JP2016004274A - Defect detection method for product - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a defect detection method for a product that can accurately and quickly detect a defect in a product by using a non-linear alignment.SOLUTION: The method compares a reference image obtained by photographing a defect-free normal product with an inspection image obtained by photographing an inspection target product which is the same type as the normal product by using a non-linear alignment, and detects the presence or absence of a defect in the inspection target product from a difference image between the images. A region of interest (ROI) in the non-linear alignment is set only when the ROI has a certain number or more of pixel patterns.

Description

  The present invention relates to a product defect detection method, and more particularly to a method that can be suitably used to detect product defects using X-rays.

  When a defect inside the product, for example, a defect such as a shrinkage nest generated inside a cast product is detected, detection using an internally transmissive X-ray is performed. In this case, since the product to be inspected usually has a three-dimensional shape, the transmission image of the product is distorted depending on the X-ray source position. On the other hand, defect detection occurred in the difference image by calculating the difference image between the reference image obtained by photographing the normal product and the inspection image obtained by photographing the same type of inspection target product as the normal product. A graphic region is determined as a defect. However, since the transmitted X-ray image has locally different degrees of distortion, the inspection image cannot be matched with the reference image by general linear coordinate transformation, and accurate defect detection cannot be performed. Therefore, it is conceivable to use so-called non-linear alignment shown in Patent Document 1 for such defect detection.

JP-A-2005-176402

  However, the subject of the invention described in Patent Document 1 is an X-ray image of a human chest and is not intended for general industrial products. When applying non-linear alignment to such a defect detection of industrial products, various countermeasures are required such as the necessity of speeding up by reducing the processing load of the calculation as well as the accuracy.

  Accordingly, the present invention has been made in view of such a demand, and an object of the present invention is to provide a product defect detection method capable of accurately and quickly detecting product defects using nonlinear alignment.

  In order to achieve the above object, in the first invention, a reference image obtained by photographing a defect-free normal product and an inspection image obtained by photographing a product to be inspected of the same type as the normal product are obtained. In the method of detecting the presence / absence of a defect in the inspection target product from the difference image by comparison by non-linear alignment, the region of interest (ROI) in non-linear alignment is set only when the pixel pattern of the ROI exceeds a certain level. Like that.

  According to the first aspect of the present invention, since the ROI is not set in the region having a flat region in which the non-linear alignment is difficult and the change of the pixel pattern is poor, the burden of the cross-correlation calculation between the ROIs is not set. Is reduced and the image alignment speed is improved. In addition, since the region where the ROI is not set is a region where the change of the pixel pattern is scarce, even if the ROI is not set in this region, the accuracy of image alignment is not lowered and accurate defect detection is possible.

  In the second aspect of the invention, the size of the target ROI and the search ROI at the time of alignment search in the nonlinear alignment is relatively increased, and the image of the target ROI and the search ROI at the time of alignment calculation is relatively reduced. To.

  According to the second aspect of the invention, since the size of the ROI at the time of registration search is relatively large, it is less affected by local pixel pattern similarity, and the accuracy of registration is improved. Further, since the ROI image at the time of alignment calculation is relatively reduced, the amount of calculation of cross-correlation calculation can be reduced, and rapid defect detection is possible.

  In the third aspect of the invention, in the nonlinear alignment, the inspection image is obtained separately for each product from a photographed image obtained by photographing a plurality of products simultaneously.

  According to the third aspect of the present invention, it is possible to obtain inspection images separated for each product for products that are stored in the tray in different postures and photographed at the same time, so that it is possible to detect defects for each product. Become.

  In the fourth aspect of the invention, in the nonlinear alignment, the inspection image is divided into partial areas, and the difference image is obtained for each partial area.

  According to the fourth aspect of the invention, the inspection image is divided into, for example, partial areas having the same luminance, and a difference image is obtained for each partial area. Defect detection is possible.

  As described above, according to the product defect detection method of the present invention, product defects can be detected accurately and quickly using nonlinear alignment.

It is a block diagram which shows the structure of the defect detection apparatus which enforces the method of this invention. It is a flowchart of the process performed by client PC. It is a flowchart of a defect detection process. It is a conceptual diagram of mask pattern generation of area division. It is a figure which shows a test | inspection image, and a figure which shows the example of arrangement | positioning of template ROI. It is a conceptual diagram which shows the setting size of ROI. It is a conceptual diagram which shows the reduction process of ROI. It is a figure which shows the reference | standard image of a product, and the figure which shows the area | region division for mask image preparation.

  FIG. 1 shows the configuration of a defect detection apparatus for carrying out the method of the present invention. The defect detection apparatus includes an X-ray irradiation apparatus 1, an X-ray image storage apparatus 2, and a determination apparatus 3. The X-ray irradiation apparatus 1 irradiates the product with X-rays and fixes the transmission photographed image of the product on the imaging plate 11. The imaging plate 11 on which the transmission image is fixed is mounted on a reading device 21 that constitutes the X-ray image storage device 2, read as a DICOM digital image, and stored in the X-ray image storage server 22. The storage server 22 is provided with a monitor 23 for image confirmation.

  The DICOM digital image is sent to the client PC 31 constituting the determination apparatus 3 and converted into a RAW image. A monitor 32, a database server 33, and an external storage device (NAS) 34 are connected to the client PC 31. Hereinafter, processing executed by the client PC 31 will be described with reference to the flowchart of FIG.

  In step 101 of FIG. 2, a RAW image file is read, and in step 102, area division is performed. A method of area division will be described with reference to FIG. Usually, a plurality of products to be inspected are housed in trays in different postures and photographed simultaneously. Therefore, it is necessary to obtain an image in which products are individually separated. For example, when two products W1 and W2 are photographed as shown in FIG. 4 (1), mask patterns M1 and M2 corresponding to the products W1 and W2 are generated as shown in FIG. 4 (2). Then, these mask patterns M1 and M2 are overlaid to obtain a RAW image separated for each single product as shown in FIG.

  In step 103 of FIG. 2, a nonlinear alignment image is generated for the RAW image of each separated product. When generating a non-linear alignment image, for example, a template region of interest (ROI) is set in the RAW image (inspection image) of each product to be inspected, and a normal product of the same type as this is imaged in advance. A search region of interest (ROI) is set in the obtained RAW image (reference image).

  In this case, the search ROI is set larger than the template ROI, and the cross-correlation value of both regions is calculated while moving the template ROI within the search ROI. Then, nonlinear alignment is performed assuming that the regions having the maximum cross-correlation values correspond to each other. This operation is a known method described in Patent Document 1 described above.

In this embodiment, the accuracy and speed of nonlinear alignment are improved in view of the characteristics of industrial products. That is, in such an image of an industrial product, there is generally an area that is flat and has little change in pixel pattern. In such a region, since it is difficult to obtain a peak of the cross-correlation value between the template ROI and the search ROI, the alignment is not performed well. Therefore, the distribution V of the luminance value I of the pixel is calculated by the following equation (1), and the template is assumed to have a pixel pattern effective for alignment only when the luminance distribution V of the pixel in the ROI is higher than a certain value. Place the ROI.
V = Σ | I (x, y) −Iv | ** 2 (1)
Here, I (x, y) is the luminance of each pixel in the template ROI, Iv is the average luminance value of the pixels in the template ROI, and ** 2 is the square (the same applies hereinafter).

  An example of the arrangement of such template ROI is shown in FIG. FIG. 5A is an inspection image, and the pixel pattern hardly changes in the inner peripheral region W11 of the circular portion and the partial region W12 of the outward protruding portion. Therefore, as shown in FIG. 5B, the template ROI 4 is not arranged in such areas W11 and W12. Thereby, while maintaining the alignment accuracy, the burden of the cross-correlation calculation between the template ROI and the search ROI is reduced, and the alignment speed is improved.

  By the way, since there are many regular changes in the case of industrial products, if the size of the template ROI or the search ROI is reduced, the cross-correlation values in a plurality of regions show similar values, and the alignment accuracy decreases. End up. Therefore, in this embodiment, the pixel size of the template ROI4 is set to, for example, an X pixel (X = 45 pixel) angle as shown in FIG. 6, and the pixel size at the time of abnormality detection for the conventional human chest is 23 pixels. The search range is expanded by increasing the pixel size of the search ROI 5 from the conventional 121 pixel angle to the Y pixel (Y = 501 pixel) angle, for example.

  However, if it does in this way, the calculation processing amount of a cross correlation process will increase to 3430 times [= (121/23) ** 2 * (501/45) ** 2]. Therefore, in the present embodiment, when the cross-correlation calculation for nonlinear alignment is performed, the images of the template ROI and the search ROI are represented by Z (for example, 1/5) as shown in FIG. The calculation processing amount of the cross-correlation processing is reduced by 1/625 [= (1/5) ** 2 × (1/5) ** 2]. Thereby, even if cross-correlation is taken in a wide range, the increase in the amount of calculation processing of cross-correlation calculation can be suppressed to about 5.5 times (= 3430 × 1/625) as a whole.

  In step 104 of FIG. 2, defect detection is performed using the nonlinear alignment image generated by the above processing. Details of the defect detection processing will be described below with reference to the flowchart of FIG. The non-linear alignment image is read in step 201 in FIG. 3, the mask image is overlaid on the non-linear alignment image in step 202, and the inspection region is extracted in step 203. Here, creation of a mask image for extracting an inspection region will be described. With respect to the reference image of the product shown in FIG. 8 (1), an area having the same luminance based on the shade (luminance) of the image is divided by a closed contour line to divide the area (FIG. 8 (2)). . In FIG. 8 (2), each part with numbers / letters is divided. The mask image is obtained by masking areas other than the area to be inspected, and is created by the number of divided areas.

  The extraction of the inspection area in step 203 is to extract only a desired area as an inspection area by overlaying the mask image on the nonlinear alignment image. In step 204, it is determined whether the inspection area includes a cast character part or a burr part. The reason for this is that it is difficult to distinguish from a defect on a non-linear alignment difference image, which will be described later, because the cast character portion includes a bright portion and a dark portion. In addition, the burr portion is difficult to align with the nonlinear alignment image.

  Therefore, in step 204, if the inspection area includes a cast character part or a burr part, the process proceeds to step 205, in which an inspection image that is a RAW image is read instead of a nonlinear alignment image, and the mask image is converted into an inspection image. The inspection area including the cast character part and the burr part is extracted by overlaying and the process proceeds to the image processing in step 207 and the subsequent steps. Here, a visual image may be used instead of the inspection image which is an X-ray transmission image.

  When the inspection area does not include a cast character portion or the like in step 204, the process proceeds to step 207, and a difference image (nonlinearity) from the reference image is obtained for each inspection area in the nonlinear alignment image extracted in step 203. A registration difference image) is calculated and generated.

  In step 208, the background luminance of each inspection area is corrected, and in step 209, the image of the inspection area is binarized with a predetermined threshold value. In step 210, the number of pixels of the graphic remaining in the binarized image is analyzed. If the number of pixels is equal to or greater than a predetermined value, the graphic is determined as a defect in step 211.

  In this case, a scale member in which a plurality of identical circular figures are drawn at different densities in several stages is placed in a tray in which the product is accommodated, and a RAW image of the scale member taken simultaneously with each product is obtained. The circle figure on the image binarized in step 209 is compared with the circle figure in the RAW image, and the determination value of the number of pixels in step 211 is calibrated based on the dimensional change rate at this time. In this case, it is possible to more accurately determine the presence / absence of a defect.

DESCRIPTION OF SYMBOLS 1 ... X-ray irradiation apparatus, 2 ... X-ray image storage apparatus, 3 ... Determination apparatus, 31 ... Client PC.

Claims (4)

A reference image obtained by photographing a defect-free normal product and an inspection image obtained by photographing the inspection target product of the same type as the normal product are compared by non-linear alignment, and the inspection target product is obtained from the difference image. In the method for detecting the presence or absence of a defect, a region of interest (ROI) in nonlinear alignment is set only when the pixel pattern of the ROI is greater than a certain value. 2. The size of the target ROI and the search ROI at the time of alignment search in the non-linear alignment is relatively increased, and the image of the target ROI and the search ROI at the time of alignment calculation is relatively reduced. Defect detection method for the described product. 3. The product defect detection method according to claim 1, wherein in the nonlinear alignment, the inspection image is obtained separately for each product from a photographed image obtained by photographing a plurality of products simultaneously. The product defect detection method according to claim 1, wherein, in the non-linear alignment, the inspection image is divided into partial areas, and the difference image is obtained for each partial area.