JPS63674A - Pattern inspection method - Google Patents

Abstract

PURPOSE:To stably inspect a pattern with high accuracy by using not only a measured data but also a correlation operation data between different window areas, for deciding whether the pattern is normal or defective. CONSTITUTION:A part of a pattern 20 is seen through each of window areas 21-27. First of all, the electric signal of a time series obtained by bringing an object pattern 50 to image pickup by an ITV camera 51 is binary-coded and divided into picture elements (dots) by a binarization/picture element dividing circuit 52, and a feature extracting circuit 53 receives a dot signal from the dividing circuit 52 and extracts its feature. Subsequently, a secondary decision processing by a feature quantity data measured directly through the window area, and a correlation secondary decision processing by a correlation operation data obtained by working a measured data area executed, and when a result shows a non-defective by one of them, an overall deciding circuit 62 outputs the result of decision as a non-defective.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a pattern inspection method for determining the quality of a pattern to be inspected.

[Conventional technology]

FIGS. 9(a) and 9(b) are diagrams explaining the principle of a conventional pattern inspection method called a template matching method, respectively. In the figure, 1 and 1' are patterns to be inspected, respectively, and 2 to 8 are sampling points indicated by cross marks.

In FIG. 9(a), the pattern to be inspected 1 is the character 8.
Suppose it was. At this time, using a template (not shown), sampling points 2 to 8 are determined as shown in the drawing, and the presence or absence of a pattern at each point is inspected. For example, if there is a pattern, it is set to logic 1, if there is no pattern, it is set to logic 0, the inspection results at each point 2 to 8 are coded, and from that code it is recognized which known pattern the pattern to be inspected is classified into. be.

In FIG. 9(a), the test result is positive (logical 1) at any of points 2 to 8, and in this case, the test target pattern 1 is recognized as the character 8. Figure 9 (b
), only point 8 has no pattern (logical 0)
In this case, an algorithm for determining that the pattern is a character 0 is determined in advance, and the character is recognized as 0 according to the algorithm.

In this pattern inspection method using template matching, the information to be examined is compressed and small (only the presence or absence of a pattern at a sampling point);
It has the advantage that it is easy to classify patterns based on the results of examination, and it is easy to classify an unknown pattern to be inspected into a standard (known) pattern, such as OCR (optical character reader rJ). It is known that it is effective if used in a pattern recognition device that can perform classification by determining whether the pattern corresponds to a pattern.

[Problem that the invention seeks to solve]

However, when the template matching method is applied to a pattern inspection method, several problems arise. One of them is that when testing the quality of a pattern, we consider a window area that extends the sampling point, and if we judge whether or not a pattern exists within the window area, we can identify a similar but defective pattern as a good item. This is because there is a fatal flaw that causes false judgments. This is because in the binary method of determining whether a pattern exists or not, if a pattern does not exist within the window area, it is determined to be defective, but even though a pattern exists, its size is different from that of a non-defective product. This is because if the product is found to be defective, it cannot be determined that the product is defective.

In general printed patterns, the area occupied by the pattern part in the window area varies greatly depending on the location, and in each window area, the pattern part has a special number (
In this case, since there are large and small areas), correct pattern inspection cannot be performed unless this special feature is extracted and measured.

The present invention has been made in view of the above-mentioned conventional technical circumstances, and therefore, an object of the present invention is to develop the template matching method based on the principle of the template matching method, and to apply it to general printing patterns, etc. It is an object of the present invention to provide a pattern inspection method that can also correctly inspect patterns.

[Means for solving problems]

A plurality of window areas are set within the field of view of the two-dimensional sequential scanning photoelectric conversion device, and a combination of each window area is set in advance. The special quantity of the pattern to be inspected is measured, the special quantity of each window region belonging to the same group is calculated by correlation, and the quality of the pattern to be inspected is determined based on the result of the correlation calculation.

[Effect]

First, the pattern to be inspected is photographed using a two-dimensional sequential scanning photoelectric conversion device (for example, an industrial television camera), and the obtained time-series electrical signals are processed to determine the quality of the pattern. A plurality (generally a large number) of window regions are set within the field of view by electrical means or the like, and the features of the pattern portion viewed from each window region are quantitatively extracted.

Even if the pattern to be inspected is a good product, if it deviates from the predetermined position within the camera field of view, the special quantity extracted through the predetermined window area may not fall within the predetermined threshold range. , As a result, the judgment result is defective. However, since the pattern is only slightly shifted to the adjacent one, if we take the correlation between the special quantity detected through a given window area and the % value detected through the adjacent window area (in this case, the correlation and (addition, subtraction, etc.), the obtained correlation amount falls within a predetermined threshold range, and the judgment result is good. Based on this idea, we determine appropriate combinations of related window areas, perform arithmetic processing (for example, addition) on each special @ quantity between them, and make judgments about the processing results (this is called correlation judgment). , and determine whether the product is good or bad based on the results.

The quality of the pattern to be inspected is determined by performing a correlation determination as shown in FIG. Note that the shape, size, etc. of the window area may be determined arbitrarily and appropriately according to the pattern to be inspected, and are not limited to a rectangle or the like. Further, as the special quantity extracted through the window area, in addition to the area of the pattern part within the window area, the circumferential length of the pattern part, and other scales may be used.

〔Example〕

The present invention will now be described in detail with reference to the drawings.

FIG. 2(a) is an explanatory diagram showing an example of setting a window area in the present invention, FIG. 2(b) is an explanatory diagram of an example of a pattern to be inspected, and FIG. FIG. 3 is an explanatory diagram of a pattern seen through a window area.

In these figures, 21 to 27 are rectangular window areas, and 20 is a pattern to be inspected (letter 8). As seen in FIG. 2(C), window areas 21-2
7, a portion of pattern 20 is visible.

FIG. 6 is an explanatory diagram showing another example of setting the window area. In the same figure, the same window area 21 as shown in FIG. 2(a) is shown.
The window areas 31 to 36 are set outside of 27, and the window areas 41 to 48 are set inside. However, if the pattern is misaligned, the window areas 21 to 36 in the center 2
7) and the outside or inside window area.

FIG. 1 is a block diagram showing one embodiment of the present invention. In the figure, 50 is a pattern to be inspected, 51 is an industrial television camera (I'rV camera), 52 is a 2-cell conversion/pixel division circuit, 53 and 53' are feature extraction circuits, and 54.
54' is a counting circuit, 55 is a counting direct memory circuit, 5
6 is a primary determination circuit, 57 is a secondary determination circuit, 58 is a correlation determination necessity circuit, 59 is a correlation calculation circuit, 60 is a correlation-order determination circuit, 61 is a correlation secondary determination circuit, 62 is a general assembly determination circuit,
63 is a window area generation circuit, 64 is a control circuit, 65 is a set value storage circuit, and 66 is a keyboard.

Next, an overview of the operation will be explained. First, the target pattern 50 is
A time-series electrical signal obtained by imaging with a TV camera 51 is divided into two in a 21i/fiii element division circuit 52 and divided into pixels (dots).

Normally, one screen's worth of electrical signals is divided into 620 dots in the horizontal X-axis direction and 240 dots in the vertical Y-axis direction, for a total of about 77,000 dots. The feature extraction circuit 53 receives the dot signal from the division circuit 52 and extracts its features. For example, if the characteristic is simply the size of the area, focus on the white dot signal, or the black dot signal, and calculate the length (number of dots) of the dot signal opening in the lateral and horizontal settings using the following counting circuit. The area is determined by counting at step 54 (the area is determined by a set of the lengths of the tabs). 53' is a circuit for extracting features other than area; for example, when detecting the boundary length between black and white, it is a circuit for extracting the boundary point;
The boundary length is determined by counting the number of boundary points by the next counting circuit 54'. Note that under the control of the window area signal generated from the window area generating circuit 63, the counting circuit 54.5
4' performs a counting operation, so the GF number results are not necessarily the special quantities extracted for each window area, but are organized by window area number and special item and stored in the memory circuit 55 once. be remembered. The window area generation circuit 65 is connected to the keyboard 66
Based on the data input to the memory circuit 65, the settings can generate and output signals representing a large number of concentrated window areas of arbitrary shapes.

Now, when the scanning of one field by the ITV camera 51 is completed, the measurement and storage of characteristic data for each condition area is completed, and based on these characteristic data, the following circuits 56 to 65 determine whether the target pattern is good or bad. judge.

The feature data for each window area read out from the memory circuit 55 is processed by the next determination circuit 56 to an upper limit setting value αij and a lower limit setting value βij set for each window area and each feature (however,
(i represents the type of feature and j represents the number of the window area), and a primary determination is made as to whether or not it is within the upper and lower limit range (threshold range). The determination processing operation for the feature item number one will be specifically explained.

The primary determination circuit 56 performs a primary determination by checking whether data DBj in the window area j regarding the feature item i (for example, the size of the area in the case of a planar shape) is within the range of the upper limit setting value αij and the lower limit setting value β1j. However, this is performed for the entire window area. - The next determination result is expressed as logic 1 if it is within the range of the upper and lower thresholds, and logic 0 if it is outside the range.

The upper and lower thresholds can be selected appropriately for each window area and each % desire item.A) These threshold data are input from the keyboard 66 and stored in the set value storage circuit 6.
5, the -next determination circuit 56 reads and uses this.

- In general, the pattern to be inspected can be considered to be composed of N small patterns (N is an arbitrary integer), so each window area has N small patterns corresponding to N small patterns. organized into groups.

For example, if the pattern to be inspected is rA B CDJ, it is considered that it consists of four small patterns (each character constitutes a small pattern). Note that the small pattern is not necessarily composed of one character or one figure, but may be composed of multiple characters.

Keyboard 66 for grouped window area numbers
and stored in the set value storage circuit 65. The secondary determination circuit 57 reads out the grouped combinations of window area numbers from the storage circuit 65, and divides the primary determination results for each window area from the primary determination circuit 56 into groups accordingly. Next, the secondary determination circuit 57 compares and collates the grouped primary determination results with the primary determination table sequentially output from the set value storage circuit 65, thereby determining the quality of the group. Note that the next determination table is determined in advance for each target pattern and stored in the set value storage circuit 65.

FIG. 4 shows a primary determination table when the window areas set as shown in FIG. 6 are used and the numbers 8 (FIG. 2 6) and 7 are used as target patterns.

When the target pattern is the number 8, the -th order determination table is the coded data (1
1111110...0] and the number 7 is C1110000...0]. It will be understood from FIG.

The secondary determination operation by the secondary determination circuit 57 is performed for all groups. Poor secondary judgment as a group (N
G), the correlation determination safety circuit 58 determines whether correlation determination is necessary. The correlation determination necessity circuit 58 determines that a correlation determination is necessary if the condition is met, and determines that the correlation determination is not necessary, if it is true, according to various preset conditions. When the correlation judgment is low, the present invention is applied and the correlation calculation circuit 59 extracts data for the set combination of window areas (data regarding this combination is subsequently obtained from the storage circuit 65). Calculate the feature data (for example, add the data of the window area 21 and the data of the window area 31), and use the result as new data DCj.
Output as k.

The new data DCjk obtained by the calculation is sent to the correlation-next determination circuit 60, where it is determined whether or not the upper limit setting value αC4K' is within the range of the lower limit setting value βCjk successively outputted by the setting value storage circuit 65. A correlation-order determination is made, and if it is within the range, a logic 1 is output, and if it is outside the range, a logic O is output. Such correlation-order determination processing is performed for all specified combinations of window areas.

Next, in the correlation secondary determination circuit 61, the correlation -
Correlation between the soft decision result and the set value storage circuit 65
A correlation secondary determination is performed by comparing and collating with the next determination table, and if -, the determination for the feature item of the group is good.

Using the window area set as shown in Figure 3, the number 8
FIG. 5 shows an example of a correlation-order determination table obtained when the position is slightly upwardly shifted from the relative position with respect to the window area shown in FIG. 2(c). In FIG. 5, since the pattern 8 has shifted slightly upward, there are few features extracted in the window area 21 alone, and the data cannot reach logic 1 and remains at logic 0. On the other hand, window areas 21 and 31 It will be understood that the data finally becomes logical 1 by the sum of the feature quantities. The same holds true for the window areas 24, 27, respectively. The correlation-secondary judgment table is also obtained in advance and stored in the set value storage circuit 65 using the keyboard 66, as in the case of the first-order judgment table described above.

As a result of the above, secondary judgment processing using feature data directly measured through the window area and correlation secondary judgment processing using correlation calculation data obtained by processing the measurement data are performed.
If either of the results shows that the product is good, the comprehensive judgment circuit 62 outputs the judgment result that the product is good.

FIG. 6 shows the operating mechanism of the embodiment shown in FIG. 1, and FIG. 7 shows the details of the operation as a flowchart. Furthermore, the seventh
In the figure, the window area is simply expressed as a window.

In the explanation of the actual examples described above, the shape of the window area was explained as a rectangle, but it may be any other shape such as a quadrangle, a pentagon, a mouth corner polygon, a circle, an annular shape, etc. In particular, the pattern to be inspected is a character. If it is a figure instead of
It is important to select a window area with an appropriate shape.

Furthermore, in the description of the embodiment, when performing the correlation determination process,
Although it has been described that the correlation secondary determination is performed using only the correlation-order determination result, the correlation-order determination result may also be referred to in addition to the correlation-order determination result. Furthermore, although an example of addition has been described as a correlation operation, depending on the pattern conditions, there may be cases where it is advantageous to take a difference or ratio. Next, in the embodiment described above, correlations were taken only for feature item i, but depending on the conditions of the pattern, there may be a significant advantage in taking correlations between different feature items. An operational flowchart for such a case is shown in FIG.

In the above description, area and boundary length have been described as feature items, but local surface information such as intersections, end points, arcuate points, diagonal lines, etc. can also be used.

As described above, in the device of the embodiment, conditions such as the shape, position, number, etc. of the window area, each hammer threshold, correlation conditions, various judgment conditions, etc. can be freely entered by keyboard input depending on the target turn. It can be set in the memory circuit. Therefore,
Conventional pattern inspection equipment had poor productivity as the hardware and algorithms differed significantly depending on the target pattern.
The apparatus of this embodiment has extremely high adaptability to changes in the pattern to be inspected, and therefore can be mass-produced.

〔Effect of the invention〕

Next, the effects of the present invention will be described.

(2) Window areas can be set in areas where no pattern exists, as shown at 31 to 48 in FIG. 3, instead of in areas where pattern components exist as in the conventional method. Therefore, it was previously impossible [inspection pattern = non-defective pattern +
[alpha]] can also be easily determined to be defective.

■ Window areas can be grouped and each group can be judged independently, so even when there are variations (minor positional deviations or density changes) in good samples, there is no waste spring (mistaking a good product as a defective one). ) will decrease.

(2) Patterns can be inspected with high precision and stability by using not only measured data but also correlation calculation data between different window areas to determine whether a pattern is acceptable or not. In particular, in non-defective samples, the absolute value of the measured value often changes depending on the location, transportation conditions, etc. In such a case, the method of comparing measurement accuracy using upper and lower limit values will result in a large number of wasted springs or a poor detection rate of defects. However, even under such conditions, there are cases where ``the relative ratio of measurement values of different window areas is constant'', ``the measurement ratio of different window areas is constant'', and ``the relative ratio of measurement values of different window areas is constant''. In many cases, there is a relationship such as "If you take the difference, the change in pattern becomes clear", so it is possible to perform highly accurate inspection using such conditions.

■ When taking correlations, it may be possible to obtain a significant effect by examining not only data on the same feature but also data on other features (even the same window area). For example, it is good to take a correlation between the area and the difference in perimeter length, as in the case of stain inspection.
The difference between defects becomes clear.

(2) By making the shape of the window area substantially arbitrary, it can be applied to various target patterns, and the number of window areas can be reduced, making it possible to perform high-speed determination.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention, and FIG. 2 (
a) is an explanatory diagram showing an example of setting a window area, FIG. 2(b) is an explanatory diagram of an example of a pattern, FIG. 2(c) is an explanatory diagram of a pattern seen through the window area, and FIG. 3 is an explanatory diagram of the window area. Fig. 4 is an explanatory diagram showing other setting examples, Fig. 4 is an explanatory diagram of the primary judgment table, Fig. 5 is an explanatory diagram showing another setting example
6 is an explanatory diagram showing an example of a correlation-order determination table, FIG. 6 is a flowchart showing an outline of the operation of the embodiment shown in FIG. 1, and FIG.
8 is a flowchart showing the details of the operation, FIG. 8 is a flowchart showing another example of the operation, and FIGS. 9(a) and 9(b) are diagrams explaining the principle of the conventional pattern inspection method. Code explanation 1.1'...Pattern, 2-8...Song sampling point, 20...Target pattern, 21-
47... Window area, 50... Target pattern, 51... Engineering TV camera, 52...
2 cage/pixel division circuit, 53.53'...Feature extraction circuit, 54.54'...Counting circuit, 55.
...Count value storage circuit, 56...-Next judgment circuit, 57...Secondary judgment circuit, 58...
・Correlation judgment safety circuit, 59...Correlation calculation circuit,
60... Correlation-order judgment circuit, 61...
Correlation secondary judgment circuit, 62... General judgment (b) path, 63... Window area generation circuit, 64...
Control circuit, 65... Setting value storage circuit, 66...
...Keyboard agent Patent attorney Akio Namiki Agent Patent attorney Kiyoshi Matsuzaki

Claims (1)

[Claims] 1) A time-series electrical signal obtained by scanning the pattern to be inspected with a two-dimensional sequential scanning photoelectric conversion device is divided into pixels and binarized, and the shape of the pattern to be inspected is inspected from the binarized signal. In a pattern inspection method, a plurality of window areas are set within the field of view of a two-dimensional sequential scanning photoelectric conversion device, and a combination of each window area is set in advance, and a two-pupil signal corresponding to each window area is generated. A certain feature amount of the pattern to be inspected in each window area is measured from , a correlation calculation is performed on the feature amounts of each window area belonging to the same group, and the quality of the pattern to be inspected is determined based on the result of the correlation calculation. A pattern inspection method characterized by determining.