JPH06174659A - Scratch inspection method by image processing - Google Patents

Abstract

(57) [Abstract] [Purpose] It is an object of the present invention to provide a flaw inspection method by image processing that can reliably distinguish dust and dirt from small scratches. In a flaw inspection method for determining directionality of each image by image processing to determine whether or not it is a flaw, image data of an object to be inspected is processed to obtain a binary image group to be inspected, and these 2 A label area 5 in which the same number is assigned to each pixel 2 forming each binary image of the value image group is created, and each label area 5 is formed.
The contour direction code 3 which is the direction between adjacent pixels is calculated for the pixels 2 forming the contour of
The direction of the label area 5 is calculated from the distribution of the directions, and the data of the label area 5 whose calculated directions are close to each other is created to form a data fusion area, and the data fusion is performed according to the feature amount indicating the area of the data fusion area. It is characterized by determining whether or not the area is scratched.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for inspecting scratches on mechanical parts and the like, and more particularly to a method for inspecting scratches by image processing.

[0002]

2. Description of the Related Art A conventional example of a flaw inspection method by image processing will be described with reference to FIGS.

FIG. 1 shows a cylinder to be inspected for flaws. FIG. 2 shows a binary image 1 of a scratch when the cylinder surface is enlarged. In the case of a scratch inspection by image processing, if a smooth surface is scratched, the scratched portion will be diffusely reflected, so if the light is obliquely applied and dark field illumination is performed so that the reflected light does not directly enter the imaging device, It is possible to obtain an image of a scratch in which only the scratched portion that is diffusely reflected shines. The binary image 1 in FIG. 2 is an image of the scratch thus obtained.

In order to obtain the binary image 1, first, the input of the television camera is stored in the image storage circuit by using the image input device.

Next, it is binarized. This binarization means that a pixel having a density larger than a certain threshold is set to "1" and a pixel having a density equal to or less than the same threshold is set to "0". Therefore, when FIG. 2 is binarized, the shining scratched portion becomes "1" and the non-scratched portion becomes "0".

Next, a labeling process and a renumbering process are performed. The labeling process divides a region of pixels having a density of “1” and a region of pixels having a density of “0” into pixels, and the pixels in the region of “1” are vertically, horizontally, and diagonally adjacent to each other. This is a process in which the connected portions are regarded as a group of connected areas, serial numbers are assigned to these connected areas, and the serial numbers of the connected areas are commonly assigned to the pixels in each connected area. The connected area with the serial number is called a label area. The renumbering process is to renumber the serial numbers so that there is no omission.

The labeling process and the renumbering process will be described with reference to the PAD charts of the labeling process and the renumbering process shown in FIGS.

FIG. 3 shows the entire labeling process, and shows that the labeling process and the renumbering process are performed in steps A to E.

FIG. 4 shows the processing / step A of the uppermost line, which scans from the uppermost line in the raster order, that is, in the order of writing a horizontal writing, and the density "1" that appears first. The pixel of interest is the pixel of interest, the label number "1" is assigned to this pixel, scanning is continued, the pixel of interest advancing to the next density of "1" is checked, the label number of the pixel on the left is checked, and the density of the pixel on the left is Is '1' and the label number is generally n including "1", the same label number n as on the left is given, and it is memorized that the pixel of interest and the pixel on the left are connected. deep. When the density of the pixel on the left side is '0' and there is no label number, generally n + 1 including the new label number "2" is included.
Attach.

FIGS. 5 and 6 show the processing of steps B and C of the second and subsequent lines, and the raster scanning of the second and subsequent lines is continued, and the reached pixel of density '1' is the target pixel. For this pixel of interest, check the label number of the 4 pixels on the left side, the upper right side, the upper side, and the upper left side that have been scanned in the raster scan so far among the 8 neighboring pixels vertically, horizontally, and diagonally. If only one type of label number is given, that number is given to the pixel of interest. If two or more types of label numbers n, m, etc. are given, data for giving any one of the label numbers to the pixel of interest and later rearranging and unifying the label numbers of the connected regions. As information, information that the areas of label numbers n and m are connected to the pixel of interest is stored. Also, the label numbers of the four pixels on the left side, the upper right side, the upper side, and the upper left side, which have been scanned by the raster scanning up to that point, are checked. .

FIG. 7 shows label number rearrangement and step D. In order to rearrange and unify the label numbers of the connected label areas after the labeling processing is completed for all the pixels, the pixels of the connected areas are unified. It is unified to one of the plurality of label numbers assigned to.

FIG. 8 shows a step of changing the label number and step E, and if necessary, performs a renumbering process for changing the number so that the numbers are not missing.

After the labeling process and the renumbering process of each binary image of the density '1' are completed as described above, each binary image (hereinafter referred to as a label area) assigned with each number is respectively Area, that is, the number of included pixels is calculated. In this case, parts other than scratches may shine due to dust reflection and slight irregular reflection of smooth parts, but these are small areas, so in order not to misidentify these as scratches, small areas are not Considering that it is not a flaw, only the label area having a large area is extracted as a flaw.

[0014]

However, the above-mentioned conventional example has the following problems. That is, in the case of scratches generated by impact on the smooth surface of the object to be inspected, as shown in the binary image 1 of FIG. Taking the label area of three scratches is small in size, and it may be difficult to judge whether scratches, dust, or dirt.

An object of the present invention is to solve the above problems and to provide a flaw inspection method by image processing capable of surely discriminating dust and dirt from a flaw.

[0016]

In order to solve the above problems, a flaw inspection method by image processing according to the present invention is a flaw inspection method for determining the directionality of each image by image processing to determine whether or not there is a flaw. In the above, the image data of the object to be inspected is processed to obtain a binary image group to be inspected, and a label area is formed by assigning the same number to the pixels forming each binary image of these binary image groups, A contour direction code which is a direction between adjacent pixels is calculated for pixels forming the contour of each label area, the direction of the label area is calculated from the direction distribution of the contour direction code, and the label area in which the calculated direction is close is calculated. It is characterized in that the data fusion area is created by fusing the data of 1. and whether or not the data fusion area is a scratch is judged by a feature amount indicating the area or the like of the data fusion area.

[0017]

According to the flaw inspection method by the image processing of the present invention, the flaws caused by the same factor have a long and narrow shape even if the flaws are divided into a plurality of pieces, and are aligned on one line or in the same direction. Therefore, the shape and directionality of the image to be inspected is determined to determine whether or not it is a scratch. Therefore, the image processing according to the present invention is performed by processing the image data of the inspection object.
Up to the step of obtaining the value image group and creating the label area in which the pixels constituting each binary image of these binary image groups are given the same numbers, the procedure is the same as in the conventional example, but after that, each label For each area, a contour direction code indicating the shape and directionality of the area is obtained, the shape and directionality of the label area are calculated from the direction distribution of this contour direction code, and the label area that extends long in the same direction. , The data of the selected label area is fused, and the data fusion area where the data is fused,
It is judged whether it is a scratch or not.

Since the contour direction code is the direction between adjacent pixels of the pixels forming the contour of each label area, the contour direction code group obtained for one label area is aligned in a certain direction. In that case, the label area will be elongated in that direction.

Therefore, in the present invention, the contour direction code of each label area is obtained, the direction distribution of the contour direction code is calculated for each label area, and the direction in which the label area extends long is calculated from this direction distribution. Then, assuming that the label areas whose calculated directions are close to each other may constitute one flaw, the data is fused to create a data fusion area, and the data fusion is performed according to the feature amount such as the area of this data fusion area. Determine if the area is scratched.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment method of the present invention will be described with reference to FIGS. In this case, FIGS. 1 to 8 are common to the conventional example.

In the flaw inspection method by image processing according to the present invention, in the image processing of the inspection object, the labeling processing performed in the same manner as the conventional example is finished, that is, the binary image, that is, the label area, the contour of each label area. Ask for direction code,
The direction distribution of the contour direction code is used to calculate in which direction the shape of the label area extends, and the data of the label area in which the direction that is calculated to be elongated is fused to create a data fusion label area. Whether or not the data fusion label area is scratched is determined based on the feature amount such as the area of the data fusion label area.

In the present embodiment, first, the surface of the cylinder shown in FIG. 1 is imaged by a television camera, the obtained image data is stored in an image storage circuit using an image input device, and binarized. A screen having a large number of binary images 1 as shown in 2 is created. After that, as in the conventional example, as shown in FIGS. 3 to 8, a large number of binary images in the screen are labeled and renumbered to obtain a large number of label areas. The processing of FIGS. 3 to 8 is the same as in the case of the above-described conventional example, and thus the description thereof is omitted.

Next, regarding the above-mentioned many label areas,
The contour direction code is obtained. The definition and the method of obtaining the contour direction code are as follows.

FIG. 9 explains the contour direction code 3 and how to obtain it. Pixel 2 is a pixel that constitutes label area 5, and contour direction code 3 is
The contour vector is a contour vector that constitutes the contour of the label area 5 and that is between adjacent pixels 2 around a specific direction.

A method for obtaining the contour direction code 3 from a screen having a large number of label areas 5 for which the labeling processing has been completed will be described with reference to FIG.

In FIG. 9, first, the screen is scanned in the raster direction, which is the order in which a horizontal writing sentence is written, and the pixel P1 in the label area 5 that has arrived first is set as the target pixel. If eight adjacent pixels in the vicinity of 8 are examined clockwise from the left, and the adjacent pixels whose image densities of these adjacent pixels have changed from “0” to “1” are defined as pixel P2, pixel P1 is set as the starting point and pixel P2 is set as the starting point. The vector whose end point is is the contour direction code 3.

Next, with the pixel P2 as the target pixel, the neighboring pixels are examined clockwise from the pixel P1 when viewed from the target pixel, and the image densities of these neighboring pixels are from "0" to "1".
The neighboring pixel that has changed to is referred to as a pixel P3. Similarly, the pixel P4, the pixel P5, ...
Continue until you return to 1. In this way, the contours of the label area 5 can be connected clockwise by the contour direction code 3.

The contour direction code obtained as described above has eight directions of up, down, left, right, and diagonal, but if the directions opposite 180 ° are the same direction, there are four directions.

In the present embodiment, next, the direction distribution of the contour direction code 3 is obtained. FIG. 10 is a direction distribution graph 4 showing the direction distribution of the contour direction code 3 for each label area 5.
The direction distribution graph 4 is composed of frequencies in each direction. Therefore, the label area 5 extends long in the direction in which the frequency is large, and if the direction with the maximum frequency is obtained from the data in FIG. 10, that direction becomes the direction in which the label area 5 extends. , The shape and directionality of the label area 5 can be known. The calculation result of the contour direction code 3 and the direction distribution data are stored.

In the present embodiment, next, using the above-mentioned stored data, the data in the label area 5 whose direction is judged to be elongated is fused to create a data fusion label area, Perform clustering processing.

The clustering process is performed as shown in the PAD chart of clustering in FIGS.

FIG. 11 shows the entire clustering process, and shows that the data is read from the reference table composed of the stored data and the clustering is performed in steps A and B.

FIG. 12 shows the label area data merging process / step A for merging the label area data, and the contours of all combinations of the label areas 5 for which the direction distribution of the contour direction code is obtained. A combination of label areas 5 having the same maximum distribution direction of direction codes is selected, and the data of the selected label areas are fused.

FIG. 13 shows a renumbering process / step B in which the numbers in the data fusion label area are renumbered so that there is no omission. This is a process of matching a plurality of label numbers in the label area 5 having the same maximum distribution direction with one of them.

This concrete work will be described with reference to FIG. FIG. 14A shows data before the renumbering process, and FIG. 14B shows data after being renumbered to 0 and 1. 1 and 5 have the same number, so 1
Integrated into. Since 0 and 2 have the same number, they are integrated into 0, and 3, 4, 5 have different numbers from 1 and 2, but I remember that they are consecutive,
0 is integrated.

In this embodiment, after the above-mentioned clustering process and renumbering process are performed, that is, regarding the data fusion label region in which data is fused by the judgment based on the direction, the fusion label region is determined according to the size of the area. To determine if it is scratched.

In this way, although it was caused by one cause, the scratches attached to it were separated into a plurality of scratches, and even when each scratch was small, the data of these separated small scratches were merged. Since it can be distinguished from the adhesion of dust or dirt, it can be accurately inspected for scratches.

[0038]

According to the flaw inspection method by image processing of the conventional example, when it is determined whether or not the image to be inspected is a flaw caused by an impact, it is determined using only the size of the image to be inspected. Therefore, it is difficult to distinguish between a small image due to adhesion of dust or dirt and a small image with small scratches, but in the present invention, the direction in which the image is elongated is recognized and separated by the contour direction code. From the images, multiple small images that are elongated in the same direction are extracted, and the data of these images are fused to determine whether or not it is a scratch.
The present invention has an effect that it is possible to more accurately determine whether or not a scratch separated into smaller pieces is a scratch as compared with the conventional example.

[Brief description of drawings]

FIG. 1 is a perspective view of a cylinder of an object to be inspected.

FIG. 2 is an image of the scratch of FIG.

FIG. 3 is a PAD chart of labeling processing.

FIG. 4 is a PAD chart of labeling processing.

FIG. 5 is a PAD chart of labeling processing.

FIG. 6 is a PAD chart of labeling processing.

FIG. 7 is a PAD chart of labeling processing.

FIG. 8 is a PAD chart of labeling processing.

FIG. 9 is a configuration diagram of a contour direction code of the present invention.

FIG. 10 is a direction distribution diagram of contour direction codes according to the present invention.

FIG. 11 is a PA of clustering processing according to the embodiment of this invention.
It is a D chart.

FIG. 12 is a PA of clustering processing according to the embodiment of this invention.
It is a D chart.

FIG. 13 is a PA of clustering processing according to the embodiment of this invention.
It is a D chart.

FIG. 14 is an operation diagram of a renumbering process according to the embodiment of this invention.

[Explanation of symbols]

 1 Binary image 2 Pixel 3 Contour direction code 4 Direction distribution graph

Claims (1)

[Claims] 1. A flaw inspection method for determining directionality of each image by image processing to determine whether or not there is a flaw, wherein image data of an object to be inspected is processed to obtain a binary image group to be inspected, and A label area in which pixels forming each binary image of the binary image group are assigned the same number, and a contour direction code which is a direction between adjacent pixels is calculated for the pixels forming the contour of each label area. , The direction of the label area is calculated from the direction distribution of the contour direction code, and the data of the label areas whose calculated directions are close to each other is created to create a data fusion area, and the data fusion area is characterized by a feature amount indicating the area or the like. A flaw inspection method by image processing, wherein it is determined whether or not the data fusion area is a flaw.