JP2787851B2 - Pattern feature extraction device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern feature extracting apparatus,
In particular, the present invention relates to a pattern feature extraction device used for inspecting a wiring pattern on a printed circuit board.

[0002]

2. Description of the Related Art As a conventional technique, for example, "printed board inspection using radial matching", Journal of the Japan Society of Precision Engineering, Vo
l. 56, No. 8, pp. 30 to 33, 1990, there is a pattern feature extraction device using a dedicated length measurement sensor and a pattern code.

FIG. 7 is a diagram for explaining a feature extracting means of a feature extracting apparatus using the above-described length measuring sensor. FIG.
(A) represents a length measuring sensor, and the length measuring sensor is the inspection center 7.
It consists of 16 length-measuring pixels 72 extending radially with respect to 1. Each measured pixel 72 is applied to a target pattern by pairing the measured pixel 72 symmetrical with respect to the inspection center 71, and codes the data measured by each measured pixel 72. Next, the coded data is compared with the coded dictionary, and the shape of each point is determined to classify each point. FIGS. 7B and 7C are schematic diagrams illustrating an example of classification by the length measurement sensor. In FIG. 7B, upper and lower length measurement pixels 73 are shown.
Measurement data is asymmetric, the left and right measurement pixels 74
In this case, the result obtained by referring to the dictionary code is determined to be a T-branch because the measurement data of the target is the target and the length is the maximum. FIG. 7C shows a case in which the branch is not a T-branch. The symmetry of each measured pixel and the code of the length data are compared with the dictionary code. This shows the case where it can be determined.

[0004]

The feature extracting means used in the above-mentioned conventional pattern feature extracting apparatus uses a length measuring sensor extending in a radial pattern, and uses a pair of measurement data of each length measuring pixel and a dictionary code. Since feature points are classified by comparison, it is necessary to prepare a dictionary code corresponding to a complicated pattern when a complicated pattern is targeted. Further, when the measurement is performed using this length measuring sensor, only the wiring pattern portion is effective, and there is a problem that the feature extraction of another means must be used for the land portion and the sub-brand portion.

[0005]

According to the present invention, there is provided a pattern feature extracting apparatus, comprising: a photoelectric conversion scanner for optically scanning a target pattern and outputting a video signal corresponding to the reflected light; and a video signal from the photoelectric conversion scanner. A / D converter for converting into a digital signal, and binarization for converting the target pattern portion of the digital signal converted by the A / D converter into binary data of "1" and the rest of the digital signal into "0" Circuit and this 2
A binary image memory for storing binary data output from the binarizing circuit as image data, a thinning circuit for thinning the image data of the binary image memory, and an output from the thinning circuit There is a thin line image memory that stores thin line image data, and a thin line image memory in which a mask having a size of 3 × 3 pixels is applied to the thin line image memory, and data inside the mask matches a preset feature candidate point mask pattern. In this case, a feature candidate point extracting unit that outputs the type of the corresponding feature candidate as a code, and an N × N memory for the thin line image memory simultaneously with the processing of the feature candidate point extracting unit
The number of intersections between the rectangular mask and the thin line data of the target pattern portion is calculated, and the result is output from the feature candidate point extracting portion. A feature point determination unit configured to accurately determine and output the position and type of the feature point of the pattern from the type of the feature candidate point and the number of branches output from the number-of-branches calculation unit; A photoelectric conversion scanner for optically scanning a pattern and outputting a video signal corresponding to the reflected light, an A / D converter for converting a video signal from the photoelectric conversion scanner into a digital signal, and an A / D converter
2. Convert the target pattern portion of the digital signal converted by the converter into binary data of "1" and the other portions into "0".
A binarizing circuit, a binary image memory for storing binary data output from the binarizing circuit as image data, a thinning circuit for thinning the image data of the binary image memory, and a thin line A thin line image memory for storing the thin line image data output from the digitizing circuit, an N × N rectangular mask applied to the thin line image memory, and an intersection between the rectangular mask and the thin line image data is calculated. A branch point extraction unit that extracts a point where the center pixel of the mask is in the vicinity of the pattern branch point, estimates the type of the feature point from the number of branches, and classifies and outputs the candidate points, and a candidate point obtained by the branch point extraction unit. A candidate point memory for storing the type and position, and clustering the data of this candidate point memory with respect to position and type, determining the center of gravity of each class, and determining the feature point position, Sometimes it constituted by including a clustering section which determines and outputs that type.

[0006]

Next, the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of one embodiment of the present invention. In FIG. 1, the target pattern is a photoelectric conversion scanner 1.
The A / D converter 2 converts the video signal d scanned and read out by the A / D converter 2 into a digital signal e. Next, the pattern portion of the digital signal e is "1" by the binarization circuit 3.
Other than that, it is set to binary data f of “0” and sent to the binary image memory 4 to obtain a binary image having a pattern portion of “1”. Subsequently, the thinning circuit 5 sequentially reads out the binary image data g from the binary image memory 4, performs thinning, sends the resulting thin line image data h to the thin line image memory 6, and obtains a thin line image of the target pattern. .

FIG. 3 is a schematic diagram showing an example of a pattern stored in the binary image memory 4. The pattern shown in FIG. 3 includes a wiring pattern 100 on a printed circuit board, a land 101, and a subland 102, and includes end points T, T branch points TJ, four branch points 4J, land connection points R, and isolated points I as characteristic points. Have been.

FIG. 4 shows the pattern shown in FIG.
FIG. 4 is a schematic diagram of a thin line image when thinning is performed by using FIG.
The thinning performed by the thinning circuit 5 is performed when the wiring pattern portion 100 has data of one pixel width and the land portion 10
With regard to 1, the processing is stopped in a state where the shape is preserved.

Next, the thin line image data o is sequentially read from the thin line image memory 6 by the feature candidate point extracting section 7, and the feature candidate points of the pattern are detected using a 3 × 3 pixel size mask for detecting the feature candidate points. On the other hand, at the same time, the branch number calculation unit 8 sequentially reads the thin line image data o from the fine line image memory 7 and calculates the number of branches of the pattern using a rectangular mask of N × N pixel size for calculating a branch point.

FIG. 5 is a view for explaining processing using a 3 × 3 pixel size mask for detecting feature candidate points and a N × N pixel size rectangular mask for calculating the number of branches. FIG. 5A shows a case where a mask 201 having a size of 3 × 3 pixels and a rectangular mask 202 having a size of N × N pixels are applied to the T-branch pattern portion, and FIG. FIG. 5 (c) shows the case of FIG.
5A shows the in-mask data 301 and 302 obtained in the T branch pattern portion, and FIG. 5D shows the in-mask data 301 and 302 obtained in the land connection point portion in FIG. 5B.

The feature candidate point extracting unit 7 applies a mask 201 having a size of 3 × 3 pixels to the thin line image data o to be read, and obtains data 301 within the mask. Here, the feature candidate point mask pattern is compared with a preset feature candidate point mask pattern, and a feature candidate point code r of the corresponding feature candidate point type is generated and sent to the feature point determination unit 9. The feature candidate point mask pattern that is set in advance is composed of a mask pattern that may occur for each of the end point pattern, the branch point pattern, the land connection point pattern, and the isolated point pattern by one logic circuit. Data 3 in the mask to be data
When 01 matches one of the feature candidate point mask patterns, a code of the type of the feature candidate point is output.

On the other hand, the branch number calculating section 8 applies a rectangular mask 202 of N × N pixels size to the read thin line image data o around the mask 201 used in the feature candidate point extracting section 7 to form a rectangle. The in-mask data 302 is obtained. Then, the number of areas where the data of the rectangular mask data 302 is “1” (the same area when they are continuous) is counted, and the rectangular mask 202 and the wiring pattern 10 are counted.
0, the intersection of the land 101 and the sub-brand 102 is obtained, and the number of branches of the pattern of the central pixel 103 is calculated. The number of branches for the T branch point shown in FIG.
By counting the area of "1" in the rectangular mask data 302 of (c), it becomes 3. The number of branches for the land connection point in FIG. 5B becomes 2 by counting the area of “1” in the rectangular in-mask data 302 in FIG. 5D. At the same time, it is also checked whether or not there are three or more consecutive “1” s in the “1” area of the rectangular mask data 302. Data 302 in the rectangular mask of FIG.
In some cases, three or more "1" s are continuous. Then, as an output of the branch number calculation unit 8, the branch number s1 and a continuity signal s2 indicating whether or not there is an area where three or more "1" s are continuous are sent to the feature point determination unit 9.

Finally, the feature judging section 9 calculates the feature candidate point code r sent from the feature candidate point extracting section 7 and the branch number s1 and the continuous point presence / absence signal s2 sent from the branch number calculating section 8. , The type of the feature point is determined based on the following determination conditions, and the feature point extraction result t is output.

The determination conditions are as follows: (a) When the feature candidate point is an isolated point and the number of branches is 0, the isolated point is determined. (B) When the feature candidate point is an end point and the number of branches is 0, it is an isolated point. (C) When the feature candidate point is an end point and the number of branches is 1, the end point. (D) When the feature candidate point is a branch point and the number of branches is 3, it is a T branch point. (E) Four branch points when the feature candidate point is a branch point and the number of branches is four. (F) If the feature candidate point is a land connection point and the number of branches is 2 and there are continuous points, it is a land connection point. (G) When the conditions of (a) to (f) are not satisfied, it is not a feature point. It is.

FIG. 2 is a block diagram of another embodiment of the present invention. In the embodiment shown in FIG. 2, a thin line image of the target pattern is also obtained in the thin line image memory 6.
This embodiment is the same as the embodiment described above, and also has the same pattern types and thinning techniques described with reference to FIGS.

Next, the branch point extracting unit 10 controls the fine line image memory 6.
, The thin line image data o are sequentially read from
The position and the type of the feature candidate point of the target pattern are obtained using a rectangular mask having a size of N pixels.

FIG. 6 is a diagram for explaining the processing performed by the branch point extraction unit 10 using a rectangular mask of N × N pixels. FIG. 6A is a schematic diagram when a rectangular mask is applied to a T branch point portion, and FIG. 6B is a schematic diagram when a rectangular mask is applied to a land connection point portion. 6 (d)
7A and 7B are diagrams showing the contents of data in a rectangular mask in the cases of FIGS. 6A and 6B, respectively. 6 (c) and FIG. 6 (d) are obtained by dividing the thin line image data o read from the thin line image memory 6 by N ×
It is obtained by applying a rectangular mask 202 of N pixel size.

Here, the branch point extracting unit 10 counts the area where the data of the rectangular mask data 302 is "1", as in the feature candidate point extracting unit 7 of FIG. , Land 101 and Subland 10
Then, the intersection of the center pixel 103 is obtained and the branch state of the pattern of the central pixel 103 is examined. At this time, if the number of intersections is 1, the end point T is set as a feature point candidate. If there is an area where the number of intersections is 2 and three or more “1” s are continuous, the land connection point R is set. If T, the T branch point is TJ and the number of intersections is 4.
If the number of intersections is 0, it is determined to be an isolated intersection I if the number of intersections is 0. Further, the branch point extracting unit 10 sequentially stores the candidate point type r1 which is the type of the characteristic candidate point and the candidate point position r2 which is the position of the central pixel 103 at that time in the candidate point memory 11 as the examined branch state. FIG. 6A shows a case where the rectangular mask 202 is applied to the T branch point. At this time, the number of “1” areas in the rectangular mask data 302 in FIG. Therefore, the candidate point is determined to be a T branch point and stored in the candidate point memory 8. Similarly, in FIG. 6B, the area of “1” in the data 302 in the rectangular mask of FIG. 6D is 2, and there are areas where three or more “1” s are continuous. Are determined as land connection points and stored in the candidate point memory 8.

Finally, the clustering unit 12 clusters the candidate point position r2 in the candidate point memory 8 and positions each class as a feature point. Since the types r1 of the feature points of the candidates in each class are the same, many types of the feature points are represented as the types of the feature points of the class. At the same time, from the candidate point position r2 belonging to each class,
The positions of the centers of gravity are shown, and the point on the thin line image data closest to the position of the center of gravity is defined as a position representing each class. The representative type and position are output as a final feature point extraction result.

[0021]

As described above, according to the present invention, an N × N pixel-size rectangular mask for extracting a branch point or a 3 × 3 pixel-size mask for detecting a feature candidate point is provided on an image obtained by thinning a target pattern. The feature points are extracted by applying a rectangular mask of N × N pixel size for calculating the number of branches. Therefore, the conventional method can be applied to complicated patterns including lands and sub-brands in addition to wiring patterns. There is an effect that feature points can be accurately extracted without making a dictionary corresponding to the pattern.

[Brief description of the drawings]

FIG. 1 is a block diagram of one embodiment of the present invention.

FIG. 2 is a block diagram of another embodiment of the present invention.

FIG. 3 is a schematic diagram showing an example of a pattern stored in a binary image memory in FIGS. 1 and 2;

FIG. 4 is a schematic diagram of a thin line image obtained by thinning the pattern shown in FIG. 3;

FIG. 5 is a diagram for explaining processing using a mask used in the embodiment of FIG. 1;

FIG. 6 is a diagram for explaining processing using a mask used in the embodiment of FIG. 2;

FIG. 7 is a diagram for explaining a feature extraction unit of a conventional feature extraction device.

[Explanation of symbols]

 Reference Signs List 1 photoelectric conversion scanner 2 A / D converter 3 binarization circuit 4 binary image memory 5 thinning circuit 6 thin line image memory 7 feature candidate point extraction unit 8 branch number calculation unit 9 feature point judgment unit 10 branch point extraction unit 11 Candidate point memory 12 Clustering unit 100 Wiring pattern 101 Land 102 Subland o Fine line image data r Feature point candidate code r1 Candidate point type r2 Candidate point position s1 Number of branches s2 Continuous point existence signal t Feature point extraction result

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06T 7/60 G01B 11/24 G01N 21/88 G06T 7/00

Claims (2)

(57) [Claims] 1. A photoelectric conversion scanner that optically scans a target pattern and outputs a video signal corresponding to the reflected light, an A / D converter that converts a video signal from the photoelectric conversion scanner into a digital signal, A binarization circuit for converting the target pattern portion of the digital signal converted by the A / D converter into binary data of "1" and the other portions into "0";
A binary image memory for storing binary data output from the binarizing circuit as image data, a thinning circuit for thinning the image data of the binary image memory, and an output from the thinning circuit There is a thin line image memory that stores thin line image data, and a thin line image memory in which a mask having a size of 3 × 3 pixels is applied to the thin line image memory, and data inside the mask matches a preset feature candidate point mask pattern. In this case, a feature candidate point extracting unit that outputs the type of the corresponding feature candidate as a code, and an N × N memory for the thin line image memory simultaneously with the processing of the feature candidate point extracting unit
The number of intersections between the rectangular mask and the thin line data of the target pattern portion is calculated, and the result is output from the feature candidate point extracting portion. A pattern feature extraction unit comprising: a feature point determination unit that accurately determines and outputs the position and type of a feature point of a pattern from the type of feature candidate point and the number of branches output from the branch number calculation unit. apparatus. 2. A photoelectric conversion scanner that optically scans a target pattern and outputs a video signal corresponding to the reflected light, an A / D converter that converts a video signal from the photoelectric conversion scanner into a digital signal, A binarization circuit for converting the target pattern portion of the digital signal converted by the A / D converter into binary data of "1" and the other portions into "0";
A binary image memory for storing binary data output from the binarizing circuit as image data, a thinning circuit for thinning the image data of the binary image memory, and a thin line output from the thinning circuit A thin line image memory for storing image data, an N × N rectangular mask is applied to the thin line image memory, and an intersection between the rectangular mask and the thin line image data is calculated, whereby the central pixel of the rectangular mask is subjected to pattern branching. At the same time as extracting a point near the point, a branch point extracting unit for estimating the type of the feature point from the number of branches and classifying and outputting as a candidate point, and storing the type and position of the candidate point obtained by the branch point extracting unit The candidate point memory and the data of the candidate point memory are clustered with respect to position and type, the center of gravity of each class is determined, and the feature point position is determined.
A clustering unit for determining and outputting the type of the pattern at the same time.