JPH02263107A - Method for inspecting linear defect - Google Patents

Abstract

PURPOSE:To accurately detect the overall length of the linear defect by extracting a defective cell from a discrimination area consisting of ranges of plural picture elements centering on defective candidate picture elements, finding the frequency distribution of those defective cells, and accumulating values exceeding a predetermined slice level and obtaining the overall length of the defect. CONSTITUTION:On plural upper and lower lines centering on an inspection line other than lines adjoining to the inspection line, defective picture elements 23 in the predetermined discrimination range 25 centering on the defective candidate picture element 24 is checked. Then if there are defective picture elements 23 in all discrimination areas 25, the defective candidate picture elements 24 are judged to be defective picture elements 23 and cells 27 including the defective candidate picture elements 24 are extracted as defective cells 28. Then, the Y-axial frequency distribution of the defective cells 28 is generated and values exceeding the predetermined slice level 5 in the frequency distribution are accumulated to obtain the overall length of the linear defect 1. Consequently, the overall length of the linear defect 1 is accurately calculated.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method of inspecting linear defects such as cracks and scratches existing on the surface of electronic parts or precision processed products by image processing.

[Conventional technology]

BACKGROUND ART Conventionally, the following methods have been proposed as methods for detecting linear defects such as cranks and scratches existing on the surface of electronic components or precision-processed products using images.

First, the method proposed in JP-A-59-37450 calculates a rectangle circumscribing a linear defect [Fig. 5(a)], as shown in Fig. 5(a) and (b). Figure 5 (b
) L The rectangle is enlarged by a certain amount and brought into contact with a rectangle existing in another nearby area, and an appropriate defect length is determined from the discontinuous defect image.

Furthermore, the method proposed in JP-A No. 63-85432 is shown in FIGS. 6(a) and 6(b), which involves converting an image containing linear defects (FIG. 6(a)) into multiple images. A plurality of cells consisting of pixels [Fig. 6(b)] are set in a matrix in the x and y axis directions, and the number of defective pixels in each cell is calculated, and cells whose number of defective pixels is on a predetermined number of groups are selected. This method extracts defective cells as defective cells, calculates the degree of connectivity of these defective cells, and judges those whose degree of continuity exceeds a predetermined value as defective.

[Problem that the invention seeks to solve]

Among the conventional examples mentioned above, those using circumscribed rectangular parallelepipeds have a lot of noise in addition to defects (if such noise exists, if the circumscribed rectangular parallelepiped of each noise is carelessly enlarged, it will come into contact with that of a linear defect, Furthermore, since the circumscribed rectangular rectangle is calculated and enlarged for each noise, it takes a long processing time.Alternatively, there is a problem that the apparatus becomes large-scale.

In addition, in an example using a defective cell, the amount of noise removed is
There was a problem in that the extracted defective cells became discontinuous and the original linear defect length could not be detected.

The present invention solves the problems existing in the prior art as described above, eliminates noise in input images, and provides a highly reliable linear defect system that completely inspects linear defects that appear continuously in apparent earthworks. The purpose is to provide an inspection method for

[Means for solving problems]

In order to solve the above problems, in the present invention, A.
It is stored in the image memory as valued pixel data.

B. For the binarized pixel data, a plurality of cells each consisting of a plurality of pixels are set in a matrix in the XY-axis directions through the central control device.

The defect candidate pixels in each cell of C1 are subjected to the following processing to determine whether they are part of a linear defect or part of noise.

That is, other than the lines adjacent to the inspection line, defective pixels are examined in a plurality of lines above and below the inspection line and within a predetermined discrimination area centered on the defective candidate pixel.

D. If there are defective pixels in all discrimination areas, the defective candidate pixel is determined to be a defective pixel. Then, the cell containing the pixel is extracted as a defective cell.

E. Create the Y-axis direction frequency distribution of defective cells.

In the F0 frequency distribution, the cumulative value of those having values exceeding a predetermined slice level is calculated and taken as the total length of linear defects.

This is a technical method that has been adopted.

[For production]

With the above configuration, noise in the input image of the inspected object is usually in the form of points or blocks. Most of the noise is removed in the extraction stage, and has the effect of eliminating errors in recognizing noise that is not originally a defect as being a defect.

Further, even if a linear defect has a discontinuous portion, it can be made to have almost no influence on the calculation of the total length of the linear defect by cumulatively calculating the defective cell frequency value.

After removing noise as described above, by accumulating the frequency values of defective cells consisting of linear defects, the total length of linear defects can be accurately calculated.

〔Example〕

FIG. 3 is a block diagram of an apparatus in an embodiment of the present invention. In FIG. 3, first, an image of an object to be inspected 31 is imaged through a lens 32 on an imaging device 33 such as a TV camera. Next, the imaging signal from the imaging device 33 is converted into binarized pixel data by the D conversion device 34 and input into the memory 35 . 36
3 is a processing circuit that processes the contents of the memory 35, and 37 is a memory that stores the processing results of the processing circuit.

Next, Figure 1 (a) (c) (d) Figure 2 (b) (c
) are schematic diagrams showing main parts of the processing procedure of the present invention.

First, FIG. 1(a) shows a part of the binarized pixel data stored in the memory 35 in FIG.

In FIG. 1(a), 1 is a scratch, which exists on the surface of the object 31 to be inspected. It is a type of linear defect.

Therefore, scratch 1 is originally continuous, but scratch 2
Part of the converted pixel data is in a discontinuous state as shown by a discontinuous portion 2.

3 is noise, which is caused by uneven background on the surface of the object 31 to be inspected and is stored as binary pixel data.

Next, FIG. 2(b) shows a state in which calculations are being made to extract cells set in this embodiment and defective cells. That is, 21 is a cell, and for example, in this embodiment, it is composed of 8×1-8 pixels,
A plurality of pieces in the x, y, and y axes are each set in a matrix via a central control device (not shown). 22 in Figure 2(b)
is a non-defective pixel (for example, a white pixel), 23 is a defective pixel (also a black pixel), and 24 is a black pixel on the inspection line, but it is a pixel that should not be determined whether it is part of noise or part of a defect. Yes, indicating that the pixel is a defective candidate pixel 24.

That is, if the pixel is a noise, it is a non-defective pixel, and if it is a defect, it is a defective pixel. This determination is made using the discrimination area 25 shown in second (b). The discrimination area has a width in the y-axis direction of a plurality of lines above and below a line adjacent to the inspection line 26, in this embodiment two lines each, and a width in the chi-axis direction of several pixels width centered on the defect candidate pixel 24. In this embodiment, it is an area having a width of 30 and 7 pixels. The location of the discrimination area 25 changes as the defective candidate pixel 24 moves.

In this discrimination area 25, area (a) (TI) (
Only when there are defective pixels 23 in all of (c) and (d), it is determined that the defective pixel 24 is a defective pixel 23. Next, the cell 27 including the defective candidate pixel 24 is the defective cell 2
day. In this judgment, the noise 3 shown in FIG. 2(b) does not indicate a defective cell.

When the above processing is performed for all pixels, Figure 2 (c)
The defective cell image is shown in .

The defective cell images extracted by the applicant are shown in Figure 1 (
Shown in c). It can be seen that the noise 3 is almost eliminated and only the defective cell 28 due to the linear defect remains.

There are 4 discontinuous parts corresponding to 2 in the defective cell image shown in FIG. 1(c). Next, the total length of the linear defect is calculated from the defective cell image (c) having such discontinuous parts 4 The method is shown in FIG. 1(d). That is, in FIG. 1(d), the horizontal axis is
This is the axial position, and the vertical axis shows the frequency distribution of defective cell images 28 counted in the y-axis direction at each X-axis position. The total length of the defect can be calculated by comparing the frequency distribution with a predetermined slice level 5 and calculating the cumulative number of frequency values exceeding the slice level 5.

Generally, the discontinuous part of a linear defect is so small that it can be ignored compared to the total length of the defect, as shown in the discontinuous part 4 shown in FIG. You can ask for it.

In addition, in this embodiment, the discrimination area 25 (() , (o
) Regarding the handling of black pixels in (c) and (d), if there are all black pixels, the cell is considered a defective cell, but according to the implementation by the applicant, noise can be detected by changing the handling as follows. and the degree of discrimination of linear defects can be freely changed.

(1) If there are black pixels in areas (0) and (c), they are determined to be defective cells.

As shown in FIG. 4(a), noise is increasing.

This judgment is also possible if the noise is reduced at the time of creating the binary image (11) If there are black pixels in areas (a) and (d), it is determined that the cells are defective.

As shown in FIG. 4(b), a considerable amount of noise remains.

is still left.

(iii) As shown in this example, area (A) (0
) If there is a black pixel or pixel in (c) or (d), it is considered a defective cell.

As shown in Figure 4(c), the noise has been greatly reduced,
Furthermore, the formation of linear defects into defective cells is not impaired.

〔Effect of the invention〕

Since the present invention has the structure and operation as described above,
It has the following effects.

(1) Even if the input image signal contains noise other than the original linear defect, it can be determined whether it is part of the defect or part of the noise using the discrimination area that goes beyond the line adjacent to the inspection line. Since noise is removed, defect detection is extremely reliable.

(2) In the input image signal, even if it is a discontinuous linear defect, the total defect length is calculated based on the cumulative value of the frequency distribution.
The total length of linear defects that are originally continuous can be determined with high reliability.

(3) The degree of discrimination between noise and linear defects can be freely changed depending on the size of the discrimination area and the method of handling black pixels within the area, so it is highly flexible.

[Brief explanation of drawings]

FIG. 1 and FIG. 2 are schematic diagrams showing the main parts of the processing procedure of the present invention, respectively. FIG. 3 is a block diagram of an apparatus in an embodiment of the present invention. FIG. 4 is a diagram showing the difference in the degree of discrimination between noise and linear defects due to the handling of black pixels within the discrimination area according to the present invention. 5 and 6 are conventional examples, respectively. (C) 2nd (a) (c) Fig. 4 (a) (b) Fig. (b) Fig.

Claims (1)

[Claims] The imaging signal of the object to be inspected obtained through the imaging means is stored in an image memory as binarized pixel data in the XY axis directions, and the binarized pixel data is sent to a cell consisting of a plurality of pixels via a central controller. For each defective candidate pixel in each cell, set a plurality of them in a matrix in the X and Y axes directions, and determine whether the defect candidate pixels in each cell are part of the defect or part of the noise. Defective cells are extracted by setting a discrimination area consisting of a range of multiple pixels centered on the defective candidate pixel in the x-axis direction on multiple lines above and below the line, and making judgments. A method for inspecting linear defects, characterized in that the total length of defects is determined by determining the frequency distribution of defects having values exceeding a predetermined slice level.