JPH02263108A - Defect detecting method - Google Patents

Abstract

PURPOSE:To decrease an error in defect detection by using a specific reference image pattern repeatedly when detecting a defect of a body to be inspected which consists of a step. CONSTITUTION:A predetermined small-range flat part pattern cut from the flat part of image pickup picture elements of the body 11 to be detected and a predetermined small-range step part pattern cut from the step part 13 are used as reference image patterns, which are used repeatedly. Namely, an inspection image pattern 1 and a flat part cut pattern 3 are compared repeatedly over the entire image plane by using the cut pattern 3 and then the calculation of the quantity of dissidences between step part cut patterns (a) - (d) and the inspection image in a predetermined range centering on the position of the step part 13 and the movement of the cut patterns (a) - (d) are performed repeatedly; and the inspection image pattern 1 is compared with the cut patterns (a) - (d) at the position where the quantity of the dissidences is minimum within the range of the presence of the step.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for inspecting the appearance of precision castings, and in particular, detects irregularities that occur at stepped portions or edge portions on the surface using images. This relates to a method of automatically inspecting during processing.

[Conventional technology]

In order to automatically use image processing to inspect concave or convex defects that occur on the surface of an object during the manufacturing process, an image is first captured.

That is, first, a light beam is irradiated onto the surface to be inspected using an illumination device. The types of light rays include parallel rays, scattered rays, etc., and the direction of irradiation may be perpendicular to the surface to be inspected or at a certain angle. Then, the reflected light of the irradiated light beam on the surface to be inspected is imaged using a linear light receiving element such as a line sensor or a planar light receiving element such as an image pickup tube.

On the other hand, a method for detecting a concave or convex defect imaged by the above method is disclosed in, for example, Japanese Patent Laid-Open No. 72679/1983. This takes advantage of the fact that the intensity of the scattered light received changes depending on the shape of the defect, binarizes the video signal at a set slice level, and calculates the feature amount of the defect from this binarized signal. It is for inspection.

Furthermore, in addition to flat areas where the reflected light is uniformly reflected on the screen captured by the above-mentioned imaging method, if there is a step on the surface to be inspected, the reflected light changes at the step as well as at a defective area. A pattern appears. The image also includes the pattern of the edges of the object to be inspected. However, this edge pattern is not a defect. However, in precision casting products, defects such as condensation and cracks are likely to occur at the stepped portions and edge portions.

As a method for detecting defects in such step portions, there is a pattern matching method in which an image of a normal crystal without defects is compared with an image of the object. This method has been put to practical use, for example, in testing semiconductor devices such as LSI wafers. As a recent example, Japanese Patent Application Laid-Open No. 61258106 proposes that the inspection image pattern includes alignment errors and errors due to distortion, and in order to prevent false detection due to these errors, the following alignment method is proposed. ′ was found. That is, while a partial area is set within the inspection image pattern, an enlarged area that includes the partial area and is wider than the partial area is set within the reference image pattern. Then, within the enlarged area, a reference area of the same size as the partial area is provided, and the reference area and the partial area are compared pixel by pixel while shifting the reference area pixel by pixel, and the number of mismatches is minimized. reference area,
This is the comparison position of the partial area. In this way, the comparison position is detected by providing partial areas over the entire area of the inspection image pattern, and comparison is performed at each comparison position.

[Problem that the invention seeks to solve]

However, precision cast products have a rough surface and the properties of the stepped portions are not uniform. Moreover,
There are microscopic defects somewhere.

From the above, first of all, with the conventional method using the hammered binary signal described above, it is possible to accurately detect only when the surface quality of the object to be inspected is relatively good and flat. In the case where the surface has a rough surface and the properties of the stepped portion are non-uniform, the stepped portion also appears as a binary image, and does not become a --like image.

Further, in the method using pattern matching, it is necessary to create a good reference pattern over the entire area, but this is extremely difficult due to the surface properties peculiar to the precision casting product. Furthermore, in order to store the reference image pattern, a memory is required to store the pattern over the entire surface of the object to be inspected. Therefore, as the memory capacity increases, the peripheral circuitry becomes complicated, the device becomes bulky, and it becomes economically expensive.

The present invention solves the problems existing in the prior art, reliably detects defects that occur in precision castings with rough surfaces and steps, especially in the step portion, and can be configured into a highly reliable and inexpensive inspection device. The purpose is to provide a method.

[Means for solving problems]

The present invention is an image defect detection method that uses a pattern matching method to detect defects by comparing an inspection image pattern with a reference image pattern. The flat part pattern cut out and the stepped part pattern cut out in a predetermined small range from the stepped part are used as reference image patterns, and these cut out standard image patterns are repeatedly used in pattern matching. It is.

In other words, after comparing the inspection image pattern and the flat area cutout pattern over the entire screen by repeatedly using the cutout pattern, the rough position of the stepped part is known in advance from the design of the subject, so this position is used as the center. The process of repeatedly calculating the amount of mismatch between the step cutout pattern and the inspection image and moving the cutout pattern within a predetermined range, and comparing the inspection image pattern and the cutout pattern at the position where the amount of mismatch is the smallest is performed. This is done repeatedly within a certain range.

〔Example〕

FIG. 3 shows the object 11 in this example, which has both a flat portion 12 and a stepped portion 13 on its surface.

FIG. 4 is a block diagram of an apparatus in an embodiment of the present invention. First, the illumination device 14 is used to irradiate the subject 11 from diagonally above. On the other hand, an imaging device 16 such as a TV camera forms an image through a lens 15. The imaging signal from the imaging device 16 is converted into grayscale image data by the AD converter 17 and stored in the memory 18. A processing circuit 19 processes the contents of the memory 18.

Note that since the test object 11 is a precision cast product, the illumination device 1
If the incident angle of 4 is set appropriately, the normal area will be bright (and the defective area will be dark).

Next, FIG. 1 shows an inspection image pattern 1 taken with the above configuration. Stepped part image 2 due to stepped part 13 of subject 11
The defect 9 generated in the stepped portion 13 also appears as an image adjacent to the stepped portion image 2. In addition to the flat part cutting pattern 3, the standard image pattern includes the stepped part 13.
From the shape of
(C) and (D) are created in this embodiment.

The main parts of the alignment and comparison processing using the step cutout pattern (a) in Fig. 1 are shown in the second section using the enlarged area 8 in Fig.
This will be explained using Figures (a), (b), and (c). The upper view of FIG. 2 (a) shows the case where the step part of the step cutout pattern (a) is on the left side of the step part of the inspection image pattern 1, and similarly, in (b) In contrast to the case where the part is exactly shaped, in (c) the step part cutting pattern (
The figure shows the case where b) is on the right side. The brightness distribution in such a case is shown in Figure 2 (a) (b) (
c) Each is shown below, but the difference is large in cases (a) and (c), and the difference is the smallest in case (b). That is, the position where the difference is the minimum is where the steps meet. In FIG. 2, since the molecule direction of the step portion is in one direction, the step cutout pattern (A) only needs to be moved in a direction perpendicular to that direction.

By the above processing, the most aligned position is determined, and when a comparison is made at this position, the defective part 9 can be left behind and separated from the stacked part image 2, as shown in FIG. 2(d).

The above explanation concerns the case of the step cutout pattern (a), but the overall flow is as shown in FIG. 5. That is, first, the flat portion cutting pattern 3 is repeatedly used and compared with the inspection image. Next, use the step cutout pattern (A) to align the above position, and inspect the inspection image pattern 1.
The comparison is repeated for all the corresponding step areas. Next, positioning is performed using the step cutout pattern (b) and comparison is made. In this case, since the stepped portion is not only in one direction, the cutout pattern is moved in the χ and Y-axis directions. Next, the cutout pattern for the step part (c)
Use to align and compare.

Finally, alignment comparison is performed using the step cutout pattern (d).

[Effects of the Invention] With the above configuration, in detecting defects in an inspection image of an object having a rough surface and uneven steps, a base image pattern having the same degree of surface roughness and unevenness as the object is repeatedly used. By using this, it is possible to create a reference pattern using a good part even from a defective object, for example. It is also possible to cut out the best parts from among a plurality of objects and use them as reference patterns. Furthermore, since the positioning process is performed using the reference pattern, it is possible to prevent a portion that is not originally defective from being mistakenly recognized as a defect due to distortion or positioning error contained in the inspection image.

(1) Since a reference inspection pattern can be created using a good portion of a captured image, a reference image pattern can be created even from a defective product, for example.

Furthermore, since it is possible to extract good parts from each of multiple objects and use them as reference image patterns, it is not only easy to create reference image patterns, but also to reduce errors in defect detection and improve reliability. I can do it.

(2) Since comparison is performed after alignment is performed using the reference image pattern, errors can be minimized even in products that are naturally distorted, such as cast products, and reliability is high.

(3) Since the reference image pattern does not require storing the entire object to be inspected, the memory capacity is reduced, and economically, the apparatus can be provided at low cost.

[Brief explanation of drawings]

Figure 1 is a diagram showing an example of creating a reference image pattern according to the present invention, and Figures 2 (a), (b), (c), and (d) are alignment and comparison processes using cutout patterns according to the present invention. 3 is an external view of the object to be inspected used in the example, FIG. 4 is a block diagram of the apparatus, and FIG. 5 is a processing flow. 2: Step image, (a) Two-step cutout pattern, 8:
Enlarged area, 9: Defect, 11: Object. 5IO arrogant chart

Claims (1)

[Claims] In a defect detection method for detecting defects by comparing an inspection image pattern with a reference image pattern, the reference image pattern is a flat area pattern in a predetermined small range from a flat area of a captured image of a subject. It is created by cutting out a pattern and creating multiple step patterns in a predetermined small range each time the step changes from step to step on the surface of the subject, and storing each cut pattern in memory. After repeatedly comparing the inspection image pattern and the flat section cutting pattern to compare the entire object imaging surface, the inspection image pattern and the flat section cutting pattern are compared in a predetermined range centered on the position determined from the design of the stepped section. A defect detection method characterized in that the amount of coincidence with a corresponding step pattern is calculated while the step pattern is moved, and the process of comparing at the position where the amount of coincidence is the smallest is repeated for the entire area of the step. .