JP6035375B1 - Defect inspection apparatus and defect inspection method - Google Patents

Abstract

A defect detection apparatus and a defect inspection method that can determine a feature amount of a defect without acquiring a two-dimensional image of the entire defect. A defect inspection apparatus includes a line sensor that reads an image of an inspection object, a binarization unit that binarizes data input from the line sensor, and binarization performed by the binarization unit. A run-length encoding unit that performs run-length encoding of data; and a connectivity processing unit that measures a feature quantity of a defect by performing connectivity processing on the run-length code. [Selection] Figure 1

Description

  The present invention is a defect inspection apparatus and a defect inspection method using a line sensor for inspecting defects existing in an inspection object such as a film that continuously runs.

Conventionally, in order to classify the shape of a defect included in a film, after detecting a defect and acquiring a two-dimensional image, binarization is performed, and the feature amount of the defect (peripheral length of the defect, area, width, length, (Aspect ratio, area ratio). The perimeter is obtained by tracing the contour of the boundary portion of each connected component, and the circularity (4πS / L ² ) and complexity (L ² / 4πS) are obtained from the area (S) and the perimeter (L). (See Patent Document 1).

FIG. 6 is a diagram showing a conventional technique for measuring the perimeter. FIG. 6 shows a technique for measuring the perimeter by contour tracking, which is known as an image processing technique. That is, FIG. 6 shows a technique for obtaining a perimeter (L) using a conventional technique for obtaining a two-dimensional image by detecting a defect and performing binarization to obtain a feature amount of the defect.
(Procedure 1) First, raster scanning is performed from the upper left to find the first pixel of the defect.
(Procedure 2) A contour portion is detected counterclockwise using the first pixel of the defect as a start point.
(Procedure 3) Since the first pixel of the defect was raster scanned from the upper left, it is determined that there are no defective pixels at the upper right, upper, upper left, and left of the start point, so first counterclockwise from the lower left part If a defective pixel is found, the search continues to the next contour.
(Procedure 4) The process ends when returning to the start point at the end.
(Procedure 5) The perimeter is counted as 1 in the vertical direction and √2 in the diagonal direction.
In this way, the peripheral length (L) is obtained as 6 + 10√2 = 20.14 pixels.

JP2013-160745A

In the prior art, after obtaining a two-dimensional image, the feature amount of the defect is obtained from the stored two-dimensional image, and thus the following problems have occurred.
(Problem 1) Problem related to measurement size Only defects smaller than the size of a two-dimensional image can be measured. For example, when the size of the two-dimensional image is 256 × 256 pixels, a defect having a size larger than this cannot be measured.
(Problem 2) Problem related to measurement speed Since a two-dimensional image has a large file capacity, it takes time to acquire it.
(Problem 3) Problems related to the number of measurements Since the file capacity of a two-dimensional image is large, there is a limitation on the number of defect images that can be stored in one inspection. For example, in the MEC inspection device LSC6000, the standard specification has 5000 images.
(Problem 4) Problems related to individual recognition When there are a plurality of defects in a two-dimensional image, it becomes a problem which defect should be measured.
(Problem 5) Resolution problem It was assumed that the width resolution and the flow resolution were the same in order to count as the number of pixels. In the case of sheet inspection that runs continuously, the width resolution and the flow resolution are often different. For example, when the width resolution / flow resolution is 0.5, a circular ellipse having an aspect ratio of 2 is closer to 1 than a circle.

  The present invention has been made in view of the above-described problems, and provides a defect detection apparatus and a defect inspection method capable of obtaining a feature amount of a defect without acquiring a two-dimensional image of the entire defect. That is.

In order to solve the above problems, a defect inspection apparatus according to the present invention includes a line sensor that reads an image of an inspection object, a binarization unit that binarizes data input from the line sensor, and the binary value. and run-length encoding unit for run length encoding the binarized data by the unit, by connectivity processing the run-length code, possess a connectivity processing unit for measuring the feature amounts of the defect, the , P is the perimeter of the nth row, Pa is the sum of the perimeters up to the (n−1) th row, Pa ′ is the right perimeter of the (n−1) th row, and Pb is the right of the nth row. Perimeter, T is the number of pixels in contact with the defect data in the (n−1) th row and the nth row, bs is the start pixel of the defect data in the nth row, and be is the end pixel of the defect data in the nth row. The connectivity processing unit is obtained on a line adjacent to the traveling direction of the inspection object. Based on the following calculation formulas (1) to (4) corresponding to the relationship of defects to be generated, (1) In the case of the generated pattern, P = 2 (be-bs), and (2) in the case of the continuous pattern, P = Pa + P, (3) P = P + Pa ′ in the case of the convergence pattern, (4) P = P + Pb in the case of the dispersion pattern. However, in the above formulas (2) to (4), the width direction is further When the start point side and the end point side are different, P = P−2T, when either one is the same position, P = P− (2T−1), and when both are the same position, P = P− ( 2T-2), and the flow direction is P = P + 1 when the defect pattern is at the same position on the start point side and the end point side, and P = P + √2 when the defect pattern is slanted. The feature is to obtain a perimeter length P that is a feature amount .

In order to solve the above-described problems, the defect inspection method of the present invention includes a reading step in which a line sensor reads an image of an inspection object, and a binarization unit binarizes data input from the line sensor. A binarizing step, a run-length encoding unit that performs run-length encoding on the data binarized by the binarizing unit, and a connectivity processing unit that converts the run-length code by treating connectivity, surrounding possess a connection processing step of measuring the characteristic of the defect, a circumferential length of P, the peripheral length of the n-th row, the Pa (n-1) to row P The total length, Pa ′ is the peripheral length on the right side to the (n−1) th line, Pb is the peripheral length on the right side of the nth line, and T is the defect data on the (n−1) th line and the nth line. The number of pixels, bs is the starting pixel of the defect data in the nth row, and be is the defect data in the nth row. Based on the following calculation formulas (1) to (4) according to the relationship of defects obtained in the line adjacent to the traveling direction of the inspection object, 1) P = 2 (be-bs) for the generated pattern, (2) P = Pa + P for the continuous pattern, (3) P = P + Pa ′ for the convergent pattern, and (4) the dispersion pattern In this case, P = P + Pb. However, in the above formulas (2) to (4), the width direction is P = P−2T when the start point side and the end point side are different, and either one is the same. In the case of the position, P = P− (2T−1), and in the case where both are the same position, P = P− (2T−2). The flow direction is the start point side and the end point side, and the defect pattern is the same position. In this case, P = P + 1, and if it is slanted, P = P + √2. Thus, the perimeter length P which is the feature amount of the defect is obtained .

According to the present invention, the above-described problems can be solved as follows by obtaining the feature amount of a defect by connectivity processing without acquiring a two-dimensional image of the entire defect.
(Problem 1) Problem concerning measurement size The width direction is the range of the number of elements of the line sensor. There are 2048-8192 elements as the line sensor to be adopted. Since the flow direction is counted by 64 bits, there is almost no practical limitation.
(Problem 2) Measurement speed problem Since large-capacity image storage is not performed, high-speed processing is possible.
(Problem 3) Problems related to the number of measurements When a large-capacity image is not stored, a large number of defects can be measured. For example, in the MEC inspection device LSC6000, the standard specification is 200,000.
(Problem 4) Problems related to individual recognition Since measurement is performed simultaneously with individual recognition, it is clear which defect is detected.
(Problem 5) Resolution problem It is possible to cope with the case where the width resolution and the flow resolution are different.
Therefore, it is possible to provide a defect detection apparatus and a defect inspection method that can obtain the feature amount of a defect without acquiring a two-dimensional image of the entire defect.

It is a figure which shows one Example of the test | inspection apparatus of this invention. It is a figure for demonstrating the content of the connectivity process currently performed in the 1st defect detection part. It is a figure which shows the circumference measurement method of this invention. It is a figure which shows the process sequence of the defect pattern shown in FIG. It is a figure which shows the case where width / flow resolution differs. It is a figure which shows the prior art which measures perimeter.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
[Configuration of defect inspection equipment]
FIG. 1 is a diagram showing an embodiment of an inspection apparatus according to the present invention.
The defect inspection apparatus according to the present embodiment includes a line illumination 1, a line sensor 3, a first defect detection unit 4, an image extraction unit 5, and a second defect detection unit 6.
The defect inspection apparatus according to the present embodiment illuminates the traveling sheet-like inspection object 2 with the line illumination 1, captures an image with the line sensor 3, and performs defect detection with the first defect detection unit 4. Further, an image centering on the defect is extracted by the image extraction unit 5 based on the detected defect coordinates, and a detailed defect measurement is performed by the second defect detection unit 6 based on the image data, thereby performing defect classification. .

  The line illumination 1 is a line-shaped illumination device arranged in a direction in which the longitudinal direction of the light irradiation range on the inspection object 2 is orthogonal to the traveling direction of the inspection object 2.

  The inspection object 2 is a continuous sheet-like film, a metal plate, a cloth, a nonwoven fabric, a resin plate, and the like. In the present embodiment, the film will be described below as an example. The inspection object 2 is preferably long in the conveying direction, and it is preferable to inspect the fixed line sensor 3 while traveling the inspection object 2 in the length direction.

The line sensor 3 is a line-shaped optical sensor arranged in a direction in which the longitudinal direction of the imaging range is orthogonal to the traveling direction of the inspection object 2.
The line sensor 3 receives reflected light or transmitted light from the inspection object 2 and supplies an electrical signal (defect data) corresponding to the color tone of the surface of the inspection object 2 to the first defect detection unit 4. . In other words, the line sensor 3 outputs an electrical signal corresponding to the light intensity distribution on the surface of the inspection object 2 in units of lines (hereinafter referred to as inspection lines) in a direction orthogonal to the traveling direction of the inspection object 2. . Examples of the line sensor 3 include a CMOS camera and a CCD camera.
For example, the line sensor 3 having 2048 to 8192 elements is used. The number and number of elements used as the inspection apparatus are selected according to the width of the inspection object, the traveling speed, the resolution, the installation space, and the like.
In the illustrated example, the line sensor 3 is arranged on the surface side of the inspection object 2 in the same manner as the line illumination 1 and receives reflected light from the inspection object 2. However, the line sensor 3 may receive light transmitted from the line illumination 1 and transmitted through the inspection object 2.

The first defect detection unit 4 includes a binarization unit 11, a run length encoding unit 12, and a connectivity processing unit 13.
The binarization unit 11 binarizes defect data based on a first threshold th1 (bright → dark) and a second threshold th2 (dark → bright) set by a threshold setting unit (not shown).
The threshold value setting unit has a first threshold value th1 (bright → dark) and a second threshold value th2 (dark → bright) from the outside in advance of defect inspection according to the material, thickness, and the like of the inspection object 2. Is set.
There are two types of detection logic in the binarization unit 11. When detecting a bright defect, the binarization unit 11 sets an area having an output level greater than the second threshold th2 to “1”, and when detecting a dark defect, the binarization unit 11 selects an area having a level lower than the first threshold th1. 1 ”. That is, as a result of binarization by the binarization unit 11, the region corresponding to the defective portion is “1” and the region corresponding to the normal portion is “0”.

The run-length encoding unit 12 obtains the address of the change point from the binary data “0” to “1” or “1” to “0” of the binary data by the binarization unit 11. Run-length encoding.
The connectivity processing unit 13 performs connectivity processing on the run-length code that has been run-length encoded by the run-length encoding unit 12. Here, the connectivity processing refers to processing while comparing defect data in a plurality of continuous scanning lines between the lines. By performing this connectivity processing, it is possible to recognize the defect of the inspection object and measure the characteristics caused by the defect form.

  With the configuration described above, the first defect detection unit 4 binarizes the defect data input from the line sensor 3 with a predetermined threshold value, compresses the defect data, performs connectivity processing, and performs defect processing. A feature amount (information indicating the feature of the defect shape. For example, the peripheral length, area, width, length, aspect ratio, and area ratio of the defect) is measured. In the present invention, the perimeter of the defect is measured at this stage to obtain the circularity and complexity. As the first defect detection unit 4, an image processing apparatus LSC600 manufactured by MEC Co., Ltd. can be used.

The image extraction unit 5 extracts an image centered on the defect based on the defect coordinates detected by the first defect detection unit 4.
The second defect detection unit 6 performs detailed defect measurement based on the image data extracted by the image extraction unit 5, and performs defect classification. As the second defect detection unit 6, for example, a Windows (registered trademark) PC can be used, and an image processing library HALCON (registered trademark) manufactured by MVTec can be used as described in Patent Document 1.

FIG. 2 is a diagram for explaining the contents of the connectivity process performed by the first defect detection unit. As shown in FIG. 2, the defect data for two consecutive rows is classified into five types of patterns (occurrence pattern, continuous pattern, convergence pattern, dispersion pattern, disappearance pattern).
(1) Occurrence pattern This is a pattern in which the defect data in the (n-1) th row is not in contact with the defect data generated in the nth row. From here, this defect is measured.
(2) Continuous pattern The defect data in the (n-1) th row is in contact with the defect data generated in the nth row. Here, both data are added.
(3) Convergence pattern A plurality of defect data (two in FIG. 2) in the (n-1) th row is in contact with one defect data generated in the nth row. Here, a plurality of (n−1) th row data (second in FIG. 2) data and the nth row data are added.
(4) Dispersion pattern One defect data in the (n-1) th row is in contact with a plurality of defect data (two in FIG. 2) generated in the nth row. (N-1) The left data in the nth row added to the data in the row and the right data in the nth row are added.
(5) Annihilation pattern One defect data in the (n-1) th row is not in contact with the defect data in the nth row. Here, the measurement of this defect ends.

FIG. 3 is a diagram showing a method for measuring the perimeter of the present invention. FIG. 3 is a diagram for explaining a calculation method for measuring the perimeter of each pattern described in FIG.
(1) Generation Pattern In the case of the start point pixel bs and end point pixel be of the defect data on the nth row, the perimeter length P (pixel) for one row is obtained by the following equation.
P = 2 (be-bs)
The peripheral length P (mm) is obtained by the following equation in the case of the width direction resolution H (mm / pixel).
P = 2 (be-bs) H
(2) Continuous pattern (n-1) Total perimeter Pa up to the line and total perimeter P of the nth line are summed.
(N-1) In the case of the number T of pixels in which the defect data in the row and the n-th row are in contact, the width direction is 2T when the start point side / end point side is different, and when either one is the same position (2T-1), if both are at the same position, (2T-2) is subtracted. When the flow direction is the start point side and the end point side and the defect patterns are at the same position, 1 is added. If it is slanted, add √2.
In addition, in the mm unit, when the flow direction resolution is V (mm / pixel), V is added when the defect pattern is at the same position. If it is slanted, (√ (H2 + V2)) is added.
(3) Convergence pattern (n-1) Summing the perimeter length Pa on the left side of the row and the perimeter length P of the nth row has been processed. (N-1) The perimeter length Pa ′ on the right side of the row and the perimeter length P of the nth row are summed.
When the defect pattern is at the same position on the end point side, 1 is added. There is no case where the defect pattern is at the same position on the start point side. If it is slanted, add √2.
In addition, in the mm unit, when the flow direction resolution is V (mm / pixel), V is added when the defect pattern is at the same position. Further, if it is slanted, (√ (H2 + V2)) is added.
(4) Dispersion pattern (n-1) Summing up the perimeter length Pa up to the row and the perimeter length P on the left side of the nth row has been processed. The peripheral length P up to the left side of the nth row and the peripheral length Pb on the right side of the nth row are summed.
When the defect pattern is at the same position on the end point side, 1 is added. There is no case where the defect pattern is at the same position on the start point side. If it is slanted, add √2.
In addition, in the mm unit, when the flow direction resolution is V (mm / pixel), V is added when the defect pattern is at the same position. If it is slanted, (√ (H2 + V2)) is added.

FIG. 4 is a diagram showing a processing procedure of the defect pattern shown in FIG. In FIG. 4, the defect patterns generated in the first to eighth rows in the traveling direction are represented by steps (1) to (10). FIG. 4 includes all the patterns shown in FIG.
[Step (1)]
Step (1) is generation.
Since the defect data in the second row is one pixel, the perimeter is 0 pixel.
[Step (2)]
Step (2) is continuous with step (1).
Since the defect data in step (2) is 2 pixels, the peripheral length is 2 pixels.
Since the defect data in step (1) is 0 pixel and the defect data in step (2) is 2 pixels, the total perimeter is 2 pixels. The portion where the second row (second row and third row) is in contact is 1 (= T) pixels, and one of them is at the same position, so the subtraction is 1 (= 2T-1) pixels. Since the starting point side is diagonal, √2 pixels are added. Since the end point is vertical, one pixel is added. As a result, the peripheral length becomes 2 + √2 (= 2-1 + √2 + 1) pixels by step (2).
[Step (3)]
Step (3) is generation.
Since the defect data in step (3) is 1 pixel, the peripheral length is 0 pixel.
[Step (4)]
Step (4) is continuous with step (2).
Since the defect data in step (4) is 4 pixels, the perimeter is 6 pixels.
Since the defect data in step (2) is 2 + √2 pixels and the defect data in step (4) is 6 pixels, the total perimeter is 8 + √2 pixels. Since the portion where the second row (the third row and the fourth row) is in contact is 2 (= T) pixels, and the start point side / end point side are different positions, the subtraction is 4 (= 2T) pixels. Since the starting point side is diagonal, √2 pixels are added. Further, since the end point side is oblique, √2 pixels are added. As a result, in step (4), the perimeter becomes 4 + 3√2 (= 2 + √2 + 6−4 + √2 + √2) pixels.
[Step (5)]
Step (5) is convergence with step (3).
The sum of the perimeter of the left side of the third row and the perimeter of the fourth row has been processed.
The peripheral length 0 pixels on the right side of the third row and the peripheral length (4 + 3√2) pixels of the fourth row are summed, and the total value of the peripheral lengths of step (5) and step (3) is 4 + 3√2 pixels It is.
Since both the start point side and the end point side are diagonal, 2√2 pixels are added. As a result, the perimeter becomes 4 + 5√2 (= 4 + 3√2 + 2√2) pixels in step (5).

[Step (6)]
Step (6) is continuous with step (5).
Since the defect data in step (6) is 5 pixels, the perimeter is 8 pixels.
Since the defect data in step (5) is 4 + 5√2 pixels and the defect data in step (6) is 8 pixels, the total perimeter is 12 + 5√2 pixels. The portion where the second row (the fourth row and the fifth row) is in contact is 4 (= T) pixels, and one of them is at the same position, so the subtraction is 7 (= 2T-1) pixels. Since the start point side is vertical, one pixel is added. Further, since the end point side is oblique, √2 pixels are added. As a result, in step (6), the perimeter becomes 6 + 6√2 (= 4 + 5√2 + 8−7 + 1 + √2) pixels.
[Step (7)]
Step (7) is continuous with step (6).
Since the defect data in step (7) is 6 pixels, the peripheral length is 10 pixels.
Since the defect data in step (6) is 6 + 6√2 pixels and the defect data in step (7) is 10 pixels, the total perimeter is 16 + 6√2 pixels. The portion where the second row (the fifth row and the sixth row) is in contact is 5 (= T) pixels, and one of them is at the same position, so the subtraction is 9 (= 2T-1) pixels. Since the starting point side is diagonal, √2 pixels are added. Since the end point is vertical, one pixel is added. Thus, the peripheral length becomes 8 + 7√2 (= 6 + 6√2 + 10−9 + √2 + 1) pixels by the step (7).

[Step (8)]
Step (8) is continuous with step (7).
Since the defect data in step (8) is 2 pixels, the peripheral length is 2 pixels.
Since the defect data in step (7) is 8 + 7√2 pixels and the defect data in step (8) is 2 pixels, the total perimeter is 10 + 7√2 pixels. The portion where the second row (the sixth row and the seventh row) is in contact is 2 (= T) pixels, and since the start point side / end point side are different positions, the subtraction is 4 (= 2T) pixels. Since the starting point side is diagonal, √2 pixels are added. Further, since the end point side is oblique, √2 pixels are added. Thereby, the perimeter becomes 6 + 9√2 (= 8 + 7√2 + 2 + 4 + √2 + √2) pixels by step (8).
[Step (9)]
Step (9) is a variance with step (8).
Summing the total perimeter up to the sixth line and the left perimeter of the seventh line has been processed.
The peripheral length of 6 + 9√2 pixels to the left side of the seventh row and the peripheral length of 0 pixels on the right side of the seventh row are totaled, and the peripheral length is 6 + 9√2 pixels. Further, since the starting point side is oblique, √2 pixels are added. Since the end point is vertical, one pixel is added. Thereby, the perimeter becomes 6 + 10√2 pixels by step (9).

[Step (10)]
Step (10) is extinction.
The perimeter of step (10) is 6 + 10√2 pixels as in the prior art shown in FIG.

That is, the first defect detection unit 4 according to the present embodiment calculates the defect data for each line by the connectivity processing unit 13 without acquiring the two-dimensional defect image of the entire defect, thereby obtaining the feature amount of the defect. The perimeter (L) can be measured.
The connectivity processing unit 13 adds the defect data in steps (1) to (10) to 0 + 1, 1 + 2, 3 + 1, 4 + 4, 8 + 5, 13 + 6, 19 + 2, and 21 + 1 in order for each line. It is possible to measure the area (S) = 22 pixels, which is the feature amount of the defect. Thereby, the connectivity processing unit 13 can calculate the circularity 4πS / L ² = 0.68.
As described above, the connectivity processing unit 13 calculates the defect feature amount (peripheral length, area, etc.) based on the calculation formula corresponding to the defect relationship obtained in the line adjacent to the traveling direction of the inspection object 2. The processing to be performed is sequentially performed in the traveling direction.

FIG. 5 is a diagram showing a case where the width / flow resolution is different.
FIG. 5A shows a case where a circular defect is read at (width resolution / flow resolution = 1), and FIG. 5B shows the same circular defect (width / flow resolution = 0.0). 5) shows the case of reading.
Hereinafter, the processing procedure of the defect pattern shown in FIGS. 5A and 5B will be described.
(Defect pattern processing procedure shown in FIG. 5A)
In FIG. 5A, the defect patterns generated in the first to twelfth lines in the traveling direction are represented by steps (1) to (11).
[Step (1)]
Step (1) is generation.
Since the defect data in the second row is 4 pixels, the peripheral length is 6 pixels.
[Step (2)]
Step (2) is continuous with step (1).
Since the defect data in step (2) is 8 pixels, the peripheral length is 14 pixels.
Since the defect data in step (1) is 6 pixels and the defect data in step (2) is 14 pixels, the total perimeter is 20 pixels. The portion where the second row (the second row and the third row) is in contact is 4 (= T) pixels, and the start point side / end point side is a different position, so the subtraction is 8 (= 2T) pixels. Since both the starting point side and the ending point side are diagonal, 2√2 pixels are added. As a result, the peripheral length becomes 12 + 2√2 (= 6 + 14−8 + 2√2) pixels by the step (2).
[Step (3)]
Step (3) is continuous with step (2).
Since the defect data in step (3) is 8 pixels, the peripheral length is 14 pixels.
Since the defect data in step (2) is 12 + 2√2 pixels and the defect data in step (3) is 14 pixels, the total perimeter is 26 + 2√2 pixels. Since the portion where the second row (the third row and the fourth row) is in contact is 8 (= T) pixels and both are 8 (= T) pixels, subtraction is 14 (= 2T-2) pixels. It is. Since both the start point side and the end point side are vertical, two pixels are added. As a result, the perimeter becomes 14 + 2√2 (= 12 + 2√2 + 14-14 + 2) pixels in step (3).

[Step (4)]
Step (4) is continuous with step (3).
Since the defect data in step (4) is 10 pixels, the perimeter is 18 pixels.
Since the defect data in step (3) is 14 + 2√2 pixels and the defect data in step (4) is 18 pixels, the total perimeter is 32 + 2√2 pixels. The portion where the second row (the fourth row and the fifth row) is in contact is 8 (= T) pixels, and since the start point side / end point side are different positions, the subtraction is 16 (= 2T) pixels. Since both the starting point side and the ending point side are diagonal, 2√2 pixels are added. As a result, the perimeter becomes 16 + 4√2 (= 14 + 2√2 + 18−16 + 2√2) pixels by step (4).
[Step (5)]
Step (5) is continuous with step (4).
Since the defect data in step (5) is 10 pixels, the perimeter is 18 pixels.
Since the defect data in step (4) is 16 + 4√2 pixels and the defect data in step (5) is 18 pixels, the total perimeter is 34 + 4√2 pixels. Since the portion where the second row (the fifth row and the sixth row) is in contact is 10 (= T) pixels, and both are 10 (= T) pixels, the subtraction is 18 (= 2T-2) pixels. It is. Since both the start point side and the end point side are vertical, two pixels are added. Thus, the peripheral length becomes 18 + 4√2 (= 16 + 4√2 + 18−18 + 2) pixels by the step (5).
[Step (6)]
Step (6) is continuous with step (5).
Since the defect data in step (6) is 10 pixels, the peripheral length is 18 pixels.
Since the defect data in step (5) is 18 + 4√2 pixels and the defect data in step (6) is 18 pixels, the total perimeter is 36 + 4√2 pixels. Since the portion where the second row (the sixth row and the seventh row) is in contact is 10 (= T) pixels, and both are 10 (= T) pixels, the subtraction is 18 (= 2T-2) pixels. It is. Since both the start point side and the end point side are vertical, two pixels are added. Thus, the perimeter becomes 20 + 4√2 (= 18 + 4√2 + 18−18 + 2) pixels by step (6).
[Step (7)]
Step (7) is continuous with step (6).
Since the defect data in step (7) is 10 pixels, the perimeter is 18 pixels.
Since the defect data in step (6) is 20 + 4√2 pixels and the defect data in step (7) is 18 pixels, the total perimeter is 38 + 4√2 pixels. Since the portion where the second row (the seventh row and the eighth row) is in contact is 10 (= T) pixels and both are 10 (= T) pixels, the subtraction is 18 (= 2T-2) pixels. It is. Since both the start point side and the end point side are vertical, two pixels are added. Thereby, the perimeter becomes 22 + 4√2 (= 20 + 4√2 + 18−18 + 2) pixels by step (7).

[Step (8)]
Step (8) is continuous with step (7).
Since the defect data in step (8) is 8 pixels, the peripheral length is 14 pixels.
Since the defect data in step (7) is 22 + 4√2 pixels and the defect data in step (8) is 14 pixels, the total perimeter is 36 + 4√2 pixels. The portion where the second row (the eighth row and the ninth row) is in contact is 8 (= T) pixels, and since the start point side / end point side are different positions, the subtraction is 16 (= 2T) pixels. Since both the starting point side and the ending point side are diagonal, 2√2 pixels are added. As a result, in step (8), the perimeter becomes 20 + 6√2 (= 22 + 4√2 + 14-16 + 2√2) pixels.
[Step (9)]
Step (9) is continuous with step (8).
Since the defect data in step (9) is 8 pixels, the perimeter is 14 pixels.
Since the defect data in step (8) is 20 + 6√2 pixels and the defect data in step (9) is 14 pixels, the total perimeter is 34 + 6√2 pixels. Since the portion where the second row (the ninth row and the tenth row) is in contact is 8 (= T) pixels, and both are 8 (= T) pixels, the subtraction is 14 (= 2T-2) pixels. It is. Since both the start point side and the end point side are vertical, two pixels are added. As a result, the perimeter becomes 22 + 6√2 (= 20 + 6√2 + 14-14 + 2) pixels in step (9).
[Step (10)]
Step (10) is continuous with step (9).
Since the defect data in step (10) is 4 pixels, the perimeter is 6 pixels.
Since the defect data in step (9) is 22 + 6√2 pixels and the defect data in step (10) is 6 pixels, the total perimeter is 28 + 6√2 pixels. The portion where the second row (the tenth row and the eleventh row) is in contact is 4 (= T) pixels, and since the start point side / end point side are different positions, the subtraction is 8 (= 2T) pixels. Since both the starting point side and the ending point side are diagonal, 2√2 pixels are added. As a result, in step (10), the perimeter becomes 20 + 8√2 (= 22 + 6√2 + 6−8 + 2√2) pixels.

[Step (11)]
Step (11) is extinction.
The perimeter of step (11) is 20 + 8√2 pixels.
That is, the first defect detection unit 4 according to the present embodiment calculates the defect data for each line by the connectivity processing unit 13 without acquiring the two-dimensional defect image of the entire defect, thereby obtaining the feature amount of the defect. Perimeter length (L) = 31.313 (= 20 + 8√2) pixels can be measured.
In addition, the connectivity processing unit 13 adds the defect data in steps (1) to (11) to 0 + 4, 4 + 8, 12 + 8, 20 + 10, 30 + 10, 40 + 10, 50 + 10, 60 + 8, 68 + 8, and 76 + 4 in order for each line. By doing so, it is possible to measure the area (S) = 80 pixels, which is the feature amount of the defect.
Thereby, the connectivity processing unit 13 can calculate the circularity 4πS / L ² = 1.025.
That is, in the case of a circle having a diameter of 10 pixels as shown in FIG. 5A, the area is 80 pixels and the peripheral length is 31.312 pixels, so the circularity is 1.025.

(Defect pattern processing procedure shown in FIG. 5B)
In FIG. 5B, the defect patterns generated in the first to seventh rows in the traveling direction are represented by steps (1) to (6).
[Step (1)]
Step (1) is generation.
Since the defect data in the second row is 6 pixels, the perimeter is 10 pixels.
[Step (2)]
Step (2) is continuous with step (1).
Since the defect data in step (2) is 8 pixels, the peripheral length is 14 pixels.
Since the defect data in step (1) is 10 pixels and the defect data in step (2) is 14 pixels, the total perimeter is 24 pixels. The portion where the second row (second row and third row) is in contact is 6 (= T) pixels, and the start point side / end point side is a different position, so the subtraction is 12 (= 2T) pixels. Since both the starting point side and the ending point side are diagonal, 2√2 pixels are added. As a result, the peripheral length becomes 12 + 2√2 (= 14 + 10−12 + 2√2) pixels by the step (2).
[Step (3)]
Step (3) is continuous with step (2).
Since the defect data in step (3) is 10 pixels, the perimeter is 18 pixels.
Since the defect data in step (2) is 12 + 2√2 pixels and the defect data in step (3) is 18 pixels, the total perimeter is 30 + 2√2 pixels. Since the portion where the second row (the third row and the fourth row) is in contact is 8 (= T) pixels, and the start point side / end point side are different positions, the subtraction is 16 (= 2T) pixels. Since both the starting point side and the ending point side are diagonal, 2√2 pixels are added. Thus, the peripheral length becomes 14 + 4√2 (= 12 + 2√2 + 18−16 + 2√2) pixels by the step (3).

[Step (4)]
Step (4) is continuous with step (3).
Since the defect data in step (4) is 8 pixels, the perimeter is 14 pixels.
Since the defect data in step (3) is 14 + 4√2 pixels and the defect data in step (4) is 14 pixels, the total perimeter is 28 + 4√2 pixels. The portion where the second row (the fourth row and the fifth row) is in contact is 8 (= T) pixels, and since the start point side / end point side are different positions, the subtraction is 16 (= 2T) pixels. Since both the starting point side and the ending point side are diagonal, 2√2 pixels are added. Thus, the peripheral length becomes 12 + 6√2 (= 14 + 4√2 + 14-16 + 2√2) pixels by the step (4).
[Step (5)]
Step (5) is continuous with step (4).
Since the defect data in step (5) is 6 pixels, the peripheral length is 10 pixels.
Since the defect data in step (4) is 12 + 6√2 pixels and the defect data in step (5) is 10 pixels, the total perimeter is 22 + 6√2 pixels. Since the portion where the second row (the fifth row and the sixth row) is in contact is 6 (= T) pixels, and the start point side / end point side are different positions, the subtraction is 12 (= 2T) pixels. Since both the starting point side and the ending point side are diagonal, 2√2 pixels are added. As a result, the perimeter becomes 10 + 8√2 (= 12 + 6√2 + 10−12 + 2√2) pixels in step (5).

[Step (6)]
Step (6) is extinction.
The perimeter of step (6) is 10 + 8√2 pixels.
That is, the first defect detection unit 4 according to the present embodiment calculates the defect data for each line by the connectivity processing unit 13 without acquiring the two-dimensional defect image of the entire defect, thereby obtaining the feature amount of the defect. Perimeter (L) = 21.312 (= 10 + 8√2) pixels can be measured.
Further, the connectivity processing unit 13 adds the defect data in steps (1) to (6) to 0 + 6, 6 + 8, 14 + 10, 24 + 8, and 32 + 6 in order for each line, so that the defect feature amount is obtained. A certain area (S) = 38 pixels can be measured.
Thereby, the connectivity processing unit 13 can calculate the circularity 4πS / L ² = 1.050.
That is, when the number of pixels in the flow direction shown in FIG. 5B is a circle of 5 pixels, the circularity is 1.050 because the area is 38 pixels and the peripheral length is 21.212 pixels. At this time, since the defect having the same width and double length becomes the same image as that in FIG. 5A, the latter is determined to be a circle as the circularity.

In the prior art, since the number of pixels is counted, it is assumed that the width resolution and the flow resolution are the same. In contrast, in the present application, even when the width resolution and the flow resolution are different, it is possible to cope.
Thus, in the defect inspection by the connectivity processing unit 13, the circularity can be calculated in real time by the process corresponding to the resolution. As a result, defect classification (for example, classification of bugs and foreign substances other than bugs) that cannot be performed sufficiently by the classification of area, width, and length is ahead of the defect inspection by the second defect detection unit 6. Can be done.

  As described above, the defect inspection apparatus of the present invention includes a line sensor 3 that reads an image of the inspection object 2, a binarization unit 11 that binarizes data input from the line sensor 3, and a binarization unit 11. A run-length encoding unit 12 that performs run-length encoding on the binarized data and a connectivity processing unit 13 that measures the feature quantity of the defect by performing the connectivity process on the run-length code.

  According to the present invention, it is possible to provide a defect detection apparatus and a defect inspection method that can determine a feature amount of a defect without acquiring a two-dimensional image of the entire defect.

  As described above, the embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to the above, and various design changes and the like can be made without departing from the scope of the present invention. It is possible to

  DESCRIPTION OF SYMBOLS 1 ... Line illumination, 2 ... Inspection object, 3 ... Line sensor, 4 ... 1st defect detection part, 5 ... Image extraction part, 6 ... 2nd defect detection part, 11 ... Binarization part, 12 ... Run Length encoding unit, 13 ... connectivity processing unit

Claims (2)

A line sensor that reads an image of the inspection object;
A binarization unit for binarizing data input from the line sensor;
A run-length encoding unit that performs run-length encoding on the data binarized by the binarization unit;
A connectivity processing unit for measuring a feature quantity of a defect by performing connectivity processing on the run-length code;
I have a,
P is the perimeter length of the nth row, Pa is the total perimeter length to the (n-1) th row, Pa 'is the circumference length of the right side to the (n-1) th row, and Pb is the circumference of the right side of the nth row Length, T is the number of pixels in contact with the defect data in the (n−1) th row and the nth row, bs is the start pixel of the defect data in the nth row, and be is the end pixel of the defect data in the nth row. When
The connectivity processing unit is based on the following calculation formulas (1) to (4) corresponding to the relationship of defects obtained in a line adjacent to the traveling direction of the inspection object.
(1) In the case of an occurrence pattern, P = 2 (be-bs),
(2) For a continuous pattern, P = Pa + P,
(3) In the case of a convergence pattern, P = P + Pa ′
(4) In the case of a dispersion pattern, P = P + Pb,
However, for the above formulas (2) to (4),
The width direction is
When the start point side and the end point side are different positions, P = P−2T,
If either one is in the same position, P = P- (2T-1)
If both are in the same position, P = P− (2T−2)
The flow direction is
When the defect pattern is at the same position on the start point side and end point side, P = P + 1,
If it is slanted, by setting P = P + √2,
A defect inspection apparatus characterized by obtaining a perimeter length P which is a feature amount of the defect. A reading process in which the line sensor reads an image of the inspection object;
A binarization step in which the binarization unit binarizes the data input from the line sensor;
A run-length encoding step in which the run-length encoding unit performs run-length encoding on the data binarized by the binarization unit;
A connectivity processing step for measuring a feature quantity of a defect by performing connectivity processing on the run-length code; and
I have a,
P is the perimeter length of the nth row, Pa is the total perimeter length to the (n-1) th row, Pa 'is the circumference length of the right side to the (n-1) th row, and Pb is the circumference of the right side of the nth row Length, T is the number of pixels in contact with the defect data in the (n−1) th row and the nth row, bs is the start pixel of the defect data in the nth row, and be is the end pixel of the defect data in the nth row. When
The connectivity processing step is based on the following calculation formulas (1) to (4) corresponding to the relationship of defects obtained in a line adjacent to the traveling direction of the inspection object.
(1) In the case of an occurrence pattern, P = 2 (be-bs),
(2) For a continuous pattern, P = Pa + P,
(3) In the case of a convergence pattern, P = P + Pa ′
(4) In the case of a dispersion pattern, P = P + Pb,
However, for the above formulas (2) to (4),
The width direction is
When the start point side and the end point side are different positions, P = P−2T,
If either one is in the same position, P = P- (2T-1)
If both are in the same position, P = P− (2T−2)
The flow direction is
When the defect pattern is at the same position on the start point side and end point side, P = P + 1,
If it is slanted, by setting P = P + √2,
A defect inspection method characterized by obtaining a perimeter length P which is a feature amount of the defect.

Images