JPS6014165Y2 - Line symmetric pattern recognition device - Google Patents

Description

[Detailed description of the invention] [Field of application of the invention] The present invention relates to a line-symmetric pattern recognition device that recognizes features such as minute defects that exist in line-symmetric patterns such as circles, ellipses, triangles, and squares. It is.

[Background of the idea]

Conventionally, as a line-symmetrical pattern inspection apparatus, there are a pattern matching method and a method of inspecting based on area 9 and perimeter length.

In the former method, as shown in FIG. 1, for example, a circular pattern 3 is prepared in the circuit to judge the acceptability of the circular pattern 1 to be inspected, and this and the circular pattern 1 to be inspected are There is a method in which the quality of the inspection pattern 1 is determined by overlapping the pattern 1, extracting the mismatched part 5, and determining the size of the pattern 5, etc., and a method in which the inspection pattern 1 and the pattern 6 obtained by rotating it by an arbitrary angle are overlapped as shown in FIG. , the inspection pattern 1 is extracted by extracting the mismatched parts 8 and 9 and determining their size, etc.
There is a known method for determining the quality of the product.

These methods are effective when detecting fairly large defects in simple patterns such as circular patterns, but when attempting to detect minute defects, they have the disadvantage that the defects are buried in quantization errors.

Here, the quantization error is the result of dividing the inspected pattern 10.13 into two-dimensional meshes 11 and 14 and binarizing it so that it can be processed by a logic circuit, as shown in Figure 3a and b. is the error that occurs at this time.

Figure 3 at b shows this phenomenon in order to make it easier to understand.
This is a case where the pattern is binarized using a mesh with the same size as the radius of the circle to be inspected, and when the proportion of patterns in one mesh is more than half, it is "1', and when it is less than half, it is "0".
It is binarized so that ゛゛.

Pattern 12 quantized with the mesh in Figures 3a and b.
15, it is clear that even circles of the same size can become one picture element or four picture elements depending on the mesh position.

This error is called a quantization error here.

That is, as shown in FIGS. 1 and 2, in the pattern matching method, both patterns 1 and 3.6 have quantization errors, so if we take the mismatched portions, we will see that on contours 4 and 7. This error is extracted as a mismatched portion.

Furthermore, in the pattern matching method described above, especially when the defect 2 is minute, the defect 2 is buried in the quantization error and cannot be detected.

In addition, the latter inspection method based on area 9 and perimeter length is similar to the pattern matching method described above, in which the number of picture elements with a value of "1°" is determined for a pattern binarized with a two-dimensional mesh, and this is calculated based on the area S1 of the pattern and its contour. Extract only the part, find the number of picture elements with the value GG 199 for the contour figure, take this as the perimeter of the pattern 1, check whether S, 1 is the correct value, and in the case of a circular pattern, (2π
γ) From 2/πγ2=12/S, 12/S=4? Using the property of r (constant), calculate 12/S, which is 4? r
This method determines whether it is a yen or not based on whether it is .

This method also has the disadvantage that minute defects are hidden in quantization errors and cannot be detected as defects.

[Purpose of invention]

An object of the present invention is to eliminate the above-mentioned drawbacks of the prior art and to provide a line-symmetric pattern recognition device that can detect even extremely small defects as defects.

[Summary of the idea]

In order to achieve the above object, the present invention provides an imaging device that detects a video signal by scanning a line-symmetrical pattern with a plurality of scanning lines in a direction perpendicular to the line, and a video signal detected by the imaging device. a binarization circuit that converts the image into a binarized video signal, and a binarization circuit that cuts out the binarized video signal outputted by the binarization circuit into two-dimensional picture elements and stores them. When it is recognized that there is an inclination in the line-symmetrical pattern, an outline extracting means extracts the line-symmetrical pattern with a binary signal as having an outline; Gradient extracting means for extracting whether or not there is a gradient with respect to the line direction of a line-symmetric pattern using a gradient signal of -1°0, +1 with a positive or negative polarity sign depending on the gradient direction; and the gradient extracting means a gradient adding means for adding the gradient signals extracted by the gradient adding means for each scanning line; and a signal outputted from the gradient adding means for each scanning line, which is ignored when it is zero and is continuously detected with the same polarity. and determining means for determining that the line-symmetric pattern has a feature portion that is not line-symmetrical when the line-symmetric pattern recognizes the line-symmetric pattern.

[Example of idea]

Hereinafter, the present invention will be specifically explained based on embodiments shown in the drawings.

FIG. 4 is a diagram showing a schematic configuration of a line-symmetric pattern inspection apparatus according to the present invention.

As shown in the figure, <16 is an imaging device that scans and images a line-symmetrical pattern in a direction perpendicular to the line and detects a series of video signals along the scanning line, such as a photodiode array, a TV camera, etc. be.

Reference numeral 17 denotes a control signal generator, which outputs a scanning signal 18 for scanning an image of a line-symmetric pattern captured by the imaging device 16, and also outputs a horizontal synchronizing signal 19.

20 is a binarization circuit that thresholds the video signal detected from the imaging device at a predetermined threshold level and converts it into a binary video signal.

21 and 22 have the number of bits of one scanning line to create delayed binary video signals of one scanning line and two scanning lines, and - store the binary video signals obtained over the scanning lines. A shift register 23 is a shift register formed of 3×3 storage elements 23a to 23f that store binary video signals of 3×3 picture elements.

25 is a contour extraction circuit, and memory elements 23b, 23d,
This circuit performs the logical product negation of 23f and 23h, and further performs the logical product of the result of this logical product negation and the storage element 23e to extract the outline of a line-symmetric pattern.

26 is a shift register having the number of bits of one scanning line and storing the contour signal obtained over the -scanning line.

Reference numeral 28 denotes a shift register formed of 2×3 storage elements 23a to 23f for storing contour signals of 2×3 picture elements.

Reference numeral 30 denotes a gradient extracting circuit, and when the value stored in the storage element 28d is ``1'' and the values stored in the storage elements 28e and 28f are ``0'', ``1'' is stored in the storage element 28b. If stored, a slope value signal of +1 is extracted, and when the value stored in the storage element 28f is "1", the slope value signal is extracted from the storage element 28d.
and the value stored in 28e is "0".

When "1'° is stored in the storage element 28b, a slope value signal of -1 is extracted, and in other cases, a slope value signal of zero is extracted because there is no slope.

Reference numeral 31 denotes a gradient addition circuit which adds the gradient values output from the gradient extraction circuit 30 for each scanning line and outputs the result.

Reference numeral 32 denotes a determination circuit which determines that a defect exists in the line-symmetric pattern when the signal output from the gradient extraction circuit 30 is not zero and the same signal is continuously output.

FIG. 5 is a diagram showing a specific embodiment of an apparatus for scanning and imaging a line-symmetric pattern when the imaging device shown in FIG. 4 is a photodiode array.

As shown in this figure, numeral 34 is an electric circuit section shown in FIGS. 17 to 32, and 35 is a microscope, which is equipped with a photodiode array 33 at the upper end and an objective lens 37 at the lower end.

36 is an illumination device, 38 is a table, 39 is an object to be inspected on which a line-symmetrical pattern is drawn, 40 is a feed screw and nut portion of the table 38, and 41 is a motor for driving the table 38 at a constant speed.

Therefore, the object to be inspected 39 illuminated by the illumination device 36
The above line-symmetrical pattern is projected by the objective lens 37 onto the light receiving surface of the photodiode array 33 as a real image.

When the table 38 is moved at a constant speed by the motor 41, the image of the line-symmetric pattern projected onto the photodiode array 33 also moves at a constant speed.

Therefore, the photodiode array 33 electrically scans the line-symmetrical pattern sequentially in the direction in which the photoelectric conversion elements are arranged, and moves in a direction perpendicular to this arrangement direction to detect video signals of several scanning lines. .

Here, the photodiode array 33 has dozens to dozens of photoelectric conversion elements arranged on a straight line, and by sequentially extracting photocharges from each element, a bright and dark image on the straight line is converted into an electric signal. It is something that converts.

The imaging device configured with the photodiode array 33 and the like as described above scans in a direction perpendicular to the line of a line-symmetrical pattern and detects a video signal for each scanning line.

This video signal is converted into a binary video signal by the binarization circuit 20.

The binarized video signal from the binarization circuit 20 and the shift registers 21 and 22 transfer the two binarized video signals 1 and 2 scanning lines before the scanning line to be scanned to the shift register 23. is entered on each line.

At this time, if the ring of the line-symmetrical pattern is located at the picture element 23e, a signal of "1" is stored in the storage element 23e.

It is assumed that the line-symmetrical pattern is brighter than its surroundings, and that a binary video signal of "1'° is detected from the line-symmetrical pattern.

Next, the contour extraction circuit 25 extracts the binary video signal 24 stored in each of the storage elements 23a to 23i of the shift register 23.
a to 24i are read, a signal “1” is stored in the memory element 23e, and the memory elements 24b, 24d, 24f, 2
A signal of 0°' is stored in one of 4h (24
bX24dX24fX24h=0), memory element 23e
Assuming that there is an outline of a line-symmetric pattern in the picture element l
It outputs a signal of "0", and at all other times it outputs a signal of "0".

In this manner, a contour signal is extracted from the contour extraction circuit 27.

The shift register 26 has already stored the contour signal output from the contour extraction circuit 25 one scan line before the scan line to be scanned.

Therefore, the memory element 28a in the first row of the shift register 28
, 28b, 28c store the contour signal of each picture element output from the contour extraction circuit 25, and the second row memory elements 28 dv 28e, 28f of the shift register 28 store the contour signal output from the shift register 26. The contour signal of the previous scanning line is stored.

These contour signals 29a to 29f are read into the gradient extraction circuit 30, and the gradient extraction circuit 30 is in the state shown in FIG. is stored in the memory element 28b, and
When the signal ``1'' is stored, a slope value of +1 is output assuming that there is a positive gradient, and when the state shown in FIG. When the "0" signal is stored in the elements 28d and 28e, and the "1°" signal is stored in the storage element 28b,
A gradient value of -1 is output assuming that there is a gradient in the negative direction, and a gradient value of zero is output in other states, assuming that there is no gradient.

Next, the gradient addition circuit 31 adds the gradient values output from the gradient extraction circuit 30 for each scanning line, and outputs the result.

The determination circuit 32 ignores when the added gradient value is zero, and determines that there is a defect in the line-symmetric pattern when +1 gradient values or -1 gradient values do not appear alternately but appear consecutively. do.

The above process will be specifically explained for the case where there is a linear defect in a circular pattern which is a line-symmetrical pattern, as shown in FIG. 7a.

That is, the imaging device 16 images the pattern while scanning it, and detects a series of video signals.

Next, this video signal is converted into a binarized video signal by the binarization circuit 20 and applied to the shift register 23 via the shift register 23.
A binarized video signal consisting of picture elements is stored.

From these binary video signals, a contour 43 is extracted along a scanning line 42 by a contour extraction circuit 25, and is formed as shown in FIG.

Next, the picture elements of this outline are stored in the shift register 28,
When the contour is inclined in the direction of rotation of the clock with respect to the direction perpendicular to the scanning line 42, it is a white point as shown in FIG. 7e.
A gradient value of 1 is output from the gradient extraction circuit 30, and when the contour is inclined counterclockwise with respect to a direction perpendicular to the scanning line 42, a gradient value of -1, which is a black dot, is extracted as a gradient as shown in FIG. 7C. A gradient value of zero, which is unmarked when there is no gradient, is output from the gradient extraction circuit 30.

This gradient value is added for each scanning line 42 by the gradient adding circuit 31, and the result is shown as a white dot when it is +1, a black dot when it is -i, and an unmarked mark when it is zero, as shown in FIG. 7d.

As is clear from FIGS. 7a, b, c, and d, when the line-symmetric pattern is bilaterally symmetrical, the sum of the gradient values of one scanning line becomes zero.

Furthermore, even if the contour is shifted by one bit due to a quantization error, it is corrected by the bits of the adjacent portion, so that the sum of the gradient values always appears alternately as +1 and -1 as shown in FIG. 7d.

However, if a defect exists in the line-symmetrical pattern, the sum of the gradient values will continue to be +1 or -1, as shown in FIG. 7d.

That is, as shown in FIG. 7e, by detecting this continuous portion, the defective portion of the axisymmetric pattern can be detected.

[Effect of idea]

As explained above, according to the present invention, compared to the conventional method in which all two-dimensional patterns are stored in a memory and comparatively inspected by pattern matching, minute defects etc. can be detected for line-symmetric patterns using a simpler device configuration. It has the effect of being able to be recognized at high speed and with high sensitivity.

[Brief explanation of drawings]

FIG. 1 is a diagram for explaining a pattern matching method of a conventional symmetrical pattern inspection device, FIG. 2 is a diagram for explaining a 90° rotation pattern matching method of a conventional symmetrical pattern inspection device, and FIG. The figure is a diagram for explaining quantization error, FIG. 4 is a block diagram showing an embodiment of the symmetrical pattern inspection device of the present invention, and FIG. 5 is an example of the imaging device shown in FIG. 4. Figures 6a and b are diagrams shown to explain the operation of the gradient extraction circuit shown in Figure 4, and Figures 7a and b! FIGS. 3C and 3E are diagrams illustrating signals output from the contour extraction circuit, gradient extraction circuit, gradient addition circuit, and determination circuit, respectively, in the form of patterns in the case of a circular pattern as a line-symmetric pattern. 16... Imaging device, 20... Binarization circuit, 23... Shift register, 25...
・Contour extraction circuit, 2...Shift register, 30
... Gradient extraction circuit, 31 ... Gradient addition circuit, 32 ... Judgment circuit.

Claims (1)

[Scope of utility model registration request] An imaging device that detects a video signal by scanning a line-symmetrical pattern with a plurality of scanning lines in a direction perpendicular to the line, and a binarization device that converts the video signal detected by the imaging device into a binary video signal. The circuit and the binarized video signal output by the binarization circuit are cut out into two-dimensional picture elements and stored, and when it is recognized that there is a slope in the surrounding picture elements with respect to the central picture element, line symmetry is established. Assuming that the pattern has a contour, there is a contour extraction means for extracting it using a binary signal, and whether there is a slope in the line direction of the line-symmetrical pattern in adjacent scanning lines of the binary contour signal extracted from the contour extraction means. gradient extracting means for extracting the difference with a positive or negative polarity sign according to the gradient direction as a gradient signal of -L Ot 1,000, and adding the above gradient signals extracted by the gradient extracting means for each scanning line. Gradient addition means, and when the signal output from the gradient addition means for each scanning line is ignored when it is zero and is continuously detected with the same polarity, there is a feature part that is not line symmetric in the line symmetric pattern. 1. A recognition device for a line-symmetric pattern, characterized in that it is equipped with a determining means for determining.