersect, a new reference value calculation line is created that encloses the entire area surrounded by the intersecting reference value calculation lines and also includes a part of both reference value calculation lines. In the method of claim 5, the reference density is characterized based on the density of the pixel above the reference value calculation line. The density is set by adding a negative offset value to the density, and when the inside of the defect candidate area is a bright defect that is brighter than the surrounding area, the reference density is set by adding a positive offset value to the reference density. In the method of claim 6, the area occupied by the pixels in the defect candidate separated based on the magnitude relationship with the threshold value and the volume that is the sum of the density difference between the density of each pixel and the threshold value are determined. In the method according to claim 7, the defect candidate is characterized based on the area and the volume, the distance between the contour lines of a plurality of defect candidates is determined, and the defect candidates whose distances are within a predetermined distance are determined. In the method of claim 8, the area and volume are summed, and the quality of the defect candidate is characterized based on the summed area and volume. The reference density of each vertex is calculated based on the density of the pixel in The density of each pixel on the threshold plane, which is obtained by shifting the reference plane in the density direction, is set as the threshold at that position, and the density of each pixel in the defect candidate area at the position corresponding to the threshold is set as each threshold. It is binarized based on the value. In the method of claim 9, when the contour line of the defect candidate is closed, the inside of the contour line is set as the defect candidate area, and when the contour line is open, the area between the edges of the contour line is set as the defect candidate area. area.

[Effect]

According to the methods of claims 1 and 2, since the closed area surrounding the outline of the defect candidate that is the target of defect determination is set as the defect candidate area, the target of determination can be limited. In addition, the density of each pixel in the defect candidate area is binary-valued based on the magnitude relationship between the inspection pair S and the threshold value set based on the density outside the vicinity of the defect candidate area inside the outline of the object. Since the pixel in the defect candidate is separated from other pixels by using This increases the accuracy of determining the quality of defect candidates. In the method of claim 3, since the defect candidate area is set in a hexagonal shape, the defect candidate area can be set in a shape that almost follows the outline of the defect candidate, and the defect candidate area can be set in a simple shape. However, the area can be set to be approximately the same as the area of the defect candidate, and an increase in the amount of calculations due to wasteful calculations within the defect candidate area can be suppressed. In the method of claim 4, reference value calculation lines are set outside the vicinity of the plurality of defect candidate areas and inside the outline of the object to be inspected, and the reference value calculation lines corresponding to different defect candidate areas intersect with each other. If not, calculate the reference density based on the density of each pixel on each reference value calculation line,
When reference value calculation lines corresponding to different defect candidate areas intersect, a new reference value calculation line is created that encloses the entire area surrounded by the intersecting reference value calculation lines and also includes a part of both reference value calculation lines. Since the standard density is set based on the density of the pixel above this standard value calculation line, there are multiple defect candidates,
Even if the reference value calculation lines corresponding to each defect candidate intersect, part of the reference density calculation line does not pass through the defect candidate area, and the reference density can be set to an appropriate value. . In the method of claim 5, when the inside of the defect candidate area is darker than the surrounding area, the threshold value is set by adding a negative offset value to the reference density, and when the inside of the defect candidate area is brighter than the surrounding area, the threshold value is set. In the case of a bright defect, since a positive offset value is added to the reference density and set, the threshold value can be set to an appropriate value depending on the brightness of the defect candidate area. In the method of claim 6, the area occupied by the pixels in the defect candidate separated based on the magnitude relationship with the threshold value and the volume that is the sum of the density difference between the density of each pixel and the threshold value are determined. Since the quality of the defect candidate is determined based on the area and the volume, the accuracy of the quality determination is higher than when the quality of the defect candidate is determined based only on the area. In the method of claim 7, the distance between the contour lines of a plurality of defect candidates is determined, the areas and volumes of defect candidates within a predetermined distance are summed, and the defect is determined based on the total area and volume. Since the quality of the candidate is determined, multiple defect candidates that were originally one defect candidate but were separated by the process of calculating the contour line can be treated as one defect candidate. An improvement in accuracy can be expected. In the method of claim 8, the reference value calculation line is set in a polygonal shape, the reference density of each vertex is calculated based on the density of pixels near each vertex of the reference value calculation line, and the position and density of each pixel are calculated. 1) Set a reference plane that approximates the density gradient between each vertex, and calculate the density of each pixel on the threshold plane by shifting the reference plane in the density direction in the three-dimensional space. is the threshold value at each position, and the density of each pixel in the defect candidate area at the position corresponding to Threshold 11 is binarized based on each threshold value. uneven illumination (even if it occurs,
In the method of claim 9, when the contour line force τ of a defect candidate is closed, the inside of the contour line is regarded as a defect candidate area, and the determination accuracy is improved. When the contour line is open, the defect candidate area is defined as the area where the edges of the contour lines are connected with straight lines, so that the defect candidate area can be set very easily.

[Embodiment 1] In the present invention, as shown in FIG. 1, an image of an object to be inspected is captured by an image input device 1 such as a television camera, and the density of each pixel is converted into a digital signal by an analog-to-digital converter 2. By performing the following preprocessing in the preprocessing unit 3G, the original image obtained from the analog-to-digital conversion unit 2 is converted into a differential absolute value image, a differential direction value image, a differential direction value image, a differential direction value image, a differential direction value image, a differential direction value image, 7 images are obtained. That is, the original image P0 obtained by imaging a spatial region including the inspection object is a grayscale image, and as shown in FIG. It becomes. The process of extracting edges such as the contour line of the inspection object O from this grayscale image is based on the idea that ``edges have large density changes and correspond to partial numbers''. , and extracts the concentration by differentiating the concentration. The differential processing is performed by dividing the original image P0 into local parallel windows W of 3×3 pixels, as shown in FIG.
Eight pixels (eight neighbors) A to D and F to I surrounding the pixel E form a local parallel window W, and the vertical density change ΔV of the density of the pixels A to I within the local parallel window W and the horizontal The concentration change ΔH in the direction is determined by the following formula, ΔV=(A+B+C)−(G+H+ 1)ΔH=(A+
D+-G)-(C+F+ I) Furthermore, the differential absolute value 1e
17 and the differential direction value l e z are determined by the following equation. e 1t=(AV'+Δ)(2)+7 However, A~I
indicates the density of the corresponding pixel. As is clear from the above equation, the absolute differential value 1e represents the rate of change in density in the area near the pixel of interest in the original image, and the differential direction value le,
calculates the outline of inspection object 0, defect X1,
It is possible to extract areas with large concentration changes, such as the presence of foreign matter X2, and the direction of the changes. Although one defect X1 and one foreign substance X2 are shown in FIG. 2(a), the same applies even if there are many more defects and foreign substances. Here, the density of each pixel is expressed as the differential absolute value te1. The image expressed by is the differential absolute value image and the differential direction value l e t
The image expressed by is called a differential direction value image. Next, a thinning process is performed. The thinning process is performed by focusing on the fact that the larger the absolute differential value, the larger the density change. That is, the differential absolute value of each pixel is the differential absolute value of the surrounding pixels. By comparing and connecting those that are larger than the surrounding pixels, an edge with a width of one pixel is extracted. As shown in Fig. 4, if we consider a differential absolute value image P in which the position of each pixel is expressed by X-Y coordinates and the differential absolute value is taken on the Z axis, the thinning process calculates the ridge line on this curved surface. It corresponds to that. Through the processing up to this point, all edges are extracted regardless of the magnitude of the absolute differential value. The ridge line obtained at this stage includes unnecessary small peaks due to noise, etc., so as shown in Figure 5,
A threshold value SL is appropriately set, and only values equal to or higher than the threshold value SL are employed to remove noise components. The image obtained by this processing tends to become discontinuous lines when the contrast of the original image is insufficient or there is a lot of noise, so edge extension processing is performed. In edge extension processing, starting from the end point of the discontinuous line, the pixel of interest and its surrounding pixels are compared, and the evaluation function f(
Extend the line in the direction where eJ) is the largest, and continue this until it collides with the end point of another line. f (e,) = Here, eo is the differential data of the central pixel (corresponding to E of the local parallel window W), and e is the differential data of the adjacent pixel (corresponding to the 8 neighbors of the local parallel window W) and j=1.2.・・・・・・It is 8. Through the above processing, as shown in FIG. 2(b), an edge image P4 is obtained that traces a portion of the original image where the density change is large. In the edge image P, each line can be regarded as representing a contour line 2° to 12 degrees of the inspection object O1, defect Xl, foreign object X2, etc. Through the above preprocessing, four types of images are obtained: an original image, a differential absolute value image, a differential direction code image, and an edge image, and each image is stored in the original image memory 4, differential absolute value image memory 5, and differential direction value image. The image is stored in the memory 6 and the edge image memory 7. In the following explanation, the position of a pixel in each image is expressed by X-Y coordinates, and the density of a pixel in each image is f+(x, y), f2(
x, y), f*(x, y), and f<(x, y). The original image is a grayscale image, and the density is usually expressed in 8 bits, so the density at each pixel is a(=f+(x,y)
) is 0≦a≦255. In addition, the density of the differential absolute value image (i.e., the differential absolute value) b(= f z (X,
y )) is represented by 6 bits, for example, and 0≦b≦6
3, and the density of the differential direction value image (that is, the differential direction value) c(-f)(X,y)) is expressed by, for example, 16 directions, and 0≦C≦15. For edge images, since there is only the presence or absence of a line, the pixel that becomes a line is "1",
Other pixels are represented as "0". In other words, f<
The value range of (x, y) is (0, 11). In the following explanation, the term "density" refers to the density of white, and the larger the density value, the brighter it is. , as shown in FIG.
Defect candidate XI I, which is the target of determining whether there is a defect in the location
+X+2 exists, and one defect candidate X11 has a contour line 11. is open, and the outline e12 of the other defect candidate By performing a rask scan inside, f 4(X,
3/)=1 is detected and set as a flag point. Starting from this flag point, f4(X,y) for 8 neighborhoods
The contour me, , is traced while extracting the pixels that become -1. Here, when the contour line /II is open as shown in FIG. 8, the starting point P becomes the end point of the contour line ll However, in such a case, tracking is performed only in one direction, and the starting point P is the contour li1. ．． If it is not an end point, it will be tracked in both directions. By tracing the contour line 111 in this manner, the coordinates of all pixels on the contour w1111 are known, and the maximum and minimum values of the X coordinate and the X coordinate are determined, respectively. There are two ways to find the maximum and minimum values of the X-coordinate and Another method is to compare the size relationship of the coordinates. As described above, contour line 1. The maximum value X m * w + Y a... and the minimum value X...
...Y..., it is possible to set a rectangular defect candidate area B circumscribing the contour line I11, as shown in FIG. After setting the defective candidate area B in this manner, as shown in FIG. 10, a reference value calculation line L is set outside the defective candidate area B by several pixels (n pixels). The reference value calculation line L is the contour line 1 of the inspection object O.
Set inside 0. Next, the densities of all pixels on the reference value calculation line L are determined from the original image and stored in a buffer, and the densities in the buffer are used to set the reference density C9. The reference density CS can be either the average value, median, mode, or average value after discarding several pieces of data above and below the density in the buffer. Which value to use depends on the state of the image. decide. The offset value d is added to the reference density Cs thus obtained, and the threshold value S L 1 (=c g+d ) is set. The offset value d is a negative value when the defect candidate Xll is a dark defect with a lower density than the surroundings, and is a positive value when the defect candidate Xll is a bright defect with a higher density than the surroundings. In the case of dark defects, the concentration f, (x. y) is smaller than the threshold SLI (i.e., f + (x, y) SL
1 < 0) as a defective candidate point, and each time a defective candidate point is found, 1 is added to an area counter to obtain a value CA corresponding to the total area occupied by the defective candidate point. At the same time, the difference (SL], f+(x, y)) between the concentration f, (xy) at the defect candidate point and the threshold value SLI is calculated and added to the volume counter to determine the concentration at the defect candidate point and the threshold value SLI. A value Cv corresponding to the sum of the density differences between the two is calculated. Here, the value CA is called the area because it corresponds to the total area of defect candidates in the original image, and if we consider a three-dimensional space in which the density is regarded as depth in the /iir image, the value CV is the total volume of defect candidates in the original image. Since it corresponds to , we call it volume. By determining the area OA and volume Cv of the defect candidate point existing in the defect candidate region B as described above, it is possible to determine whether the defect candidate Xl+ is good or bad based on the area CA and volume Cv. That is, the primary judgment of acceptability is made based on the size of the area C, and when the area CA is within a predetermined range, the secondary judgment is made based on the volume CV. For example, the predetermined reference value a for the area CA is as follows:
, b (a>b), and a predetermined reference value C is also set for the volume CV. ■ CA>a ■ When either of the conditions a≧cA>b and Cv>c is satisfied, it is determined to be defective, and defect candidate X is determined to be defective. In addition, in other cases, the product is determined to be non-defective. When the defect candidate
The pixel that becomes the defect candidate point is defined as the defect candidate point. Further, the volume of each defect candidate point is determined as fl(X, 3/)-3LI. The conditions for determining pass/fail are the same as for dark defects. After determining the quality of defect candidate Xl+ as described above,
Continue rask scanning within the inspection area A and select the next defect candidate X1.
Similar processing is performed for No. 2 to determine whether it is good or bad. Here, after determining the quality of one defect candidate X, the next defect candidate -X+2 is searched for, but after setting the reference value calculation line L for one defect candidate X11, the next defect candidate
After searching for □ and setting the reference value calculation line L for all defect candidates, the quality determination may be performed. As described above, by setting the defect candidate area B that circumscribes the contour line 1e12 of the defect candidates X + +, X +□, the contour line 'll+112 of the defect candidates Xz, X, 2
Even if is not closed, defect candidates X + + , X 1□
It is possible to judge the quality of the product. In addition, a reference value calculation line L surrounding the defect candidate area B is set, and the defect candidate area B is determined based on the density of each pixel within the reference value calculation line B.
Since the density of pixels within is binarized, defect candidate X11
．． Even if unnecessary lines are continuous on the contour line 111+l+2 of X12, they will be ignored during the pass/fail judgment due to binarization, and the accuracy of the judgment will be increased. By the way, for setting the defect candidate area B, a differential binary image can also be used instead of using an edge image. A differential binary image is an image that is obtained by differentiating the original image (the method of differentiation is not limited to the above-mentioned method) and then binarizing it using an appropriate threshold, so that it becomes a figure whose border with the background is always closed. It has the property of becoming. Therefore, if the boundary line with the background is found within the inspection area A by rask scanning and this boundary line is tracked, processing is almost the same as tracing the contour line in an edge image. After tracing the boundary line with the background, the defect candidate area B can be set by finding the maximum and minimum values at the X coordinate and the X coordinate of the boundary line, respectively. The subsequent processing is the same as when using edge images. [Example 2] In Example 1, the contour line 1l12 of defect candidate Xz,
The defect candidate area B was set as a rectangle circumscribing the defect candidate area B, but if the defect candidate area B is a circumscribed rectangle
The area of defect candidate region B is considerably larger than the area surrounded by the contour line 111+l+2 of
, , X, contains many parts other than 2, and depending on the shape of the defect candidates X 1+ , X + □, there may be many unnecessary operations. Further, when a plurality of defect candidates X, . This embodiment compensates for the shortcomings of the first embodiment, and the first
Figure 1 (,) and 12th Q? 1(a), in order to make the defect candidate area B narrower than in the case of Example 1, the defect candidate area
A straight line tangent to the contour line 11 is set. As a method of setting such a straight line, for example, there is the following method. That is, first, as in Example 1, the X coordinate of the contour line 11 and the maximum value of the X coordinate X II @
It l Y * a 11 and the minimum value X s ln *
Y m r n is determined, and a rectangle circumscribing the outline 11 is set. For each vertex of this rectangle, the search starting point is moved pixel by pixel in the X direction (or Y direction) starting from the vertex, and the search is performed in a direction of 45 degrees from the search starting point. In other words, when the search starts from one search starting point and reaches the other side of the quadrangle without detecting any pixels on the contour line II, the search starting point is shifted by one pixel in the X direction (or in the Y direction). When a pixel on the contour line 11 is detected during the search, this search line is regarded as a straight line that intersects each side of the rectangle at 45° and is in contact with the contour line. By performing the above procedure for each vertex of the rectangle, it is possible to set a straight line that intersects each side of the rectangle at an angle of 45 degrees, and it is possible to set a hexagonal defect candidate area B. . As a result, compared to Example 1,
The defect candidate area B can be narrowed. Defect candidate area B
After setting, the same process as in Example 1 is performed, and as shown in FIG. 11(b) and FIG. 12(b), a reference value calculation line L is set outside the defect candidates @B. The quality of the defect candidate area B is determined. According to the method of this embodiment, since the area of the defect candidate region B is reduced, the amount of calculations required for pass/fail determination is reduced, and even when multiple defect candidates exist, only one defect candidate The probability that the corresponding defect candidate area B overlaps with other defect candidates is greatly reduced, and accurate pass/fail determination can be expected. In addition, the above procedure can set the rectangular defect candidate area B without using a calculation formula, and the contour line e
It can be used even if l is not closed. As an application of this embodiment, it is also possible to set the defect candidate area B to a polygon other than a quadrangle or a hexagon, but even if the defect candidate area B is set to a polygon larger than a hexagon, the practical effect will hardly change, and the calculation will be difficult. This is not a good idea as the amount will increase.

[Example 3] By the way, as shown in FIG. 13, a plurality of defect candidates X,
,,X,2 exists, one defect candidate,X,,,
Reference value calculation line for X, 2 L l l * L l 2
may pass over the other defect candidate X1llXI2. In this case, an appropriate value cannot be obtained as the reference density c3. Therefore, in this embodiment, it is necessary to determine whether or not the reference value calculation lines LL+2 intersect with each other (this can be done by comparing the coordinates of the vertices of both reference value calculation lines L + + + L l 2). If so, the reference value calculation line L is a rectangle obtained by extending the sides of the reference value calculation lines L, , L, and □ that do not intersect. The reference value calculation line L obtained in this way does not intersect with other reference value calculation lines, and the reference value calculation line L is based on the density of each pixel on this reference value calculation line L.
can be set to an appropriate value. The subsequent processing is the same as in the first embodiment.

[Embodiment 4] In each of the above embodiments, the defect candidate area B is replaced by the defect candidate Xz'
, X) z, but in this embodiment, the defect candidate area B is set as follows. That is, when the contour line 'll+l+2 is closed, the inside of the contour lines 111 and 112 is defined as defect candidate area B, and when the contour line 111+l12 is open, the end points are connected with a straight line S as shown in FIG. As a result, the contour line '11+'+2 is closed and the inside of the contour line 111+112 is designated as the defect candidate area B. Thereafter, the reference value calculation line L may be set in the same manner as in the first embodiment, and then the processing may be performed in the same manner as in the first embodiment. If defect candidate area B is set in this way, defect candidate area B
Compared to the case where is a polygon, the defect candidate area B contains almost no areas other than the defect candidates Xll and X12, and a reduction in the amount of calculation can be expected.

[Embodiment 5] In this embodiment, as shown in FIG. 15, a plurality of defect candidates X l l to X15 exist, and several defect candidates X l l to x1. An example of processing that is effective when the two are located relatively close to each other will be exemplified. That is, the contour lines l to Z1 of each defect candidate X, 1 to XI5
Based on the coordinates of each pixel on 5, the shortest distance between contour lines l to L, dl 2 + dl 3 + dl 4 +
......, d, and 5 are determined respectively. Here, the shortest path Md1J is a pair of defect candidates X It , X 1 h
Indicates the distance between. When the shortest path Nd1J obtained in this way is smaller than the preset threshold value dIIL (d, r < dst,), the pair of defect candidates X, , , X, are considered to be continuous. I regard it as such. Defect candidates X I L , X considered to be continuous
Regarding No. 14, the value obtained by adding the area CA and the volume CV, respectively, is used as the entire area and volume for the quality determination. For example, in FIG. 15, defect candidates X, 1 to X+4
are considered to be continuous, each defect candidate X,
If each defect candidate X, the surface ffcA, ~CA4 of l~x14, and the volume CVl~Cv4 of each defect candidate Xll~X14 are added together, and are continuous The area C10 and volume Cvo of the entire portion being considered are determined. That is, the following equation is obtained. CAo = CA l + CA 2 + CAff+
CA 4Cv a ” Cv + + CV 2 +
After Cv s + Cv 4, CA:=CA0,C
By performing the same processing as in the first embodiment with v:=Cvo, it is possible to determine the quality. Defect candidates considered to be independent X1. The quality determination is made in the same manner as in the first embodiment.

[Embodiment 6] In each of the above embodiments, the density of pixels in the defect candidate area B is set to 2.
The threshold value for converting into a value was determined using the reference density c8 based on the density of the pixels on the reference value calculation line L set to surround the outer periphery of the defect candidate area B, but the Depending on the position of the light source, etc., the illuminance may vary even within the range of the reference value calculation line L. If such unevenness exists, if the reference density C is fixed to a constant value, pixels that should originally have the same density will have different densities, which will affect the accuracy of the pass/fail determination. Therefore, in this embodiment, as shown in FIG. 16, there are reference value calculation areas at the corner portions of the reference value calculation line L.
D, and each reference value calculation area. Each reference value calculation area is determined from the density of each pixel in ~D, and the reference density of ~D is defined as the density of each vertex of the reference value calculation line L, and the xY coordinates of the pixel and the density are three-dimensional. A reference plane is set in space based on these four points. That is, a reference plane is set such that the sum of the distances from these four points is the smallest using the method of least squares or the like. Adding an offset value to this reference plane in the same manner as the reference density c6 to shift the reference plane in the three-dimensional space,
Set the threshold plane corresponding to the threshold. Generally, the unevenness of illuminance caused by a light source changes in a negative manner, so by setting a reference plane in the three-dimensional space, it is possible to deal with the tendency of change in unevenness, that is, the concentration gradient. By binarizing the density of each pixel on the original image that has the same XY coordinates as each point on the threshold plane set in the above manner according to the density difference, the influence of uneven illuminance can be removed. This allows accurate determination of pass/fail.

【Effect of the invention】

As described above, according to the methods of claims 1 and 2, the closed area surrounding the outline of the defect candidate to be determined as a defective part is set as the defect candidate area. Moreover, the density of each pixel in the defect candidate area can be determined based on the magnitude relationship with the threshold value set based on the density outside the vicinity of the defect candidate area inside the outline of the inspection object. Since the pixels in the defect candidate are binarized and separated from other pixels, it is possible to set an appropriate value as a threshold value, and it is possible to reliably separate pixels in the defect candidate from other pixels. This has the advantage of increasing the accuracy of determining the quality of defective candidates. In the method of claim 3, since the defect candidate area is set in a hexagonal shape, the defect candidate area can be set in a shape that almost follows the outline of the defect candidate, and the defect candidate area can be set in a simple shape. However, since the area can be set to be approximately the same as the area of the defect candidate, an increase in the amount of calculation due to wasteful calculation within the defect candidate area can be suppressed. In the method of claim 4, reference value calculation lines are set outside the vicinity of the plurality of defect candidate areas and inside the outline of the object to be inspected, and the reference value calculation lines corresponding to different defect candidate areas intersect with each other. If not, calculate the reference density based on the density of each pixel on each reference value calculation line,
When reference value calculation lines corresponding to different defect candidate areas intersect, a new reference value calculation line is created that encloses the entire area surrounded by the intersecting reference value calculation lines and also includes a part of both reference value calculation lines. Since the standard density is set based on the density of the pixel above this standard value calculation line, there are multiple defect candidates,
Even if the reference value calculation lines corresponding to each defect candidate intersect, part of the reference density calculation line does not pass through the defect candidate area, and the reference density can be set to an appropriate value. has. In the method of the claims, the threshold value is set by adding a negative offset value to the reference density when the inside of the defect candidate area is darker than the surrounding area, and the threshold value is set by adding a negative offset value to the above reference density. In the case of a bright defect, a positive offset value is added to the reference density to set the threshold value, which has the advantage that the threshold value can be set to an appropriate value depending on the brightness of the defect candidate area. In the method of claim 6, the area occupied by the pixels in the defect candidate separated based on the magnitude relationship with the threshold value and the volume that is the sum of the density difference between the density of each pixel and the threshold value are determined. Since the quality of the defect candidate is determined based on the area and the volume, the accuracy of the quality determination is improved compared to the case where the quality of the defect candidate is determined based only on the area. In the method of claim 7, the distance between the contour lines of a plurality of defect candidates is determined, the areas and volumes of defect candidates within a predetermined distance are summed, and the defect is determined based on the total area and volume. Since the quality of the candidate is determined, multiple defect candidates that were originally one defect candidate but were separated by the process of calculating the contour line can be treated as one defect candidate. This has the advantage that improved accuracy can be expected. In the method of claim 8, the reference value calculation line is set in a polygonal shape, the reference density of each vertex is calculated based on the density of pixels near each vertex of the reference value calculation line, and the position and density of each pixel are calculated. In the three-dimensional space described above, a reference plane that approximates the concentration gradient between each vertex is set, and the density of each pixel on the threshold plane is determined by shifting the reference plane in the density direction in the three-dimensional space described above. Since the density of each pixel in the defect candidate area at the position corresponding to the above threshold value is binarized based on each threshold value, the illuminance of the defect candidate area can be adjusted depending on the position of the light source, etc. Even if unevenness occurs, the influence of uneven illuminance can be removed, improving determination accuracy. In the method of claim 9, when the contour line of the defect candidate is closed, the inside of the contour line is set as the defect candidate area, and when the contour line is open, the area between the edges of the contour line is set as the defect candidate area. Since the area is defined as a region, there is an effect that setting of the defect candidate area becomes very easy.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a processing circuit corresponding to the method of Embodiment 1 of the present invention, FIG. is an explanatory diagram showing locally parallel windows in the same as above, FIG. 4 is an explanatory diagram showing a differential absolute value image in the same as above, and FIG. An explanatory diagram, FIG. 6 is an operation explanatory diagram showing the relationship between the outline of the inspection object and the inspection area in the same as above, and FIG. 7 is an operation explanatory diagram showing the search operation for defect candidates within the inspection area in the same as above Figure 8 is an operation explanatory diagram showing the relationship between the contour line of the defect candidate and the flag point in the same as above;
9 is an explanatory diagram of the operation showing the relationship between the defect candidate and the defect candidate area in the same as above, FIG. 10 is an explanatory diagram of the operation showing the relationship between the defect candidate area and the reference value calculation line in the same as the above,
1 and 12 are operation explanatory diagrams showing a method for setting a defect candidate area in Example 2 of the present invention, and FIG. 13 is an operation explanatory diagram showing a method for setting a reference value calculation line in Example 3 of the present invention, FIG. 14 is an operational explanatory diagram showing a defect candidate area setting method in Embodiment 4 of the present invention, FIG. 15 is an operational explanatory diagram showing Embodiment 5 of the present invention, and FIG. FIG. A... Inspection area, B... Defect candidate area, lo~12
．． IZI2...contour line, L +' L + +,
L+□...Reference value calculation line, 0...Inspection object, Xl...Defect, X2...Foreign object, XX12...Defect candidate.

Claims (9)

[Claims] (1) After capturing an image of a spatial region including the object to be inspected using the image input means, an edge image including the outline of the object to be inspected is set based on the change in density of each pixel of the original image obtained by the image input means. However, in a defect extraction method using visual inspection that extracts defects such as chips, cracks, dirt, etc. that exist inside the outline of the object to be inspected based on the original image and the edge image, The closed area surrounding the outline of the defect candidate that is the target of defect determination is set as the defect candidate area, and the reference density is determined based on the density outside the vicinity of the defect candidate area and inside the outline of the object to be inspected. and set
A feature is that pixels in the defect candidate area are separated from other pixels by binarizing the density of each pixel in the defect candidate area based on the magnitude relationship with a threshold value set based on the reference density. Defect extraction method using visual inspection. (2) After capturing an image of a spatial region including the object to be inspected by the image input means, a differential binary image containing the outline of the object to be inspected is based on the change in density of each pixel of the original image obtained by the image input means. In the defect extraction method using visual inspection, which extracts defects such as chips, cracks, and stains that exist inside the outline of the object to be inspected based on the original image and the differential binary image, A closed area surrounding the outline of the defect candidate to be determined as a defect is set as the defect candidate area inside the outline, and the density is set inside the outline of the object to be inspected outside the vicinity of the defect candidate area. By setting a reference density based on the above, and binarizing the density of each pixel in the defect candidate area based on the magnitude relationship with the threshold value set based on the reference density, pixels in the defect candidate area and other pixels are A defect extraction method by visual inspection characterized by separating pixels. (3) The defect candidate area is set as an octagon surrounded by a pair of sides running in the vertical and horizontal directions of the screen, and four sides intersecting each of the above sides at an angle of 45 degrees. 3. The defect extraction method by visual inspection according to claim 1 or 2, characterized in that: (4) Set reference value calculation lines outside the vicinity of multiple defect candidate areas and inside the outline of the object to be inspected, and when the reference value calculation lines corresponding to different defect candidate areas do not intersect, each A reference density is calculated based on the density of each pixel on the reference value calculation line, and when reference value calculation lines corresponding to different defect candidate areas intersect, the entire area surrounded by the intersecting reference value calculation lines is enclosed. According to claim 1, a new reference value calculation line that includes part of both reference value calculation lines is set, and the reference density is set based on the density of the pixel above this reference value calculation line. Or the defect extraction method by visual inspection according to claim 2. (5) The above threshold value is set by adding a negative offset value to the above reference density for a dark defect where the inside of the defect candidate area is darker than the surrounding area, and for a bright defect where the inside of the defect candidate area is brighter than the surrounding area. 3. The defect extraction method by visual inspection according to claim 1, wherein the reference density is sometimes set by adding a positive offset value to the reference density. (6) Calculate the area occupied by the pixels in the defect candidate separated based on the magnitude relationship with the threshold value and the volume that is the sum of the density difference between the density of each pixel and the above threshold value, and 3. The defect extraction method by visual inspection according to claim 1, wherein the quality of the defect candidate is determined based on the volume. (7) Find the distance between the contour lines of multiple defect candidates, total the area and volume of the defect candidates whose distance is within a predetermined distance, and evaluate the quality of the defect candidate based on the total area and volume. Claim 6 characterized in that it determines
Defect extraction method using visual inspection as described. (8) Set the reference value calculation line in a polygonal shape, calculate the reference density of each vertex based on the density of pixels near each vertex of the reference value calculation line, and create a three-dimensional space between the position and density of each pixel. , a reference plane that approximates the density gradient between each vertex is set, and the density of each pixel on the threshold plane, which is obtained by shifting the reference plane in the density direction in the three-dimensional space, is set as the threshold value at that position. , wherein the density of each pixel in the defect candidate area at a position corresponding to the threshold is binarized based on each threshold. Extraction method. (9) When the contour line of the defect candidate is closed, the inside of the contour line is the defect candidate area, and when the contour line is open,
3. The method for extracting defects by visual inspection according to claim 1 or 2, characterized in that the area defined by connecting the edges of the contour lines with straight lines is defined as a defect candidate area.