JPH08189904A - Surface defect detector - Google Patents

Abstract

PURPOSE: To easily and accurately detect the surface defect and judge the shape or defect kind and class thereof with excellent accuracy by paying attention to a part of gray level picture inputted from a sensing element, and judging the presence or absence of defect by a defect judging part first. CONSTITUTION: While a steel plate 1 is illuminated by an illumination 2, the surface thereof having a defect 3, etc., is picked up by a camera 4. A signal processing circuit 5 stores an A/D converted picture data in a picture memory 6, and the data read out from the memory 6 is differentiated by a differentiating circuit 7. The obtained differentiated value shows the change ratio of illuminance value in the area adjacent to respective picture elements, so an area setting circuit 8 sets a rectangular area picture element with an apporpriate size centering the picture element of the maximum differentiated value. A presence/absence information extracting circuit 9 extracts a presence/absence information for judging the presence or absence of defect in respective areas, by using party informations in the respective areas. A presence/absence judging circuit 10 calculates the squared sum and the number of picture elements of differentiated information with specified value or more reagarded as the maximum differentiated value and defective edge in the area, so as to judge the presence or absence of defect.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a surface defect detecting device for inspecting a surface defect of a product such as a steel plate carried out from a production line.

[0002]

2. Description of the Related Art In a conventional surface defect inspection apparatus, for example,
While the metal plate is being transferred, its surface is imaged by a CCD area sensor, and the surface defect of the metal plate is inspected based on the surface image. In such a surface defect inspection apparatus, since a defect is often a luminance change portion, the defect is often detected based on an emphasized image such as a differential image obtained by differentiating the surface image. Japanese Unexamined Patent Publication No. 1-25363
According to Japanese Patent Laid-Open No. 9 and Japanese Patent Application Laid-Open No. 5-126759, several kinds of predetermined differential processing or differential filters matching the shape and direction of the target defect are made to act, and the defect is detected by the differential value calculated by this. .

[0003]

However, the above-mentioned differential processing does not necessarily emphasize only the edge portion of the defect but may also emphasize the noise included in the background portion, which may cause over detection. turn into. On the contrary, when a defective image having a blurry edge is differentiated, the edge is not emphasized so much that the edge and the noise cannot be distinguished. Therefore, it is necessary to judge the defect from the differential value of the area having a predetermined size. Further, since the surface defect inspection apparatus is often used in various FA fields including steel lines, and higher speed processing is required, it is better that the differential processing for one surface image is small.

When there is a defect in the surface image, in order to obtain the type and grade of the surface defect, for example, the surface of the inspection object to be inspected is imaged and the features of the surface defect such as a flaw are detected from the image information. However, there is known a technique of performing a function operation on this feature together with parameter information set separately. However, for example, the types of flaws that occur in a steel sheet production line are generally as diverse as more than 50 types, and the grade is 5
It can be divided into stages. Such a precisely to determine the number of types and grades, are extremely important to set the parameters to accurately, what parameters because many defect type is effective in the determination Is difficult to determine, and it becomes impossible to deal with the increase / decrease in defect types, especially when the number increases. Further, in the case of an uneven defect, it may be possible to determine the shape or defect by detecting the shape feature amount, but in the case of a non-uneven planar defect, it is difficult to detect the shape feature amount and the defect type is determined. It is difficult to judge from the shape.
Therefore, it is necessary to determine the defect type from the image information.

Generally, in order to calculate the shape feature amount, the input grayscale image is binarized and measured. Conventionally, image binarization is performed by comparing an image signal obtained by scanning an inspection object with a threshold value obtained by a predetermined calculation. However, in general, such an image signal may have a large variation in its output level. For example, when an image is read by a CCD line sensor or CCD area sensor, the output level fluctuates greatly due to non-uniformity of the characteristics of each pixel of the sensor and uneven illumination. Further, the output level greatly changes due to the change of the incident angle of the scanning beam on the surface to be inspected and the change of the incident angle on the optical system. In this way, when the output level includes a large fluctuation, a signal indicating a defect is buried in this large fluctuation when compared with the threshold value, or conversely, the noise included in the simple background part is eliminated. It may be detected as a defect. Although it depends on such a reason, generally, when binarization is performed, image information is extremely lost, so that an accurate shape feature amount cannot be obtained, and an accurate defect type cannot be determined.

In view of the above circumstances, the present invention is capable of easily and highly accurately detecting a surface defect, accurately and flexibly determining the shape of a defect, a defect type, and a grade, and increasing or decreasing the defect type. It is an object of the present invention to provide a surface defect detection device that can flexibly deal with the problem.

[0007]

A surface defect detecting apparatus of the present invention which achieves the above object is a surface defect detecting apparatus for detecting a surface defect of an object based on image data representing a grayscale image of the surface of the object. In (1), an image data storage unit for storing the image data, (2) defect presence / absence determining unit, and (3) a defect evaluation unit operating in parallel with the defect presence / absence determining unit.

Of these, the defect presence / absence determining unit of (2) is (2-1) differential processing means for generating differential image data by differentiating the image data stored in the image data storage unit (2- 2) Based on the differential image data, area setting means for setting one or a plurality of image areas having the above-mentioned surface defects and having high probability in the grayscale image (2-3) set by the area setting means For each image area, there is a defect presence / absence determining means for determining the presence / absence of the surface defect based on the differential image data corresponding to the image area.

Further, the defect evaluation section of (3) is (3-1) image data or differential image data corresponding to each image area judged to have a defect by the defect presence / absence judgment section, for each image thereof. Image memory for storing each area (3-2) A feature amount representing the characteristics of the surface defect for each image area based on the image data or the differential image data for each image area stored in the image memory (3-3) Surface defect identifying means for identifying the surface defect for each image region based on the characteristic amount (3-4) For identifying the image defect for each image region A gradation image evaluation means is provided for evaluating the surface defects over the entire gradation image by integrating the surface defects.

Here, the area setting means searches for the maximum absolute value of the differential image data corresponding to the grayscale image, sets an image area including the maximum value, and after the setting, performs the grayscale processing. Repeating the step of searching for the maximum absolute value of the differential image data corresponding to the part of the image excluding the set image area and setting the image area including the maximum value until a predetermined condition is satisfied. Is preferred.

Further, the defect presence / absence determining unit corresponds to the first index indicating the degree of variation of the differential image data corresponding to each of the image areas and the differential image data exceeding a predetermined value in the differential image data. It is preferable that a second index indicating the number of pixels to be obtained is determined, and the presence or absence of the surface defect in each image area is determined based on the first and second indices.

The feature quantity extracting means may extract the feature quantity by causing a template corresponding to the surface defect to act on image data or differential image data corresponding to each of the image areas. . Alternatively, the feature amount extraction means generates a density co-occurrence matrix from the image data or differential image data corresponding to each of the image regions, and based on the density co-occurrence matrix, the feature amount is the contrast or the second moment. , And correlation may be obtained.

[0013]

The surface defect detecting apparatus according to the present invention responds to the line speed by paying attention to a part of the grayscale image input from the image pickup device or the like, and first determining whether or not there is a defect in the defect presence / absence determining unit. The inspection can be performed with high accuracy. Further, when it is determined that there is a defect, in the defect evaluation unit, the defect type, the grade, etc. are determined independently of the presence / absence determination process of the defect. You can judge the grade.

According to the present invention, several image areas of appropriate size are set in a place where a defect is likely to exist in a grayscale image, so that defect inspection of the entire grayscale image can be performed with high accuracy. The accuracy can be comprehensively done.

[0015]

Embodiments of the present invention will be described below. FIG. 1 is a schematic configuration diagram of a first embodiment of a surface defect detecting device of the present invention. The steel sheet 1 as the transported inspection object is illuminated by the illumination 2. The surface of the steel sheet 1 has defects 3 such as unevenness. The surface image of the steel plate 1 captured by the camera 4 is A / D converted by the signal processing circuit 5, and the output image data is stored in the image memory 6 (an example of the image data storage section referred to in the present invention). . A differential filter is applied to the image data read from the image memory 6, and a differential processing circuit 7 performs differential processing.

FIG. 2 is an explanatory diagram of the differential processing in the differential processing circuit. In this differentiation processing, as shown in FIG. 2, for example, a differentiation filter F is superposed on a 3 × 3 area centering on a processing target pixel, a product of corresponding pixels is obtained, and the sum of them is differentiated from the processing target pixel. The value. This operation is performed from the upper left pixel to the lower right pixel.

FIG. 3 is a diagram showing a Sober filter as an example of the differential filter shown in FIG. The Sober filter shown in FIG. 3 is composed of an X-direction filter (a) having a weighting coefficient in the X direction and a Y-direction filter (b) having a weighting coefficient in the Y direction. When such a Sober filter is used, not only an edge in one direction but also an edge in all directions can be detected and a differentiation process can be performed in which the influence of background noise is suppressed. In this case, Δxy is a differential value, where Δx is the result of applying the X-direction filter (a) and Δy is the result of applying the Y-direction filter (b). Here, the differential value Δxy is expressed as Δxy = √ (Δx ² + Δy ² ).

Since the differential value Δxy thus obtained represents the rate of change of the brightness value in the neighborhood area of each pixel,
The pixel having the largest differential value is the place where the defect is highly likely to exist in the grayscale image. Therefore, the area setting circuit 8 shown in FIG. 1 sets a rectangular area (for example, 64 × 64) pixels having an appropriate size centering on the pixel having the maximum differential value. Next, the rectangular area centered on the pixel having the maximum differential value is set again so that the pixel in the rectangular area once set is not set as the center of the rectangular area. This is repeated until a predetermined number (for example, 10) or a differential value at the center of the rectangular area reaches a predetermined value (for example, 50). Then, in the presence / absence information extraction circuit 9, by using the differential information in each rectangular area,
Presence / absence information for determining whether or not a defect exists in each rectangular area is extracted. As the presence / absence information, the maximum differential value in the rectangular area, the sum or squared sum of the differential information equal to or larger than a predetermined value that can be regarded as the defect edge, and the number of pixels of the differential information that is equal to or larger than the predetermined value that can be regarded as the defect edge are calculated. First, if the maximum differential value is greater than a predetermined value that is considerably large, it is unconditionally determined to be defective. If the maximum partial value is less than or equal to the predetermined value and there is a possibility of noise, the sum of the absolute values of the difference between the predetermined value of the differential information that can be regarded as the defect edge and the value greater than that or the sum of squares, and the defect The presence / absence of a defect on the surface of the steel 1 is determined based on the number of pixels whose differential information, which can be regarded as the edge, has a value equal to or larger than a predetermined value. The sum of the absolute values of the differences or the sum of squares represents the edge strength or contrast of the entire image area, and the number of pixels represents the number of defective edges.

FIG. 4 is a diagram showing the presence / absence of surface defects on the steel sheet. The presence / absence determination circuit 10 determines the presence / absence of a surface defect in the steel sheet 1 within the range shown in FIG. If it is determined that there is a defect, the image information or the differential information of each rectangular area that is determined to have a defect is stored in the image memory 11_1 shown in FIG.
~ 11_n. Characteristic information extraction circuits 12_1 to 1
At 2_n, a template prepared for each defect type, grade, or shape is applied to the image information in the image memory. This template conforms to the shape of the defect, and if the defect part has a higher brightness value than the background, the weighting factor at the defect part is a positive number, and the weighting factor at the background part is a negative number. The sum of the weighting factors is set to a large value to some extent so that when the template is applied, a difference is generated between the defective portion and the other portion, and considering the influence of noise and the like. If the luminance of the defective portion is lower than that of the background, the positive number portion and the negative number portion of the weighting factor are exchanged.

FIG. 5 is a diagram showing each example of the template. FIG. 5 (a) is a linear defect template, and FIG. 5 (b).
Is a point defect template. FIG. 6 is an explanatory diagram of processing by the template shown in FIG. Prepare a template as shown in Fig. 5 for each defect shape,
As shown in FIG. 6, the template is superimposed on a region having the same size as the template in which the pixel to be processed is the upper left pixel, the product of the corresponding pixels is obtained, and the sum of them is used as the feature amount. This operation is performed from the upper left pixel to the lower right pixel. Since the absolute value of the feature amount becomes larger as the defect is closer to the shape of the template and the contrast is higher, the maximum value of the absolute value of each template is obtained and the shape or type and the grade of the surface defect are obtained by using them as input data. In the rectangular area judged to have defects,
When the maximum values of the absolute values of the two templates of FIG. 5 are compared and the maximum values of the point defect template shown in FIG. The shape of the existing defect can be judged as a defect close to the shape of the point defect template among the shapes of the two templates. However, even if the shapes are different, the value of the feature amount becomes large if the contrast is strong, so to compare the feature amount between templates, equalize the maximum and minimum absolute values of the feature amount. It is necessary to normalize the weights. For example, when the image information has a brightness value in the range of 0 to a, the range of absolute values that can be taken when each template in FIG. 5 is applied is 0 to 24a. The shape or type of the surface defect in the defect type determination circuits 13_1 to 13_n using these feature amounts as input data,
The grade is calculated for each rectangular area. Gray image evaluation circuit 14
At, the number of rectangular areas including point defects based on the surface defect information for each rectangular area, the number of rectangular areas including linear defects is calculated, and the shape or type of the surface defects of the entire grayscale image,
Judge the grade comprehensively.

Next, a second embodiment of the surface defect detecting device of the present invention will be described. The process up to the point where it is determined that there is a defect is the same as in the first embodiment, and the feature information extraction circuit 12_1
12 to 12_n, a density co-occurrence matrix is generated from image information of a rectangular area in which a defect exists, and three types of feature amounts related to texture obtained from the density co-occurrence matrix, that is, the contrast, the second moment, and the correlation are calculated for each rectangular area. Ask for each.

Here, the contrast is an index representing the contrast of the image, the second moment is an index representing the uniformity of the texture, and the correlation is an index representing the directionality of the texture. FIG. 7 is a diagram (b) showing the image information (a) and the density co-occurrence matrix M (m, n). From a pixel (i, j) on the image information shown in FIG.
i, Δj) (This is σ. (−Δi, −Δj) = −
Pixels (i + Δi, j) that are separated by σ including σ)
+ (Δj) luminance value pair (a (i, j), a (i + Δi,
The number of appearances of (j + Δj)) becomes an element of the density co-occurrence matrix M (m, n).

Further, each element of the density co-occurrence matrix M (m, n) is normalized by dividing it by the sum of all elements, and this is normalized by p
(M, n)

[0024]

[Equation 1]

It is represented by However, since these feature quantities also differ depending on σ, it is necessary to determine σ that matches each defect. Further, since these feature values evaluate the texture of the entire image information in each rectangular area, a planar defect that spreads like unevenness is more characteristic than a defect that occurs in a small number suddenly. It easily appears in the quantity.

FIG. 8 shows the ripples (a), burn spots (b) and streaks (c) generated on the surface of the surface-treated steel sheet 1.
3A and 3B are schematic views showing three types of planar defect images and an image (d) having no planar defect. Here, the texture feature amount of the image of three types of planar defects and the image without defects shown in FIG. 8 is calculated (σ =
(0, 3), calculated with density gradation 16). These texture feature quantities are input to a neural network (not shown) as input data, each defect kind is made to correspond to each neuron in the output layer, the output neuron indicating the defect kind is set to 1 and the other neurons are set to 0, and backpropagation is performed. When the defect type is learned by the gating method, the defect type corresponding to the output neuron having the largest value is set as the defect type that can be determined from the texture feature amount. When data different from the learning data was input and judged, the results shown in Table 1 were obtained.

[0027]

[Table 1]

This data was calculated by setting the vector direction of σ in the same direction as the streak, particularly by detecting the streak. However, by changing σ or using different σ simultaneously, Data for discriminating with higher accuracy can be obtained. Further, it is possible to discriminate with higher accuracy by increasing the density gradation.

[0029]

As described above, the surface defect detecting apparatus of the present invention reduces the influence of noise by observing a part of the input grayscale image and judging only the presence or absence of a defect in a small area. Highly accurate inspection that can correspond to the line speed can be performed. Further, the processing time can be shortened by increasing the number of pixels of the input grayscale image and collectively processing the grayscale images in a wider range. Further, if a differential filter that detects edges in all directions and is not easily affected by background noise or a differential filter that operates in a similar manner is used as the differential filter, the detection accuracy increases.

Further, a method such as comparing the image information with a preset threshold value is affected by shading, but the differential processing is not affected by shading. There is no need to consider uniform illumination. Further, since the edge of the defect in the small area is observed by paying attention to a part of the grayscale image, even a defect having an unclear edge can be recognized, and a highly accurate inspection can be performed.

When it is determined that there is a defect, the defect type, grade, etc. are determined independently of the defect presence / absence determination process. You can judge the grade. By setting several rectangular areas of appropriate size in places where there is a high possibility that defects exist in the grayscale image, comprehensive defect inspection of the entire grayscale image can be performed with high accuracy. You can When a template prepared for each defect type, grade, or shape is applied to the image information of the rectangular area determined to have a defect, the defect for each defect type, grade, or shape can be determined. Further, since it is sufficient to increase or decrease the number of templates to increase or decrease the number of defect types, it is not necessary to change the parameters for discrimination. On the other hand, with respect to a planar defect that spreads like unevenness, if the texture feature amount is calculated and determined, the defect determination can be performed with higher accuracy.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of a first embodiment of a surface defect detection device of the present invention.

FIG. 2 is an explanatory diagram of differential processing in a differential processing circuit.

FIG. 3 is a diagram showing a Sober filter as an example of the differential filter shown in FIG.

FIG. 4 is a diagram illustrating the presence / absence of surface defects on a steel sheet.

FIG. 5 is a diagram showing a linear defect template and a point defect template.

FIG. 6 is an explanatory diagram of processing by the template shown in FIG.

FIG. 7 is a diagram showing image information and a density co-occurrence matrix M (m, n).

FIG. 8 is a diagram showing an image of three types of sheet-like defects such as ripples, uneven burns, and streaks occurring on the surface of a surface-treated steel sheet, and an image without sheet-like defects. It is a schematic diagram of the example of a defect of the steel plate surface shown in FIG.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Steel plate 2 Illumination 3 Defect 4 Camera 5 Signal processing circuit 6 Image memory 7 Differentiation processing circuit 8 Area setting circuit 9 Presence / absence information extraction circuit 10 Presence / absence determination circuit 11_1, ..., 11_n Image memory 12_1, ..., 12_n Feature information extraction circuit 13_1 ..., 13_n Defect type judgment circuit 14 Grayscale image evaluation circuit

Claims (5)

[Claims] 1. A surface defect detecting apparatus for detecting a surface defect of a subject based on image data representing a grayscale image of the subject surface, comprising: an image data storage unit for storing the image data; Differential processing means for generating differential image data by differentiating the image data stored in, and based on the differential image data, in the grayscale image,
Area setting means for setting one or more image areas having a high probability of having the surface defect, and for each image area set by the area setting means, based on differential image data corresponding to the respective image areas. A defect presence / absence determining unit having a defect presence / absence determining unit for determining the presence / absence of the surface defect; and image data or differential image data corresponding to each image region determined to have a defect by the defect presence / absence determining unit, Based on the image memory stored for each image area and the image data or differential image data for each image area stored in the image memory, a feature amount representing the feature of the surface defect is obtained for each partial image. Feature amount extraction means for extracting, surface defect identification means for identifying the surface defects for each of the image regions based on the feature amounts, and total surface defects identified for each of the image regions To the and a gray image evaluation means for evaluating surface defects over the entire gray scale image, the surface defect detection apparatus comprising the defect evaluation unit which operates in parallel with the defective state determining unit. 2. The area setting means searches the maximum value of absolute values of the differential image data corresponding to the grayscale image to set an image area including the maximum value, and after the setting, the grayscale image. A step of searching the maximum value of the absolute value of the differential image data corresponding to the part excluding the already set image area and setting the image area including the maximum value until the predetermined condition is satisfied The surface defect detection device according to claim 1, wherein 3. The defect presence / absence determining means corresponds to a first index indicating a degree of variation of differential image data corresponding to each of the image areas and differential image data exceeding a predetermined value in the differential image data. A second index representing the number of pixels to be processed is determined, and the presence or absence of the surface defect in each of the image areas is determined based on the first and second indexes. 2. The surface defect detection device according to 2. 4. The feature amount extracting means extracts the feature amount by applying a template corresponding to the surface defect to image data or differential image data corresponding to each of the image regions. The surface defect detection device according to claim 1, wherein 5. The feature amount extraction means generates a density co-occurrence matrix from image data or differential image data corresponding to each of the image regions, and based on the density co-occurrence matrix, the feature amount is a contrast, 4. The method according to claim 1, wherein the second moment and the correlation are obtained.
The surface defect detection device according to claim 1.