KR101675532B1

KR101675532B1 - Apparatus and Method for detecting defect of thick steel plate

Info

Publication number: KR101675532B1
Application number: KR1020140065455A
Authority: KR
Inventors: 윤종필; 박창현; 이주승
Original assignee: 주식회사 포스코
Priority date: 2014-05-29
Filing date: 2014-05-29
Publication date: 2016-11-22
Also published as: KR20150137497A

Abstract

The present invention relates to a surface image photographing apparatus for photographing a surface of a thick plate to generate an original image; A first binarization unit that binarizes the filtered image to generate a first binarized image, and a first binarization unit that binarizes the filtered image to generate a first binarized image, A detection unit; A second binarization unit for generating a second binarized image of the original image by binarizing the image of the corresponding column based on a reference value of a brightness value of the column for each column of the original image; And a second defect detecting unit for detecting at least two second defects for each blob of the binarized image step by step, the second defect detecting unit including a defect candidate detecting unit for generating a defect candidate binarized image by combining the binarized images, The present invention provides an advantageous effect of classifying and detecting defects for which no pattern appears by cross illumination by detecting oil defects, chip defects, scrap and scale in the order of increasing feature values in addition to the irregular defect do.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a defect inspection apparatus,

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and a method for detecting surface defects on a heavy plate, and more particularly, to an apparatus and a method for detecting surface defects on a surface of a heavy plate.

Generally, the surface defect inspection system of steel plate obtains the surface image of the thick plate through the camera placed on the upper plate, and then detects various types of plate defect from the surface image obtained by the image processing technique.

Korean Patent No. 10-1372787 (published on Apr. 3, 2014) discloses a defect detection apparatus using cross illumination.

Particularly, the apparatus for inspecting a thick plate surface defect using cross illumination has a line camera for acquiring a surface image of a material as a thick plate and two lights at the front and rear ends of the line camera along the traveling direction of the material. A plurality of the two lights may be arranged in the width direction of the material.

The above-described two lights are turned on and off alternately by an illumination control signal, respectively. When a first illumination control signal is turned on, a line image is displayed by a line camera, When the signal is turned ON, the image of the next row is obtained by the line camera. When all images of one frame are acquired, defects of the material can be detected by analyzing the obtained image of one frame.

However, such a plate surface defect inspection apparatus has a problem that the scale and the irregular defect can not be classified because the depth is the same as the irregular defect when the scale is peeled from the surface or the peeled scale is attached to the surface.

In addition, oil defects caused by oil drop do not occur in depth, and thus there is a problem that classification can not be performed using cross illumination.

In addition, in the scrap, there is a problem that the depth is not generated due to being pressed on the rear surface, so that it can not be classified using cross illumination. Such scrap may fall off at the time of heavy plate processing and cause irregularities.

Korean Patent No. 10-1372787 (Announcement of Mar. 10, 2014)

Accordingly, the present invention has been made to solve the above-mentioned problems, and it is an object of the present invention to provide a method and apparatus for detecting defects such as oil defects, defects without scales or scales, as well as irregularities detected using patterns generated by cross- And an object thereof is to provide a defect inspection apparatus and method.

The problems to be solved by the present invention are not limited to the above-mentioned problems, and other problems not mentioned here can be understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided an image processing apparatus including a surface image capturing unit for capturing an image of a surface of a thick plate to generate an original image, a filtering unit for filtering the original image to generate a filtered image, A first binarization unit for generating a binarized image, the first binarization unit detecting a first defect in the first binarized image, and a second binarization unit for detecting a first defect in the first binarized image based on a reference value of a brightness value of the column, And a defect candidate detecting unit for generating a defect candidate binarized image by combining the first binarized image and the second binarized image, wherein the defect candidate detecting unit detects the defect candidate binarized image by combining the first binarized image and the second binarized image, And a second defect detecting section for classifying at least two second defects for each blob of the binarized image step by step, Device can be provided.

Preferably, the surface imaging portion comprises a first illumination and a second illumination that illuminate the first point at a location different from the camera that takes a first point of the slab surface, wherein the first illumination and the second illumination And a second image photographed in a state in which only the other one of the first illumination and the second illumination is in an ON state is generated can do.

The filtering unit may generate the filtering image by performing Gabor filtering on the original image.

Preferably, the first binarization unit may obtain an energy matrix from the filtered image and generate the first binarized image by Gaussian filtering the energy matrix.

The first binarization unit may generate the first binarized image by binarizing the filtered image by setting a double-threshold value.

Preferably, the reference value may be a double boundary value determined by an average value and a standard deviation of a brightness value of a corresponding column of a pixel of the original image.

Preferably, the second binarization unit corresponds to 1 if the first boundary value is greater than or equal to the first boundary value, and corresponds to -1 if the second boundary value is less than or equal to the second boundary value. Otherwise, the second binarization unit corresponds to 0 Thereby generating the second binarized image.

Preferably, the second defect is an oil defect, a chip defect, and a scrap, and the second defect detecting unit classifies the oil defect and the first residual image in the defect candidate binarized image, The defect and the second residual image may be classified and the scrap and the scale may be classified in the second residual image.

Preferably, the second defect detector may classify the oil defect and the first residual image through a learned artificial neural network or a support vector machine based on the first characteristic parameter of the defect candidate binarized image .

Preferably, the first characteristic parameter may be texture information, a morphological feature, a shape change value, and a rotation invariant using a histogram of the defect candidate binarized image.

Preferably, the shape change value may be a variance value of the circular judgment distance to the edge of the blob with respect to the center of gravity of the blob.

Preferably, the rotation invariant calculates an average value of the circular determination distances based on at least two reference directions symmetrical with respect to the center of gravity, and corresponds to 1 when the calculated average value is larger than the calculated average value, 0, a first mapping value is calculated for each reference direction around the center of gravity and arranged in a matrix form, a bit operator corresponding to the first mapping value is generated, and the first mapping value Calculating a second mapped value by multiplying the first mapped value by the bit operator and calculating the feature value by summing the calculated second mapped values, and rotating the bit operator stepwise in the reference direction around the center of gravity, The feature value may be one of the feature values after calculating the feature value by the number of feature values.

Preferably, the rotation discomfort may be a minimum value or a maximum value of the characteristic values.

Preferably, the second defect detector may classify a chip defect and a second residual image through a learned artificial neural network or a support vector machine based on a second characteristic parameter of the first residual image .

Preferably, the second characteristic parameter may be a Gray-Level Co-Occurrence Matrix (GLCM) of the first residual image, an average energy of the Gabor filter, an edge direction, a morphological characteristic, and a statistical characteristic.

Preferably, the second defect detector may classify the scrap and the scale through a learned artificial neural network or a support vector machine based on the third characteristic parameter of the second residual image.

Preferably, the third characteristic parameter may be texture information using a histogram of the second residual image, roughness of perimeter (RoP), and shape information.

According to another aspect of the present invention, there is provided a method of generating a binary image, the method comprising: generating an original image by photographing a surface of a thick plate; generating a filtered image by filtering the original image; generating a first binarized image by binarizing the filtered image; Generating a second binarized image of the original image by binarizing an image of the corresponding column based on a reference value of a brightness value of the column for each column of the original image, Generating a defect candidate binarized image by combining the first binarization image and the second binarization image, and classifying at least two second defects for each blob of the binarized image and detecting stepwise.

According to one embodiment of the present invention, it is possible to classify and detect defects for which no pattern appears by cross illumination, by detecting oil defects, chip defects, scrap and scale in stages in the order of increasing feature values, Effect.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing a heavy plate surface defect inspection apparatus according to a preferred embodiment of the present invention;
2 is a view showing a surface photographing unit,
3 is a diagram showing a first defect detecting section.
FIG. 4 is a view showing a process in which the surface defects of the thick plate are sorted and detected step by step,
5 is a view showing a second defect detecting portion,
6 is a photograph showing an oil defect.
Fig. 7 is a diagram showing a circular judgment distance of the blob.
8 shows eight first mapping values arranged in a matrix around the center of gravity G of the block blob,
Figure 9 illustrates bit operators corresponding to the eight first mapping values of Figure 7;
10 illustrates a state in which first mapping values are rotated around a center of gravity G of a block blob,
11 is a flowchart illustrating a method of detecting a surface defect of a thick plate according to a preferred embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS The objectives, specific advantages, and novel features of the present invention will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, in which: FIG. The terms and words used in the present specification and claims should not be construed to be limited to ordinary or dictionary terms and the inventor should appropriately define the concept of terms in order to describe his invention in the best way possible The present invention should be construed in accordance with the meaning and concept consistent with the technical idea of the present invention. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail.

1 is a block diagram showing a heavy plate surface defect inspection apparatus according to a preferred embodiment of the present invention.

1, a thick plate surface defect inspection apparatus 100 according to a preferred embodiment of the present invention includes a surface image pickup unit 110, a first defect detection unit 120, and a second defect detection unit 130 .

2 is a view showing the surface photographing unit.

1 and 2, the surface image pickup unit 110 may include a camera 111, a first illumination light 112, and a second illumination light 113. FIG. The first illumination 112 and the second illumination 113 are symmetrically disposed on the left and right sides of the camera 111 located at the widthwise center C of the thick plate 1, .

The surface image pickup unit 110 photographs the surface of the thick plate 1 in a state where the first illumination 112 is on and the second illumination 113 is off, And the second illumination 113 can take an image of the surface of the thick plate 1 in an on state to generate an original image. Here, in the case where the concavo-convex defect A having the concave depth exists, the pixel values of the images photographed under the two conditions described above differ from each other.

3 is a diagram showing a first defect detecting section.

The first defect detecting unit 120 detects the irregular defect using a pattern generated due to cross illumination. The first defect detector 120 may include a filtering unit 121 and a first binarization unit 122, as shown in FIG.

The filtering unit 121 may generate a filtered image by filtering an original image and performing Gabor filtering. For duty-free cracks, Gabor filtering can be used to exploit this characteristic, since the direction of the crossed pattern is vertical and black lines and white lines appear. At this time, the energy matrix can be obtained by taking the square of each pixel value and converting the minus value to a positive value.

The first binarization unit 122 may generate a first binarized image by Gaussian filtering the squared filtered image. At this time, the first binarization unit 122 may binarize by setting a double-thresholding. Binarization refers to the use of two values, 0 and 1, for color values that are widely distributed over a pixel. That is, only one candidate group that may correspond to a surface defect in the original image can be expressed as 1, and the remaining part can be expressed as 0. The part represented by 1 is represented by white, and the part represented by 0 is represented by black.

FIG. 4 is a view showing a process in which the surface defects of the thick plate are sorted and detected step by step, and FIG. 5 is a view showing the second defect detecting portion.

As shown in FIG. 4, the second defect detector 130 serves to detect an oil defect, a chip defect, a scrap, and a scale that do not show a pattern by cross illumination step by step.

The second defect detector 130 may include a second binarizer 131 and a defect candidate detector 132, as shown in FIG.

The second defect detector 130 can detect a defect by generating a vertical profile. The vertical profile is an average of the pixel values included in each column in the vertical direction of the original image. Specifically, the second binarization unit 131 may generate a second binarized image of the original image by binarizing the image of the corresponding column based on the reference value for the brightness value of the corresponding column for each column of the pixels.

Also, as the vertical profile, the first boundary value and the second boundary value can be obtained by obtaining the standard deviation of the average of the pixel values included in each column in the longitudinal direction of the original image. A first boundary value may be used to detect a bright visible defect in the image, and a second boundary value may be used to detect a dark visible defect.

The second binarization unit 131 may correspond to 1 if it is greater than or equal to the first boundary value and correspond to -1 if the second boundary value is less than or equal to the second boundary value.

The defect candidate detecting unit 132 may generate a defect candidate binarized image by combining the first binarized image and the second binarized image.

The second defect detector 130 may classify and detect the oil defect, the chip defect, the scrap, and the scale step by step in the defect candidate binarized image.

6 is a photograph showing an oil defect.

First, as shown in Fig. 6, the oil defect can be detected by extracting a feature whose shape is similar to a circle, the difference between the background and the pixel value is large, and the change in the pixel value is small within the oil defect.

Specifically, the second defect detector 130 can classify the oil defect and the first residual image through the artificial neural network or the support vector machine learned based on the first characteristic parameter of the defect candidate binarized image have.

Here, the first characteristic parameter may be a texture information using a histogram of a defect candidate binarized image, a morphological feature, a shape variance, and a rotation invariant value.

The texture information using the histogram may be Mean, Standard Deviation, Smoothness, Uniformity, Entropy. The morphological features can be the oil mark's extent ratio, solidity, elongation, compactness (circumference), compactness (area) have.

Fig. 7 is a diagram showing a circular judgment distance of the blob.

Referring to FIG. 7, the shape variation value may be a variance value of the circular determination distance d1 from the center of gravity G of the block blob to the edge of the block blob in eight directions. The smaller the variance value of the circular judgment distance d1 is, the closer to a circle.

The rotation invariant value may be a feature value calculated by the following procedure.

8 is a view showing eight first mapping values arranged in a matrix around the center of gravity G of a block blob, and FIG. 9 is a diagram showing a bit map Fig.

First, an average value of the circular judgment distance d1 is calculated based on eight directions symmetric about the center of gravity G of the block blob. If the average value is larger than the calculated average value, The first mapping values may be calculated in eight directions around the center of gravity G as shown in FIG. 8, and arranged in a matrix (A) form.

As shown in FIG. 9, the second defect detector 130 may generate a bit operator B corresponding to the eight first memory values.

The second defect detector 130 may calculate the second mapping values by multiplying the first mapping value by each bit operator. For example, the re-2 matching value 0 can be calculated by multiplying the first mapping value A1 of A1 in Fig. 9 by 1 and the bit operator 1 of B1 in Fig. In the clockwise direction, all eight second matching values such as 0, 2, 0, 8, 16, 0, and 128 can be calculated.

The feature value sum1 can be calculated by summing all the second matching values.

For example, the feature value (sum1) = 000 + 002 + 000 + 008 + 016 + 000 + 000 + 128 = 154

10 is a view showing a state in which the first mapping values are rotated about the center of gravity G of the block blob.

Referring to FIG. 10, a matrix A in which first mapping values are arranged in eight directions is stepwise rotated around a center of gravity G to multiply corresponding bit operators to calculate a second mapping value, The sum of the mapping values can be used to calculate the feature values sum2, sum3, ..., sum8.

E.g,

Feature value (sum1) = 000 + 002 + 000 + 008 + 016 + 000 + 000 + 128 = 154

Feature value (sum2) = 001 + 000 + 004 + 000 + 016 + 032 + 000 + 000 = 53

Feature value (sum3) = 000 + 002 + 000 + 008 + 000 + 032 + 064 + 000 = 106

.

Feature value (sum8) = 001 + 000 + 004 + 008 + 000 + 000 + 064 + 000 = 77

At this time, the rotation invariant can be determined as the minimum value 53 (sum2) of the feature values. The rotation invariant may be a maximum value among the feature values.

Next, the second defect detector 130 classifies the chip defect and the second residual image through the artificial neural network or the support vector machine learned based on the second characteristic parameter of the first residual image can do.

Here, the second characteristic parameter may be a Gray-Level Co-Occurrence Matrix (GLCM) of the first residual image, an average energy of the Gabor filter, an edge direction, a morphological characteristic, or a statistical feature. In the case of a chip defect, a narrow and long line-shaped defect is formed. Since the crossing pattern is generated due to the irregularities, the crossing pattern can be quantified and the average energy of the Gabor filter in the defect region can be used. In addition, the number of frequencies along the edge direction can be calculated and utilized.

The statistical feature can be obtained by obtaining the difference between the odd image and the even image and quantizing the mean and variance within the defective area.

Then, the second defect detector 130 can classify the second defect image into scraps and scales through the learned artificial neural network or the support vector machine based on the third characteristic parameter of the second residual image.

Scrap is a defect that occurs when pressure is applied in the state of foreign matter, and there are concavities and convexities, and a cross pattern is generated.

Here, the third characteristic parameter may be texture information using a histogram of the second residual image, roughness of perimeter (RoP), and shape information. Here, RoP is a measure for determining the roughness of the periphery.

11 is a flowchart illustrating a method of detecting a surface defect of a thick plate according to a preferred embodiment of the present invention.

Referring to FIG. 11, in the method of detecting a surface defect of a thick plate according to an exemplary embodiment of the present invention, a surface image capturing unit 110 may generate an original image by photographing a surface of a thick plate.

Next, the first defect detector 120 generates a filtered image by filtering the original image, generates a first binarized image by binarizing the filtered image, and detects an oil defect in the first binarized image (S200 )

Next, the second defect detector 130 generates a second binarized image of the original image by binarizing the image of the corresponding column based on the reference value of the brightness value of the corresponding column for each column of the original image, The defect candidate binarization image may be generated by combining the first and second binarization images, and at least two second defects may be classified for each blob of the binarized image and detected step by step.

The apparatus and method for detecting surface defects of a heavy plate according to one preferred embodiment of the present invention have been described in detail with reference to the accompanying drawings.

It will be apparent to those skilled in the art that various modifications, substitutions and substitutions are possible, without departing from the scope and spirit of the invention as disclosed in the accompanying claims. will be. Therefore, the embodiments disclosed in the present invention and the accompanying drawings are intended to illustrate and not to limit the technical spirit of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments and the accompanying drawings . The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents should be construed as falling within the scope of the present invention.

100: Plate surface defect inspection device
110:
111: camera
112: 1st illumination
113: Second illumination
120: first defect detector
121: Filtering section
122: first binarization unit
130: second defect detector
131: second binarization unit
132: defect candidate detector

Claims

A surface image photographing unit for photographing the surface of the thick plate to generate an original image;
A first binarization unit that binarizes the filtered image to generate a first binarized image, and a first binarization unit that binarizes the filtered image to generate a first binarized image, A detection unit;
Calculating a first boundary value by adding a standard deviation to the calculated average value and calculating a mean value and a standard deviation of the brightness value of the corresponding column for each column of the original image; calculating a value obtained by subtracting the standard deviation from the calculated average value; 2 if the brightness value of the corresponding pixel of the original image is equal to or greater than the first threshold value and to -1 if the brightness value of the corresponding pixel is equal to or less than the second threshold value, And a defect candidate detecting unit for generating a defect candidate binarized image by adding the first binarized image and the second binarized image to each other in units of pixels to form a binarized image, ) For detecting at least two second defects for each of the plurality of defects
Wherein the plate surface defect inspection device comprises:

The method according to claim 1,
Wherein the surface imaging unit comprises a first illumination and a second illumination that illuminate the first point at a location different from the camera that takes a first point of the slab surface, wherein only one of the first illumination and the second illumination Which generates an original image in which a first image photographed in a state of On is combined with a second image photographed in a state where only the other of the first illumination and the second illumination is on, Flaw detection device.

3. The method of claim 2,
Wherein the filtering unit generates the filtering image by performing Gabor filtering on the original image.

The method of claim 3,
Wherein the first binarization unit obtains an energy matrix from the filtered image and performs Gaussian filtering on the energy matrix to generate the first binarized image.

5. The method of claim 4,
Wherein the first binarization unit binarizes the filtered image by setting a double-threshold value to generate the first binarized image.

delete

The method according to claim 1,
The second defect is oil defect, chip defect, and scrap,
Wherein the second defect detector classifies the chip defect and the second residual image in the first residual image after classifying the defect candidate binarized image into the oil defect and the first residual image, Defect plate surface defect inspection system for classifying scales.

9. The method of claim 8,
Wherein the second defect detector classifies the oil defect and the first residual image through a learned artificial neural network or a support vector machine based on the first characteristic parameter of the defect candidate binarized image.

10. The method of claim 9,
Wherein the first characteristic parameter is texture information using a histogram of the defect candidate binarized image, a morphological feature, a shape change value, and a rotation invariance.

11. The method of claim 10,
Wherein the shape change value is a variance value of the circular judgment distance up to the edge of the blob with respect to the center of gravity of the blob.

12. The method of claim 11,
The rotation invariant calculates an average value of the circular judgment distance based on at least two reference directions symmetrical about the center of gravity, and corresponds to 1 when the calculated value is larger than the calculated average value and to 0 when the average value is smaller than the average value Calculating a first mapping value for each reference direction around the center of gravity and arranging the first mapping value in a matrix form to generate a bit operator corresponding to the first mapping value, And calculating the feature value by summing the second mapping values calculated by calculating the second mapping value, and stepwise rotating the bit operator in the reference direction around the center of gravity, And calculating the feature value, and then calculating the feature value.

13. The method of claim 12,
Wherein the rotation invariant is a minimum value or a maximum value of the characteristic values.

9. The method of claim 8,
Wherein the second defect detector classifies a chip defect and a second residual image through a learned artificial neural network or a support vector machine based on a second characteristic parameter of a first residual image.

15. The method of claim 14,
Wherein the second characteristic parameter is a gray-level Co-Occurrence Matrix (GLCM) of the first residual image, an average energy of the Gabor filter, an edge direction, a morphological characteristic, and a statistical characteristic.

9. The method of claim 8,
Wherein the second defect detector classifies the scrap and the scale through a learned artificial neural network or a support vector machine based on a third characteristic parameter of a second residual image.

17. The method of claim 16,
Wherein the third characteristic parameter is texture information, a roughness of perimeter (RoP), and shape information using a histogram of a second residual image.

Photographing the surface of the thick plate to generate an original image;
Generating a filtered image by filtering the original image, generating a first binarized image by binarizing the filtered image, and detecting a first defect in the first binarized image;
Calculating a first boundary value by adding a standard deviation to the calculated average value and calculating a mean value and a standard deviation of the brightness value of the corresponding column for each column of the original image; calculating a value obtained by subtracting the standard deviation from the calculated average value; 2 if the brightness value of the corresponding pixel of the original image is equal to or greater than the first threshold value and to -1 if the brightness value of the corresponding pixel is equal to or less than the second threshold value, 2 binary images, combining the first binarized image and the second binarized image for each pixel to generate a defect candidate binarized image, classifying at least two second defects for each blob of the binarized image, Stepwise detection
And the surface defect detection method.