KR20050022320A - Defect inspecting method and apparatus - Google Patents

Abstract

PURPOSE: An apparatus and a system for inspecting a defect of an object are provided to effectively inspect a surface of the object at a high speed by obtaining a plurality of critical values for the defect. CONSTITUTION: An inputted image is divided into overlapped areas. Then, pixels formed in the overlapped areas are binary-combined with a predetermined brightness critical value(S70). Then, a predetermined number of the pixels are selected from among the pixels existing in the overlapped areas. If the number of selected pixels exceeds the predetermined level, the area having the pixels is determined as it has the defect(S90). Then, a binary image is obtained based on the inputted image. Then, the logical OR between the binary image and the defected image is obtained(S100).

Description

DEFECT INSPECTING METHOD AND APPARATUS}

BACKGROUND OF THE INVENTION Field of the Invention The present invention relates to a defect inspection method and apparatus, and for example, to a method and apparatus for inspecting defects such as scratches on an object surface.

Background Art Conventionally, inspection of surface defects, for example, using steel sheets, aluminum plates, or the like as inspection targets has been performed. One method is to obtain a surface image of an inspection object and extract a portion that is brighter than other places, or conversely, a portion that is darkened to determine whether or not it is a defect. According to this inspection method, scratches are obtained by signal processing the acquired surface image. It can be specified.

In such a surface flaw detection method, it is determined for each pixel of the obtained surface image whether it is more than a predetermined | prescribed luminance threshold value, and the pixel more than the threshold value (or less) is extracted as a pixel which may comprise a flaw. Thereafter, labeling processing (pixel connection processing) is performed on the extracted pixels. It is determined that the scratches result in the area exceeding the predetermined threshold as a result of the labeling.

For example, this method is effective in the case that there are some scratches on the mirror-finished surface. However, when there is a land shape or contamination on the inspection object, these are detected as scratches. Further, as a result, even all pixels which become noise become a target of the labeling process, which makes it difficult to increase the processing speed and also requires a high performance processing apparatus for processing.

Moreover, although not aiming at the improvement of a processing speed, the prior art which provided several threshold values according to the kind of defect in order to raise detection accuracy is known (refer patent document 1).

[Patent Document 1]

Japanese Patent No. 2890801

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

In order to achieve the above object, the inventors first divide an input image into overlapping regions, binarize the regions within a predetermined luminance threshold, and calculate a predetermined number of pixels ("0" or "1") among the binarized pixels. When the number of pixels exceeds a predetermined pixel number threshold, the area is referred to as a defective area, and the threshold value is provided in the number of pixels therein. Thus, noise that will be taken conventionally (for example, low luminance and small size). It was found that only defects can be taken by removing. In the case where the regions are arranged so as not to overlap, such as tile arrangement, even if they are divided at the boundary of the region and the count value of the number of pixels is halved and it is not judged to be a scratch, the reduction of the count value can be prevented by overlapping the regions. It has been found that omission of detection can be prevented.

The present invention obtains a defect area image by obtaining the defect area, obtains a binarized image that binarizes the input image as a whole, and performs defect inspection based on a detected defect binarized image obtained by taking a logical product of the binarized image and the defective area image. Since noise can be removed at the previous stage of the labeling process, the burden of subsequent labeling can be greatly reduced.

In addition, in the present invention, the shape of the area for dividing the input image can be rectangular, and the area and the amount of overlap are also provided in plural sets, and the plural sets can be used to divide the input image into overlapping areas. It is also possible to detect only the vertically long scratches, for example, by changing the size of the scratches according to the scratches to be detected. In this case, the processing can be performed at a much lower load as compared with the detection of vertically long scratches using a conventional spatial filter.

In the present invention, the area of the scratches occupied in the rectangular area is increased, the area of the noise occupied in the rectangular area is reduced to increase the noise removal performance, and the pixel number threshold for recognizing the scratches is set to a value close to the rectangular pixel number. do. Since the noise image is smaller than the scratch image, the area of the scratches occupied in the rectangular area can be enlarged by using a rectangle having the same area as the scratch area actually generated.

However, when processing a scratch image larger than a rectangular area, a problem arises that a part of a scratch image is missing. For example, as shown in Fig. 15, when the vertical ellipse-shaped scratches are divided into non-overlapping rectangles such as ①, ②, and ③, the rectangle of ① contains only the tip of the vertical ellipse-shaped scratches. I do not go. Therefore, since the number of pixels in which the scratch image occupies a rectangle does not satisfy the predetermined pixel number threshold, it is treated as noise, resulting in a problem that a part of the scratch image is missing. In order to be processed without segmentation of the scratched image, the rectangular areas need to overlap as shown in (1), (2), (3), (4) and (5) in FIG.

As described above, in order to prevent a part of the scratch image larger than the rectangular area from being missing, it is necessary to take a large amount of overlap of the rectangle, but as a result, the amount of the rectangle to be processed increases, resulting in slowing down the processing speed of the entire image. A problem arises.

Thus, the inventors found that the shape of the area is a rectangle, and the size of the rectangle is determined to include scratches to be detected for each process to be inspected, and the size of the rectangle is a vertical (l) pixel and a horizontal (m) pixel. In this case, if the amount of overlapping rectangles (number of pixels) is at least 1/4 of l × m, it does not interfere with the processing speed of the entire image, and does not divide a large scratched image, thereby providing sufficient noise removal performance. It was found to exert.

14 shows a method of determining the rectangular size.

The scratch image and noise image to be detected are prepared, and the size of the rectangle is determined so as to substantially include the scratch image. In Fig. 14A, a suitable rectangular size is shown. If the size of the rectangle is too small as shown in Fig. 14 (b) or the size of the rectangle is too large as shown in Fig. 14 (c), it is difficult to distinguish the scratched image from the noise.

When the size of the rectangle is vertical (l) and horizontal (m), for example, as shown in Fig. 2, the entire processed image is divided while overlapping l / 2 in the longitudinal direction and m / 2 in the lateral direction. The rectangle is cut out of the processed image. If the number of pixels of the scratch candidate having a luminance higher than or equal to the threshold for each rectangle is greater than or equal to the predetermined value, the rectangle is recognized as including real scratches and the rectangle remains as scratched. The rectangle is erased with no scratches.

Then, by taking the logical product of the binarized image of the entire processed image and the scratch image in each rectangular unit, only the real scratch image is extracted to remove the noise component.

In addition, the luminance threshold value and the pixel number threshold value are also provided in plural sets, and a defective area can be obtained by each set, and these can be combined to form a defective area 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 count threshold may be combined.

Thus, by having a some threshold value, the threshold value according to a flaw can be applied. For example, a defect with a large change in luminance may be taken even a small one, and a defect with a small change in luminance may be taken only a large one.

In addition, the input image can be an image in which luminance is normalized.

According to this configuration, even when it is difficult to binarize the screen to the same threshold value due to uneven illumination, the full screen can be processed to the same luminance threshold value by obtaining a normalized image and processing it based thereon.

First, the outline | summary of embodiment of this invention is demonstrated.

Fig. 1 shows a binarized image by a conventional method, which binarizes an image on an inspection surface using a luminance threshold. The flaw K and the noise part N are shown 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 this embodiment, the original image (the image before binarization) is divided into overlapping rectangles, and it is determined whether or not there are scratches in the rectangular units. This is shown in FIG. 2 shows two rectangles covering the scratches K, but actually covers the entire screen with these rectangles. The rectangle B is horizontal (l) and vertical (m), and the horizontal (x) is overlapped by m / 2 and the vertical (y) by l / 2. By dividing into overlapping rectangles as described above, it is determined whether or not scratches exist for each rectangle. In other words, if the data is first binarized using the luminance threshold and the number of pixels exceeding the luminance threshold is counted for each rectangle, and the result exceeds the pixel threshold given to the rectangle, the rectangle becomes a scratched rectangle. In addition, the reason why the rectangles overlap is to prevent the missing of defects at the boundary of the rectangle.

Fig. 3 shows that the scratched rectangle is taken out. In Fig. 3, only the rectangle covering the scratch portion 3 remains as a scratched rectangle. Referring to the noise portion N of FIG. 1, even if the luminance threshold is satisfied, the pixel threshold given to the rectangle has not been exceeded, and thus is not shown in FIG. As a result, the noise portion is removed so that only a scratched rectangle indicating the presence of the scratch K is obtained. Here, if the AND of Figs. 1 and 3 is taken, a binarized image can be obtained in which the noise portion N is removed and only the scratch K is present from Fig. 1. By ANDing only the scratched rectangle and the original image, the noise is removed at the previous stage of the labeling process, thereby improving the processing efficiency.

4 to 12, the outline of the embodiment of the present invention will be described.

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

As shown in Fig. 4, first, in step S10, the surface image of the inspection object 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 in step S50, a normalized image is obtained. Therefore, in step S20, each pixel of the input image is multiplied by the luminance reference value (128 in this example), and in step S30, an averaging process by a low pass filter is performed on the input image, and the average luminance around each pixel is adjusted. Get For example, a moving average around each pixel 32 × 32 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 to obtain a normalized image in step S50. In addition, the multiplication process of step S20 is for storing the gradation represented by the integer (0 to 255).

Fig. 7 shows an image which has finished the luminance normalization process in step S50. In the upper left, there is a light colored portion P1 having a large area, a small but black portion P2 in the lower right, and a set of many thin and small portions P3 in the central longitudinal direction.

If the input image is normalized in this manner, even if the screen cannot be binarized to the same threshold value in the input image state from the detector due to illumination unevenness or the like, processing can be performed using the same threshold value for the entire screen.

In the next step S60 (Fig. 5), the normalized image is divided into rectangular areas with overlaps. In this example, the same rectangle as in Fig. 2 described above is used. As described above, the rectangle B is horizontal (l) and vertical (m), and the horizontal (x) is overlapped by m / 2 and the vertical (y) by l / 2.

The rectangular area specifies the location of the absorption, and its shape and size can be determined according to anticipated scratches. In general, it is better to set the shape and size so that the scratch can be covered slightly larger than the expected scratch. For example, if the vertically long scratches are part of a dotted line shape, the vertically long rectangular area can be divided into the shorter rectangular areas in the longitudinal direction even though the vertically long scratches can be accurately recognized. The dotted line portion is judged as noise and discarded, which may misinterpret the shape of the correct collection. As a matter of course, the rectangular area does not have to be set larger than the scratch. The rectangular area may be usually about 16 x 32 pixels, and for the vertically long scratches, for example, 16 x 64 pixels may be taken.

Since the shape and size of the rectangular area can be changed corresponding to the scratch to be detected, for example, only the vertically long scratches can be detected using the vertically long rectangular area. Conventionally, in the case of vertically long scratches, processing was performed using a spatial filter. However, the spatial filter had a large computational load, and a heavy load was applied to the processing. By using the rectangular area of the present invention, it is possible to handle with a much lower load.

In the present example, the overlapped portion is m / 2 and the verticaly y is l / 2, but the size of the overlapped portion is arbitrary and can be appropriately determined. The meaning of overlapping rectangles is that when the rectangles are arranged without overlapping, the damage may be omitted if there is damage in the boundary portion of the rectangle. In the present invention, the number of pixels that satisfies a predetermined condition is counted, and a process is performed in which a rectangle that is equal to or larger than a predetermined number of pixels is formed as a scratched rectangle. At that time, the scratch is divided at a rectangular boundary, for example, both rectangles. This is because when the number of pixels of the scratches is reduced to half, missing counts may occur and may not be recognized as scratches. Therefore, in order to prevent a missing count, it is better to make the overlap part (x, y) larger and 1/2 or more. However, when the overlapping portion is enlarged, many rectangles are required, so processing time is required. In this example, it is made equal to 1/2 as mentioned above. In addition, although the image segmentation area was made into a rectangle, other polygons, such as a triangle and a parallelogram, may be sufficient. Either way, what is necessary is just an area | region which can be divided so that a test | inspection image may be covered.

In the following steps S70 to S90 and S71 to S91, scratched rectangles are specified using sets of different luminance thresholds and pixel number thresholds, respectively.

In step S70, the rectangular area is binarized using the first luminance threshold value. In step S80, the number of pixels exceeding the brightness threshold is counted for each rectangular area, and in step S90, the rectangle counting the result exceeds the first pixel number threshold. It is set as a scratched rectangle. Here, it is conceivable that the set of the luminance threshold value and the pixel number threshold value increases the first pixel number threshold value so that the first luminance threshold value is close to 128 (average value), for example. In the first set, the thinly extended region can be determined as a scratch.

An example of processing the image of FIG. 7 with such a set of thresholds is shown in FIG. Since the luminance threshold value is close to the average value and the pixel count threshold value is large, the color P and the wide area P1 are covered with a rectangle B1 and detected. Rectangles corresponding to other small areas P2 and P3 are not shown.

Similarly, the rectangular region is binarized using the second luminance threshold in step S71, the number of pixels exceeding the luminance threshold is counted for each rectangular region in step S81, and the result counted in step S91 exceeds the second pixel count threshold. Let the rectangle be a scratched rectangle. Here, the second luminance threshold is farther from 128 (that is, closer to 0 or 255), and the second pixel count threshold is made smaller. In the second set, black and small areas can be determined as scratches.

This result is shown in FIG. Since the pixel value is small but the luminance threshold is far from the average value, only the rectangle B2 covering the portion P2 which appears as a black dot is shown. The other portions P1 and P3 are excluded because they do not satisfy the luminance threshold.

In the present example, two sets of sets are used, but the present invention is not limited thereto and may be three or more sets. The combination of the luminance threshold value and the pixel number threshold value can be variously changed depending on the defect to be detected. It is also possible to use only a single set without using a plurality of sets.

In the next step S100, a scratched rectangular mask image is obtained in step 110 by combining the rectangles determined as scratched rectangles in step S90 or step S91 (usually by an OR operation).

Fig. 10 is a scratched rectangular mask obtained by taking a logical sum of the scratched rectangular masks B1 and B2 (FIGS. 7 and 8) obtained by using the first and second threshold sets, in accordance with step S100.

If the purpose is only to determine whether there is a scratch, the place of the scratch is specified by the scratched rectangular mask thus obtained, and may be terminated here. In addition, although it combines by taking a logical sum here, you may perform another logical operation according to a set of a luminance threshold value and a pixel number threshold value.

To find the exact shape of the scratch, go to the next step.

Therefore, using the normalized image of step S50 in step S120, normal binarization using a predetermined luminance threshold value (though the threshold value used in step S70 may be sufficient or other values may be sufficient) is performed, and a normal binarization image is carried out in step S130. Get

Fig. 11 is obtained by the 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 labeling of many of these pixels has placed a heavy load on the processing apparatus.

Next, in step S140 (Fig. 6), the logical AND of the normal binarized image obtained in step S130 and the scratched rectangular image obtained in step S110 is taken.

12 is a logical product AND of a scratched rectangular mask (FIG. 10) and a normal binarized image (FIG. 11), which is the scratch detection binarized image of step S150. As shown in Fig. 12, the light and large scratches P1 and the small and dark scratches P2 are preserved in size and shape and can be detected, and the portion P3 that becomes a noise is removed. In this manner, surface noise and the like appearing in a normal binarized image are erased, and only the scratches to be found are detected.

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

In general, a labeling process is required to know the shape and size of the scratch as well as the presence or absence of the scratch. It is preferable to binarize to a low luminance threshold in order to obtain a binary image showing the appearance of accurate scratches, but if the threshold is lowered, the surface noise is also taken out and a labeling process must be performed on a large number of pixels, thereby increasing the load of the labeling process. It was a result. In this embodiment, the scratched area is identified and the noise is removed at the previous stage of the labeling process by using a scratched rectangular mask. Therefore, the burden of labeling could be greatly reduced.

In step S180, the harmfulness of the scratch is determined on the basis of the length, width, and peripheral length of the scratch, and in step S190, the scratch image is displayed.

According to the present invention, it is possible to obtain a rectangular mask having scratches before image processing such as labeling processing, which requires a large load for each pixel, thereby eliminating the scratches such as surface noise, thereby limiting the target of the image processing. Therefore, the efficiency of the flaw detection processing can be greatly increased.

Next, with reference to FIG. 13, the outline | summary of one Embodiment of this apparatus is demonstrated.

For example, an inspection image is taken into the image input unit 1 from an inspection object such as steel plate ST through 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 rectangular image with scratches and a normal binarizing image, respectively, and an AND of both images is taken by the calculating unit 5 to remove the scratch detection binary image. Get Thereafter, labeling and feature extraction are performed in the image processing unit based on the scratch detection binary image, and the shape and size of the scratch are determined to determine the degree of harmfulness of the scratch. In the image display unit, a scratch image after the image processing is displayed.

In this example, the steel sheet is described as an example, but the inspection object is not limited to the steel sheet. For example, a slab may be used. For example, the steel sheet may be applied to a fluorescent image or an ultrasonic image obtained by a magnetic image. In addition, the present invention can be applied as long as the image of the inspection object is obtained.

According to the present invention, there is provided a surface inspection method and apparatus which can perform an efficient treatment at high speed.

1 is a schematic diagram showing a conventional binarized image.

2 is an explanatory diagram showing a rectangular region of one embodiment of the present invention;

3 is an explanatory diagram showing a scratched rectangle of an embodiment of the present invention;

4 is a first flowchart illustrating a flow of an embodiment of the present invention.

5 is a second flowchart illustrating a flow of an embodiment of the present invention.

6 is a third flowchart illustrating a flow of an embodiment of the present invention.

7 shows a normalized image of one embodiment of the present invention;

Figure 8 shows a scratched rectangle obtained by the first set of threshold values in one embodiment of the present invention.

Fig. 9 shows a scratched rectangle obtained by the second threshold set in one embodiment of the present invention.

Fig. 10 is a view showing a combination of scratched rectangles shown in Figs. 8 and 9;

Fig. 11 is a diagram showing a normal binarized image in one embodiment of the present invention.

Fig. 12 is a diagram showing a scratch detection binarization image in one embodiment of the present invention.

Figure 13 illustrates a device configuration of one embodiment of the present invention.

14 shows a method of determining the rectangular size.

Fig. 15 shows a method for determining the rectangular size.

<Explanation of symbols for the main parts of the drawings>

K: scratch

B, B1, B2: rectangular area

N: Noise part

Claims (14)

Dividing the input image into overlapping regions; Binarizing the region to a predetermined luminance threshold; Calculating the number of pixels having a predetermined value among the binarized pixels in the area; And said area is referred to as a defect area when the number of pixels exceeds a predetermined pixel number threshold. Dividing the input image into overlapping regions; Binarizing the region to a predetermined luminance threshold; Calculating the number of pixels having a predetermined value among the binarized pixels in the area; Obtaining a defective region image as a region where the region is scratched when the pixel number exceeds a predetermined pixel number threshold; Binarizing the input image to a predetermined luminance threshold to obtain a binarized image; And taking a logical product of the binarized image and the defective area image. The defect inspection method according to claim 1 or 2, wherein the shape of the region is rectangular. 4. The method according to any one of claims 1 to 3, wherein the dividing of the input image into overlapping regions has a plurality of sets of sizes and overlapping amounts of the regions, and in each set, a defect for dividing the input image into overlapping regions. method of inspection. The said luminance threshold value and the said pixel-number threshold value consist of a plurality of sets which combined the several threshold value, The combination of the flawless area | region obtained using each set of Claims 1-4. The defect inspection method of obtaining the said defect area image. 6. The defect inspection method according to claim 5, wherein the combination of the plurality of thresholds has a combination of increasing the pixel number threshold when the luminance threshold is large and increasing the pixel number threshold when the luminance threshold is small. 7. The defect inspection method according to any one of claims 1 to 6, wherein the input image is an image in which luminance is normalized. An image input unit for inputting an image of an inspection target; When the input image is divided into overlapping regions, binned in the region to a predetermined luminance threshold, the pixel number of a predetermined value among the binarized pixels in the region is calculated, and the pixel number exceeds a predetermined pixel number threshold. An area binarizing unit which generates a defect area image by calling the area a defect area; The defect inspection apparatus provided with the image display part which displays a defect area image. An image input unit for inputting an image of an inspection target; The input image is divided into overlapping regions, binned within a region to a predetermined luminance threshold, and the pixel number of a predetermined value among the binarized pixels within the region is calculated, and the pixel number exceeds the predetermined pixel number threshold. An area binarizing unit for generating a defective area image by calling the area a defective area; A normal binarizing unit for binarizing the input image at a predetermined luminance threshold to obtain a binarized image; An operation unit for obtaining a defect inspection binarization image from the defect area image and the normal binarization unit; The defect inspection apparatus provided with the image display part which displays the said defect inspection binarization image. The defect inspection apparatus according to claim 8 or 9, wherein the shape of the region is rectangular. The defect inspection apparatus according to any one of claims 8 to 10, wherein the area binarization unit has a plurality of sets of sizes and overlapping amounts of the area, and divides the input image into an overlapping area in each set. 12. The flaw obtained according to any one of claims 8 to 11, wherein in the area binarization unit, 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. The defect inspection apparatus which produces | generates the said defect area | region image by the combination of the area | region which exists. The defect inspection apparatus according to claim 12, wherein the combination of the plurality of thresholds has a combination of increasing the pixel number threshold when the luminance threshold is large and increasing the pixel number threshold when the luminance threshold is small. The defect inspection apparatus according to any one of claims 8 to 13, wherein the input image is an image in which luminance is normalized.