JP6546125B2 - Defect inspection device - Google Patents

Description

  The present invention relates to a defect inspection apparatus.

  As a method of inspecting a surface defect such as a flaw or unevenness on an outer surface of an automobile body, a method of photographing an inspection object and performing image processing is known. For example, according to Patent Document 1, a photographed image of an inspection object is divided into a plurality of areas, and the luminance values of pixels included in each area are averaged to create a divided image. Then, a defect candidate area is extracted by comparing the luminance value of each area with the luminance value of the area present in the periphery.

JP, 2013-185862, A

  Conventional inspection devices can detect local flaws and irregularities. However, in the conventional inspection apparatus, for example, it is an image of whether it is a defect or not with respect to a slight unevenness of about 10 μm (loose undulation or minute surface distortion) in about 10 cm square existing on the panel surface of 1 meter square. It was difficult to detect using the process. Therefore, conventionally, inspections relying on visual judgment of experienced inspectors have been performed.

  The present invention has been made to solve such a problem, and provides a defect detection device capable of detecting slight surface distortion while suppressing the influence of individual differences existing within the variation range of non-defective products. The purpose is to

  A defect inspection apparatus according to the present invention comprises: an imaging unit configured to capture an inspection image of an inspection object on which an irradiation pattern of predetermined light is projected; a processing unit configured to process the inspection image captured by the imaging unit; The processing means arranges the inspection image and a plurality of non-defective images obtained by photographing the site identical to the inspection image by the imaging means to create a first alignment image, and the first alignment image Is divided into a plurality of areas, the average luminance value of each of the areas is calculated, the average luminance values of the adjacent areas are compared, and the difference between the average luminance values exceeds the reference value. A score is added, and the inspection image and the plurality of non-defective images of the first aligned image are rearranged to create a second aligned image, and the division of the area is performed on the second aligned image Repeat up to the point addition And, based on the accumulated the number, and outputs whether include defects in the inspection image.

  According to the present invention, by performing image processing on a plurality of aligned images, compared to a region having surface distortion within the variation range of non-defective products, to a region having surface distortion exceeding the variation range of non-defective products More points will be added. Therefore, it is possible to provide a defect detection device capable of detecting a slight surface distortion by suppressing the influence of individual differences existing in the non-defective product variation range.

FIG. 1 is a diagram showing a configuration of an apparatus for realizing a defect detection method according to a first embodiment. 5 is a flowchart of image processing according to the first embodiment. 5 is an illustration of an image according to the first embodiment. 5 is an illustration of an image according to the first embodiment. FIG. 2 is a diagram of a power spectrum according to the first embodiment. 5 is a flowchart of the defect inspection apparatus according to the first embodiment. 5 is an illustration of an aligned image according to a first embodiment. 5 is an illustration of an aligned image divided into grids according to the first embodiment. 6 is an illustration of experimental results according to the first embodiment. 6 is an illustration of experimental results according to the first embodiment. 7 is a flowchart of a defect inspection apparatus according to a second embodiment. 15 is an illustration of an aligned image according to the second embodiment in which the grid size is changed.

Embodiment 1
First, a first embodiment according to the present invention will be described.

  FIG. 1 is a diagram showing the configuration of an apparatus for realizing the defect detection method according to the first embodiment, that is, a defect inspection apparatus. The defect inspection apparatus 10 includes an imaging unit and a processing unit. The photographing means comprises an imaging unit 11, an illumination unit 13 and a lens 14. The processing means is a PC (personal computer) 12. The imaging unit 11 shoots the inspection object 15 and generates an image. As the imaging unit 11, a camera or the like provided with an imaging device such as a CCD (charge-coupled device) can be used. When photographing with the imaging unit 11, the illumination unit 13 including the lens 14 irradiates the inspection target 15. In the first embodiment, the illumination unit 13 including the lens 14 projects an irradiation pattern of predetermined light on the surface of the inspection object 15. The irradiation pattern is, for example, a stripe pattern. The image taken by the imaging unit 11 is taken into the PC 12. The PC 12 executes defect candidate extraction by a process described later based on the captured image.

  When the defect candidate extraction is performed, the image processing described below is performed on the captured image. FIG. 2 is a flowchart showing an example of image processing performed on a captured image.

  First, the imaging unit 11 captures a fringe projection image (step S101). Next, the PC 12 takes in a fringe projection image and performs local dispersion processing (step S102). By performing the local dispersion processing, the PC 12 reduces unevenness of the image generated at the time of shooting, such as shading, from the fringe projection image. FIG. 3 is an example of a fringe projection image before the local dispersion processing is performed. Since the image 21 has a shadow at the upper left, the contrast in the image is uneven. FIG. 4 is an example of a fringe projection image after local dispersion processing. The image 21 has reduced unevenness due to shadows, and the contrast in the image is uniform. In addition, since the local distributed processing is a general method, the detailed description here is omitted.

  Next, the PC 12 Fourier-transforms the image (step S103). By Fourier transform, the PC 12 decomposes the image into frequency components. Next, the PC 12 performs filtering to cut low frequency and high frequency components from the image data subjected to Fourier transform (step S104). Furthermore, the PC 12 performs inverse Fourier transform on the image data that has been subjected to the filtering (step S105).

  FIG. 5 is a diagram for explaining the principle of filtering in step S104, which shows the power spectrum after Fourier transform. In the image after the local dispersion processing, a portion with a relatively small distortion on the surface of the inspection object becomes a curve such as a curve 26 after Fourier transform. On the other hand, in a portion where the distortion of the surface of the inspection object is relatively large, the distortion of the image may deteriorate the contrast of the image. In such a case, for example, the curve after Fourier transform looks like the curve 25. Therefore, the PC 12 cuts lower frequency components than the frequency fa of the curves 25 and 26 and higher frequency components than the frequency fb. Then, the contrast of the image after the inverse Fourier transform (step S105) is improved, and false detection is suppressed in the image processing to be performed later.

  As described above, when the PC 12 performs the process illustrated in FIG. 2, it is possible to improve the unevenness and the contrast of the image, and to suppress the possibility of erroneous detection when extracting a defect candidate in the subsequent image processing. The PC 12 stores the image whose image quality has been adjusted by the above process (step S106). The PC 12 stores a plurality of the image quality-adjusted images, and proceeds to a process for extracting defect candidates.

  Next, the defect candidate extraction process performed by the PC 12 will be described. FIG. 6 is a flowchart of image processing according to the first embodiment.

  First, the PC 12 aligns the image quality-adjusted images to generate an aligned image (step S210). FIG. 7 shows the aligned image 30 generated by the PC 12. The alignment image 30 is formed by aligning and combining four images in the X row and four images in the Y row. In the first embodiment, the alignment image 30 is illustrated as 4 × 4 sheets, but the number of sheets to be aligned may be two or more, and the number of X columns and Y columns may not be the same. Here, the defect inspection apparatus 10 inspects whether or not there is a defect in the inspection image 31. The 15 images other than the inspection image 31 are 15 types of non-defective images obtained by photographing the same portions as the inspection image. Each non-defective image has surface distortion and the like within the design tolerance of the product. In the example of FIG. 7, the inspection image 31 is arranged at the position of X3Y3 of the alignment image. The PC 12 randomly performs this arrangement.

  Next, the PC 12 divides the aligned image 30 into predetermined areas, and averages the luminance values in each area (step S220). In the first embodiment, the area is called a grid. FIG. 8 is an example of a state in which the alignment image 30 is divided into grids. In FIG. 8, the PC 12 assigns p grids in the x column and q grids in the y column, and let it be assumed that the i-th grid in the x-th column and the j-th grid in the y-column are G (i, j). . The PC 12 calculates the average of the luminance values for the pixels in each grid. Further, in the first embodiment, the average luminance value in the grid G (i, j) is referred to as B (i, j).

  Next, the PC 12 determines the start grid for comparing the average luminance value of each grid (step S230). For example, in the aligned image 30 shown in FIG. 8, the PC 12 can set G (1, 1) as the start grid.

  Next, the PC 12 compares the average luminance value of the inspection grid with the average luminance value of the adjacent grid (step S240). For example, when G (i, j) is an inspection grid, adjacent grids are G (i-1, j), G (i + 1, j), G (i, j-1), G (i, j + 1). ). Thus, the PC 12 compares B (i, j) with B (i-1, j). Similarly, the PC 12 compares B (i, j) with B (i + 1, j), B (i, j-1), and B (i, j + 1), respectively.

  Next, based on the comparison result, the PC 12 determines whether the difference is larger than the reference value (step S250). For example, as a result of comparing B (i, j) and B (i-1, j), if the difference is larger than the reference value (step S250: Yes), the PC 12 adds points (step S260). ). And as a result of comparing B (i, j) and B (i + 1, j), when the difference is not larger than a reference value (Step S250: No), PC12 does not add a score. As described above, the PC 12 compares the average luminance values of all adjacent grids in the inspection grid, and determines the comparison result and adds points. The PC 12 then proceeds to the next step.

  Next, the PC 12 determines whether or not all grids have been inspected (step S270). If the PC 12 has performed processing for all the grids in the aligned image 30 (step S270: Yes), the PC 12 proceeds to the next step S290. If the PC 12 has not processed all the grids in the aligned image 30 (step S270: No), the PC 12 performs a process of changing the inspection grid (step S280). Then, the PC 12 returns to step S240 again, and compares the average luminance value with the adjacent grid. For example, when the inspection grid is G (i, j), the PC 12 changes the inspection grid to G (i + 1, j).

  If the PC 12 has performed processing for all the grids in the aligned image 30 (step S270: Yes), the PC 12 determines whether a predetermined number of inspections have been performed (step S290). If the number of examinations has not reached the predetermined number (Step S290: No), the PC 12 randomly sorts 16 image quality-adjusted images in the aligned image 30 (S292). Then, the PC 12 generates a new aligned image based on the randomly rearranged array (step S210). Then, the PC 12 compares the average luminance value in the grid with the adjacent grid again, and repeats the process of adding the score. Then, when the number of examinations has reached a predetermined number (step S290: Yes), the PC 12 outputs the examination result (step S295).

  The inspection result is obtained by accumulating the addition of the points performed in step S260. The PC 12 outputs the number of points accumulated in each pixel of the inspection image 31. Since the addition of points is performed on all the pixels in the aligned image, as illustrated in FIGS. 9 and 10, not only the points accumulated in each pixel of the inspection image 31 but also the result output. , For all pixels in the aligned image.

  The number of examinations is two or more and can be set by the user. If the number of inspections is increased, the combination of the inspection image and the adjacent non-defective image will increase. That is, the defect inspection apparatus 10 can improve the determination accuracy as to whether or not it is a defect image by increasing the comparison with a non-defective image having individual surface distortion within the design tolerance.

  9 and 10 show the results of experiments conducted by the inventors according to the first embodiment. In the example of FIG. 9 and FIG. 10, an aligned image is generated by aligning seven images in the X column and seven images in the Y column. The examination image is arranged at X3Y5. When the comparison of the average luminance value based on this Embodiment 1 is performed, the position of a test | inspection image is arranged at random. However, here, in order to display the result in an easy-to-understand manner, the inspection image after image processing is fixed to X3Y5 and described.

  FIG. 9 shows the result of generating the alignment image once and comparing the brightness values of the respective grids. According to the result shown in FIG. 9, the area 50 and the area 51 can be mentioned as an area having a high score, and a part which is considered to be a defect can be widely seen. That is, the PC 12 adds a high score to the non-defective image in addition to the original inspection target image. Therefore, this result indicates that the non-defective image is highly likely to contain a defect.

  On the other hand, the result shown in FIG. 10 is the result of generating the aligned image a plurality of times and repeatedly comparing the brightness values of the respective grids. According to the result shown in FIG. 10, the PC 12 adds a high score to the area 60. This result indicates that the image located at X3Y5 contains a defect.

  According to the first embodiment described above, it is possible to provide a defect inspection apparatus that detects slight surface distortion while suppressing the influence of individual differences existing in the variation range of non-defective products.

Embodiment 2
Next, Embodiment 2 according to the present invention will be described. First, the configuration of the apparatus according to the second embodiment is the same as the configuration of the defect inspection apparatus 10 illustrated in FIG. 1 in the first embodiment.

  Next, the process according to the second embodiment will be described. FIG. 11 is a flowchart for explaining the process according to the second embodiment. In the second embodiment, the same steps as those in the first embodiment are denoted by the same reference numerals, and the redundant description will be omitted.

  First, as in the first embodiment, the PC 12 generates an aligned image (step S210). Next, the PC 12 determines the grid size (step S215). In the second embodiment, for example, a grid shown in FIG. 8 is adopted as an initial value. The description related to FIG. 8 is the same as that of the first embodiment and thus will be omitted.

  Next, the PC 12 performs the processing from step S220 to step S280 as in the first embodiment. Since these processes are the same as in the first embodiment, the description will be omitted.

  Next, in step S270, when the luminance value comparison for all grids is completed (step S270: Yes), the PC 12 determines whether or not all grid sizes have been checked (step S275). If all grid sizes have not been checked (step S275: No), the PC 12 changes the grid size (step S276). FIG. 12 is an illustration of the aligned image 30 whose grid size has been changed by the PC 12. In FIG. 12, the PC 12 assigns m grids in x columns and n grids in y columns. The grid size in FIG. 12 is larger than that in FIG. The setting of the grid size is appropriately determined by the user according to the size of the inspection image and the size of the defect to be inspected. In FIG. 12, the inspection image is arranged at X2Y3.

  After changing the grid size in step S276, the PC 12 returns to step S220 for calculating the average luminance value in the grid, and again compares the average luminance values for all adjacent grids, and determines the comparison result and scores Perform the addition of When all grid sizes have been checked (step S275: Yes), the PC 12 proceeds to the next step.

  After that, as in the first embodiment, the PC 12 compares the average luminance values of the grids and adds the points until the number of inspections reaches a predetermined number. However, in the second embodiment, the PC 12 also changes the grid size. When the number of inspections reaches a predetermined number (Step S290: Yes), the PC 12 outputs the inspection result (Step S295).

  In the second embodiment, the processing loop for changing the grid size and the processing loop for randomly generating the aligned image may be interchanged.

  According to the second embodiment described above, the defect inspection apparatus 10 can detect various defects by changing the grid size and performing inspection. That is, by setting the grid size to be small, the defect inspection apparatus 10 can detect local flaws and irregularities. Then, by setting the grid size to a large size, the defect inspection apparatus 10 can also detect slight surface distortion that has been difficult to detect.

  In the embodiment described above, the grid assignment and the comparison of the average luminance value in the grid may assign the grid to the entire range of the aligned image as described above, but another A method may be used. For example, the PC 12 may set a range smaller than the aligned image and larger than the grid size, compare the average luminance value for each range, and perform score addition. For example, this range is called a block. By adopting such a method, although the repetition frequency of processing is increased, processing can be performed for each block size. Therefore, the PC 12 can execute calculations with a small memory size.

  The present invention is not limited to the above-described embodiment, and in addition to the contents described above, the present invention can be appropriately modified without departing from the scope of the present invention. For example, the image quality adjustment of the photographed image described with reference to FIG. 2 is not limited to the example described here, and these processes may be omitted if the detection accuracy can be ensured. Moreover, about addition of the score demonstrated in FIG.6 S260, addition of the score which used weighting may be employ | adopted and you may employ | adopt multiplication. Various calculations may be performed as long as the calculation is for the purpose of detecting a defect candidate.

10 defect inspection apparatus 11 imaging unit 12 PC
13 illumination unit 14 lens 15 inspection object 30 aligned image 31 inspection image G (i, j) area B (i, j) average luminance value

Claims (2)

An imaging unit configured to capture an inspection image of an inspection object on which an irradiation pattern of a predetermined light is projected;
Processing means for performing image processing on the inspection image photographed by the photographing means;
Equipped with
The processing means
The first inspection image is created by arranging the inspection image and a plurality of non-defective images in which the same portion as the inspection image is captured by the imaging unit.
Dividing the first aligned image into a plurality of regions;
Calculating the average luminance value of the area,
Comparing the average luminance values of the regions adjacent to each other, and adding a score to the regions when the difference between the average luminance values exceeds a reference value;
The inspection image and the plurality of non-defective images of the first alignment image are rearranged to create a second alignment image,
The division of the area to the addition of the score is repeated for the second aligned image,
Based on the accumulated points, it is output whether or not a defect is included in the inspection image.
Defect inspection device. The processing means
After adding the points,
Changing the size of the area and repeating the calculation of the average luminance value to the addition of the points;
Furthermore, the inspection image and the plurality of non-defective images of the first aligned image are rearranged to create a second aligned image.
The defect inspection apparatus according to claim 1.