JP2005069887A - Defect inspection method and apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a surface inspection method and an apparatus capable of high-speed and efficient processing. <P>SOLUTION: An input image is divided into overlapping rectangles (S60). The insides of the rectangles are binarized by a prescribed luminance threshold (S70). The number of prescribed ("0" or "1") pixels among binarized pixels is computed (S80). In the case that the number of pixels exceeds a prescribed pixel number threshold, it is determined that the rectangles are flawed rectangles (S90). It is possible to specify the locations of flaws on the basis of a flawed rectangle mask image (S110) constituted of the flawed rectangles. By further taking the AND of a normal binarized image (S130) and the flawed rectangle mask image (S110), it is possible to acquire a binarized image in which only flaws are detected. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a defect inspection method and apparatus, and more particularly to a method and apparatus for inspecting defects such as wrinkles on an object surface.

  Conventionally, for example, a surface defect is inspected using a steel plate, an aluminum plate, or the like as an inspection target. As the method, there is one that obtains an image of the surface of the object to be inspected, extracts a part that is shining brighter than others or a part that is darker on the contrary, and determines whether it is a defect or not, according to this inspection method, A wrinkle can be specified by signal processing the acquired surface image.

  In such a surface wrinkle detection method, for each pixel of the obtained surface image, it is determined whether or not it is greater than or equal to a predetermined luminance threshold value, and there is a possibility that pixels that are equal to or greater than (or less than) the threshold value constitute a wrinkle. Extract as a certain pixel. Thereafter, labeling processing (pixel connection processing) is performed on the extracted pixels. When the labeling result area exceeds a predetermined threshold, it is determined to be 疵.

  For example, when there are slight scratches on the mirror-finished surface, this method is also effective, but if there is a ground pattern or dirt on the inspection object, these are detected as wrinkles. It will be. In addition, even if pixels result in noise, it is difficult to increase the processing speed by making them all subject to labeling processing, and there is a problem that a high-performance processing device is required for processing. there were.

  Although not aimed at improving the processing speed, a conventional technique is known in which a plurality of thresholds are provided in accordance with the type of wrinkles in order to increase detection accuracy (see Patent Document 1).

Japanese Patent No. 2890801

  In view of the above problems, an object of the present invention is to provide a surface inspection method and apparatus capable of high-speed and efficient processing.

  In order to achieve the above object, the present invention first divides an input image into overlapping areas, binarizes the area with a predetermined luminance threshold value, and selects a predetermined (“0” or “1” among the binarized pixels. The number of pixels “)” is calculated, and when the number of pixels exceeds a predetermined pixel number threshold, the region is defined as a defective region.

  According to this configuration, since the threshold is set for the number of pixels in the area divided, it is possible to remove noise (for example, the brightness is low and the size is small) that would otherwise be picked up and pick up only the bag. it can. In addition, when tiles are placed without overlapping areas, even if they are divided at the boundary of the area and the count value of the number of pixels is halved and it is not determined that there is a defect, counting is performed by overlapping the areas. Since it is possible to prevent a decrease in value, it is possible to prevent omission of detection of soot.

  The present invention obtains the defect area image by obtaining the defect area, obtains a binarized image obtained by binarizing the input image as a whole, and obtains a logical product of the binarized image and the defect area image. A defect inspection is performed based on the obtained detected defect binarized image.

  Since the present invention employs such a configuration, noise can be removed before the labeling process, so that the burden of subsequent labeling can be greatly reduced.

  In the present invention, the shape of the area into which the input image is divided can be rectangular, the area size and the overlap amount are provided in multiple sets, and the input image can be divided into overlapping areas using the multiple sets. it can.

  According to this configuration, by changing the size of the area according to the wrinkle to be detected, it is possible to detect, for example, only a vertically long flaw. Compared with the detection of the vertical fold using a conventional spatial filter, processing can be performed with a much lower load.

  Furthermore, a plurality of sets of luminance threshold values and pixel number threshold values are provided, and a defective region can be obtained by each set, and these can be combined to form a defective region image. A set of a large luminance threshold and a small pixel number threshold, and a set of a small luminance threshold and a large pixel number threshold may be combined.

  In this way, by having a plurality of threshold values, it is possible to apply a threshold value corresponding to wrinkles. For example, it is possible to pick up a small bag with a large change in luminance, and pick up a large bag with a small change in luminance.

  Furthermore, the input image can be an image with normalized brightness.

  According to this configuration, even when it is difficult to binarize the screen with a uniform threshold due to uneven illumination, a normalized brightness threshold is obtained by obtaining a normalized image and processing based on the normalized image. Can be processed.

  First, an outline of an embodiment of the present invention will be described.

  FIG. 1 shows a binarized image obtained by a conventional method, in which an image of an inspection surface is binarized using a luminance threshold value.疵 K and noise part N appear by binarization. Conventionally, the labeling process is performed based on this image, but the labeling process is also performed on the noise portion N, and the load is large.

  In the present embodiment, the original image (image before binarization) is divided into overlapping rectangles, and it is determined whether or not there are wrinkles in units of rectangles. This is shown in FIG. FIG. 2 shows two rectangles covering the ridge K, but actually the entire screen is covered with such rectangles. The rectangle B has a width of l and a length of m, and the width x overlaps by m / 2 and the length y by l / 2. In this way, it is divided into overlapping rectangles, and it is determined whether or not there are wrinkles for each rectangle. That is, first, binarization is performed using a luminance threshold, and the number of pixels exceeding the luminance threshold is counted for each rectangle. If the result exceeds the pixel threshold given to the rectangle, the rectangle is To do. The reason why the rectangles are overlapped is to prevent the wrinkles from being missed at the boundaries of the rectangles.

  FIG. 3 shows a rectangle with a ridge taken out. In FIG. 3, only the rectangle covering the ridge portion 3 remains as a ridged rectangle. Speaking of the noise part N in FIG. 1, even if the luminance threshold value is satisfied, the pixel threshold value given to the rectangle is not exceeded, so it does not appear in FIG. Eventually, the noise portion was removed, and only a creased rectangle indicating the presence of 疵 K was obtained. Here, if the AND of FIG. 1 and FIG. 3 is taken, a binarized image in which the noise portion N is removed from FIG. The outline of the present embodiment is as described above. Since noise is removed in the previous stage of the labeling process, the processing efficiency is improved.

  Hereinafter, an outline of an embodiment of the present invention will be described with reference to FIGS.

  4 to 6 are flowcharts showing the processing steps of the present invention, and FIGS. 7 to 12 show images obtained in the processing steps.

  As shown in FIG. 4, first, in step S10, a surface image to be inspected is input from a detector such as a CCD camera. Normally, an 8-bit 256-gradation image is used.

  In the next steps S20 to S50, the luminance of the image is normalized, and a normalized image is obtained in step S50. For this purpose, in step S20, each pixel of the input image is multiplied by a luminance reference value (128 in this example), and in step S30, the input image is averaged by a low-pass filter, To obtain the average brightness. For example, a moving average of 32 × 32 around each pixel is taken. In step S40, the result of step S20 is divided by the result of step S30. In this way, the luminance is normalized, and a normalized image is obtained in step S50. In addition, the meaning of the multiplication process of step S20 is for preserving the gradation represented by the integer (0-255).

  FIG. 7 shows an image after the luminance normalization processing in step S50. A light-colored portion P1 having a large area at the upper left, a small but black portion P2 at the lower right, and a set P3 of a number of thin and small portions in the central vertical direction can be seen.

  If the input image is normalized in this way, even if the screen cannot be binarized with a uniform threshold if the input image from the detector remains as it is due to uneven illumination, etc. Processing can be performed using a uniform threshold.

  In the next step S60 (FIG. 5), the normalized image is divided into overlapping rectangular regions. In this example, the same rectangle as that shown in FIG. 2 is used. As described above, the rectangle B has the horizontal l and the vertical m, and the horizontal x overlaps with m / 2 and the vertical y with l / 2.

  The rectangular area specifies the location of the wrinkles, and the shape and size can be determined according to the expected wrinkles. In general, the shape and size should be a little larger than the expected wrinkles and can cover the wrinkles. For example, in the case of a vertically long ridge and a part thereof is dotted, a rectangular region that is short in the vertical direction even if the vertically long rectangular region can accurately recognize the vertically long cocoon as a whole If it is divided by, the dotted line portion is judged as noise and discarded, and there is a risk of mistaking the exact shape of the eyelid. However, as a matter of course, the rectangular area does not have to be set larger than 疵. The rectangular area may normally be about 16 × 32 pixels, and for a vertically long wrinkle, for example, 16 × 64 pixels may be taken.

  Since the shape and size of the rectangular area can be changed according to the wrinkle to be detected, for example, only a vertically long wrinkle can be detected using a vertically long rectangular area. In the past, when a vertically long basket was picked up, processing was performed using a spatial filter. However, the spatial filter has a large calculation amount, and processing is burdened. When the rectangular area of the present invention is used, processing can be performed with a much lower load.

  In this example, the overlap portion is m / 2 in the horizontal direction and 1/2 in the vertical direction, but the size of the overlap portion is arbitrary and can be determined as appropriate. The meaning of overlapping the rectangles is that if the rectangles are arranged without being overlapped, the scratches may be removed if there are scratches on the boundary of the rectangles. In the present invention, the number of pixels that satisfy a predetermined condition is counted, and a rectangle that is equal to or larger than the predetermined number of pixels is processed to be a rectangle with a wrinkle. At that time, the wrinkles are divided at a rectangular boundary. This is because if the number of pixels of the eyelid is halved, the count may be lost and may not be recognized as an eyelid. Therefore, in order to prevent counting omission, it is better to make the overlapping portions x and y large and ½ or more. However, if the overlapping portion is enlarged, so many rectangles are required, so that processing time is required. In this example, it is set equal to 1/2 as described above. Further, although the image segmentation area is a rectangle, it may be another polygon such as a triangle or a parallelogram. In any case, it may be an area that can be divided so as to cover the inspection image.

  In the next steps S70 to S90 and S71 to S91, a wrinkled rectangle is specified using a set of different luminance threshold values and pixel number threshold values, respectively.

  In step S70, the rectangular area is binarized using the first luminance threshold. In step S80, the number of pixels exceeding the luminance threshold is counted for each rectangular area. In step S90, the counted result is the first value. A rectangle that exceeds the threshold value for the number of pixels is defined as a rectangle with a wrinkle. Here, in the set of the luminance threshold value and the pixel number threshold value, for example, the first luminance threshold value may be close to 128 (average value), and the first pixel number threshold value may be increased. In the first set, an area that is thinly spread can be determined as a wrinkle.

  An example of processing the image of FIG. 7 with such a set of threshold values is shown in FIG. Since the brightness threshold value is close to the average value and the pixel number threshold value is large, the light and wide area P1 is detected by being covered with the rectangle B1. The rectangles corresponding to the other portions P2 and P3 having a smaller area do not appear.

  Similarly, in step S71, the rectangular area is binarized using the second luminance threshold value, and in step S81, the number of pixels exceeding the luminance threshold value is counted for each rectangular area. In step S91, the counted result is A rectangle that exceeds the second pixel count threshold is defined as a rectangle with a wrinkle. Here, the second luminance threshold is far from 128 (that is, close to 0 or 255), and the second pixel number threshold is small. In the second set, a small black area can be determined as a wrinkle.

  The result is shown in FIG. Although the pixel value is small, since the luminance threshold is far from the average value, only the rectangle B2 covering the portion P2 that appears to be a black dot appears. The other portions P1 and P3 are excluded because they do not satisfy the luminance threshold.

  In this example, two sets are used. However, the present invention is not limited to this. The number of sets may be three or more, and the combination of the luminance threshold value and the pixel number threshold value can be variously changed according to the trap to be detected. . Moreover, only a single set can be used without using a plurality of sets.

  In the next step S100, a rectangle mask image having a wrinkle is obtained in step 110 by combining the rectangles determined to be wrinkled rectangles in step S90 or S91 (usually by a logical sum operation).

  FIG. 10 shows the wrinkled rectangle obtained by combining the logical sums of the wrinkled rectangular masks B1 and B2 (FIGS. 7 and 8) obtained using the first and second threshold sets according to step S100. It is a mask.

  If the purpose is only to determine whether or not there is a wrinkle, the location of the wrinkle is specified by the obtained rectangular mask with the wrinkle, and the processing may be ended here. Although the combination is performed by taking a logical sum here, other logical operations may be performed depending on the set of the luminance threshold value and the pixel number threshold value.

  To determine the exact shape of the ridge, proceed to the next step.

  Therefore, in step S120, normal binarization using a predetermined luminance threshold value (the threshold value used in step S70 or another value may be used) is performed using the normalized image in step S50. In step S130, a normal binarized image is obtained.

  FIG. 11 is obtained by normal binarization processing of the normalized image of FIG. 7, and all pixels P1, P2, and P3 that are blacker than a certain threshold are detected. Conventionally, labeling processing or the like has been performed based on this, but performing a labeling on these many pixels puts a heavy load on the processing apparatus.

  Next, in step S140 (FIG. 6), the logical product (AND) of the normal binarized image obtained in step S130 and the wrinkled rectangular image obtained in step S110 is calculated.

  FIG. 12 is an AND of the rectangular mask with wrinkles (FIG. 10) and the normal binarized image (FIG. 11), and is the wrinkle detection binarized image in step S150. As shown in FIG. 12, even if the color is light, a large haze P1 and a small color P2 that is dark even if it is small can detect the size and shape, and can be detected as a noise P3. Has been removed. In this way, the background noise and the like appearing in the normal binarized image disappear, and only the desired wrinkles are detected.

  Thereafter, image processing is performed to determine the shape and size of the ridge. That is, a labeling process is performed in step S160, and a feature extraction process is performed in step S170 to determine the shape and size of the heel.

  In general, in order to know not only the presence or absence of wrinkles but also the shape and size of the wrinkles, a labeling process is required. In order to obtain a binary image that accurately represents the outline of the eyelid, it is better to binarize with a low luminance threshold, but if the threshold is lowered, background noise will also be picked up and labeling processing for a large number of pixels The result is that the load of the labeling process increases. In the present embodiment, a wrinkled region is identified, and noise is removed before the labeling process using a wrinkled rectangular mask. Therefore, the burden of labeling can be greatly reduced.

  In step S180, the harmfulness level of the heel is determined based on the length, width, and perimeter of the heel, and the heel image is displayed in step S190.

  According to the present invention, prior to image processing with a large load that must be performed for each pixel such as labeling processing, a rectangular mask with wrinkles is obtained, thereby removing non- wrinkle parts such as background noise and performing image processing. Since the target to be performed can be limited, the efficiency of wrinkle detection processing can be greatly increased.

  Next, an outline of an embodiment of the device of the present invention will be described with reference to FIG.

  For example, an inspection image is taken into the image input unit 1 from an inspection target such as a steel plate ST via an image input device such as a CCD camera C. The inspection image is sent from the image input unit 1 to the luminance normalization unit 2 to obtain a normalized image. The normalized image is sent to the rectangular binarizing unit 3 and the normal binarizing unit 4 to generate a wrinkled rectangular image and a normal binarized image, respectively, and the arithmetic unit 5 ANDs both images. , A wrinkle detection binary image is obtained. Then, based on the wrinkle detection binary image, labeling and feature extraction are performed in the image processing unit, the shape and size of the wrinkle are obtained, and the harmfulness level of the wrinkle is determined. In the image display unit, the haze image after the image processing is displayed.

  In this example, a steel plate has been described as an example. However, what is to be inspected is not limited to a steel plate, and may be, for example, a slab, or may be applied to, for example, a fluorescent image or an ultrasonic flaw detection image obtained by magnetic particle flaw detection. In addition, the present invention can be applied as long as an image to be inspected can be obtained.

It is the schematic which shows the conventional binarized image. It is explanatory drawing which shows the rectangular area of one Embodiment of this invention. It is explanatory drawing which shows the rectangle with a crease of one Embodiment of this invention. It is a figure which shows the flow (the 1) of one Embodiment of this invention. It is a figure which shows the flow (the 2) of 1 embodiment of this invention. It is a figure which shows the flow (the 3) of one Embodiment of this invention. It is a figure which shows the normalized image of one Embodiment of this invention. It is a figure which shows the wrinkled rectangle obtained by the 1st threshold value set of 1 embodiment of this invention. It is a figure which shows the wrinkled rectangle obtained by the 2nd threshold value set of 1 embodiment of this invention. It is a figure which shows the combination of the rectangle with a hook shown in FIG.8 and FIG.9. It is a figure which shows the normal binarized image in one Embodiment of this invention. It is a figure which shows the wrinkle detection binarized image in one Embodiment of this invention. It is a figure which shows the apparatus structure of one Embodiment of this invention.

Explanation of symbols

K ... 疵 B, B1, B2 ... Rectangular area N ... Noise part

Claims (14)

Dividing the input image into overlapping regions;
Binarizing an area with a predetermined luminance threshold;
Calculating the number of pixels of a predetermined value among the binarized pixels in the region;
When the number of pixels exceeds a predetermined pixel number threshold, the region as a defective region;
Including defect inspection method. Dividing the input image into overlapping regions;
Binarizing an area with a predetermined luminance threshold;
Calculating the number of pixels of a predetermined value among the binarized pixels in the region;
When the number of pixels exceeds a predetermined pixel number threshold, obtaining a defective region image with the region as a defective region;
Binarizing the input image with a predetermined luminance threshold to obtain a binarized image;
A defect inspection method comprising: calculating a logical product of the binarized image and the defect area image.   The defect inspection method according to claim 1, wherein the shape of the region is a rectangle.   The step of dividing the input image into overlapping regions has a plurality of sets of sizes and overlapping amounts of the regions, and the input image is divided into overlapping regions in each set. Defect inspection method.   The luminance threshold value and the pixel number threshold value are composed of a plurality of sets in which a plurality of threshold values are combined, and the defective region image is obtained by combining defective regions obtained by using each set. The defect inspection method according to any one of the above items.   The defect inspection method according to claim 5, wherein the combination of the plurality of threshold values includes a combination of decreasing the pixel number threshold value when the luminance threshold value is large and increasing the pixel number threshold value when the luminance threshold value is small. .   The defect inspection method according to claim 1, wherein the input image is an image whose luminance is normalized. An image input unit for inputting an image to be inspected;
The input image is divided into overlapping regions, the inside of the region is binarized with a predetermined luminance threshold, the number of pixels having a predetermined value among the binarized pixels in the region is calculated, and the number of pixels is predetermined A region binarization unit that generates a defect region image, using the region as a defect region when the pixel number threshold of is exceeded,
A defect inspection apparatus comprising an image display unit for displaying a defect area image. An image input unit for inputting an image to be inspected;
The input image is divided into overlapping regions, the inside of the region is binarized with a predetermined luminance threshold, the number of pixels having a predetermined value among the binarized pixels in the region is calculated, and the number of pixels is predetermined When the pixel number threshold is exceeded, the region as a defect region, a region binarization unit that generates a defect region image,
A normal binarization unit that binarizes the input image with a predetermined luminance threshold to obtain a binarized image;
A calculation unit for obtaining a defect inspection binarized image from the defect region image and the normal binarization unit,
An image display unit for displaying the defect inspection binarized image;
A defect inspection apparatus comprising:   The defect inspection apparatus according to claim 8 or 9, wherein the shape of the region is a rectangle.   The defect inspection apparatus according to claim 8, wherein the region binarization unit has a plurality of sets of sizes and overlap amounts of the regions, and divides the input image into overlapping regions in each set. .   In the region binarization unit, the luminance threshold value and the pixel number threshold value include a plurality of sets in which a plurality of threshold values are combined, and the defect region image is obtained by combining defective regions obtained by using each set. The defect inspection apparatus of any one of Claims 8-11 which produces | generates.   The defect inspection apparatus according to claim 12, wherein the combination of the plurality of threshold values includes a combination of decreasing the pixel number threshold value when the luminance threshold value is large and increasing the pixel number threshold value when the luminance threshold value is small. The defect inspection apparatus according to claim 8, wherein the input image is an image whose luminance is normalized.

Images