JP2825359B2 - Pattern recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for detecting the position and inclination of a surface-mounted electronic component by linearly approximating a boundary point between the object and a background in an image picked up by an image pickup means. The present invention relates to a pattern recognition method for detecting an outline of a pattern.

[0002]

2. Description of the Related Art An example of a conventional pattern recognition method for detecting the inclination and center of an object (QFP) using linear approximation will be described below with reference to FIGS.

In an image 11 as shown in FIG. 5A, a coarse center detection area 12 is set in a horizontal direction and a vertical direction. Then, the center in the horizontal direction is obtained by searching from both ends of the coarse center detection area 12 in the horizontal direction and detecting both ends in the horizontal direction of the object. In addition, the vertical center is obtained by searching from both ends of the vertical coarse center detection area 12 and extracting the vertical both ends of the object, and the rough of the object is obtained from the horizontal center and the vertical center. The center 13 can be detected. From the coarse center 13 obtained above and the dimensions of the object, FIG.
(b) As shown in FIG.
Are set to the four sides of the object (QFP). Next, the detection area
By searching the lead 15 from the direction in which the tip of the lead 14 can be detected, the boundary point 16 can be extracted as shown in FIG. By applying the least squares method to the boundary points 16 obtained as described above, a straight line as shown in FIG. 6B can be obtained. If the distance from the straight line to each boundary point 16 is more than a predetermined reference distance, the boundary point 16 is regarded as a noise point and removed, and the least square method is applied to the boundary point 16 within the reference distance. By doing so, an approximate straight line 17 as shown in FIG. 6 (c) can be detected. By performing the above processing on the four sides of the object (QFP), the four sides as shown in FIG.
The approximate line 17 of the side can be detected. Then, an intersection is obtained between the straight lines adjacent to the four approximate straight lines 17,
Finding the center of gravity of the intersection is the detection of the center position of the object.

[0004]

However, with the above configuration, it becomes difficult to detect an accurate approximate straight line when the number of noise points other than the original boundary point 16 increases.

For example, in the case of an image as shown in FIG. 8A, when a detection area 15 is set, a boundary point 16 as shown in FIG. 8B is detected. When the prior art is applied to FIG. 8B, the approximate straight line 17 as shown in FIG. 9 is obtained, and the present invention has a problem that the center position of the object cannot be detected. An object of the present invention is to provide a pattern recognition method capable of setting an approximate straight line indicating the contour of an object.

[0006]

In order to solve the above-mentioned problems, a pattern recognition method according to the present invention provides an image obtained by capturing an image of an object by an image pickup means, in an image obtained by imaging the object and a background within a predetermined area. Detecting all boundary points with the boundary as a boundary point, dividing the predetermined area into a plurality of parts in a certain direction, and approximating the boundary points by the least square method within one of the plurality of divided areas. A boundary point separated from the approximate line by a predetermined reference distance or more is excluded as a noise point, and an approximate line is obtained again at the remaining boundary points. The variation of the boundary points present in the as the evaluation value in the area, the evaluation value is detected straight line of the area having the evaluation value closest to the predetermined reference value,
The approximate line of the region obtained above is defined as a straight line in a predetermined region, and a boundary point whose distance from the straight line to the boundary point is larger than a predetermined reference distance is deleted. In this configuration, an approximate straight line is obtained from the points by the least square method.

[0007]

According to the above arrangement, since the detection area is divided into a plurality of areas, the approximate straight lines of some of the divided areas indicate the contour positions of the object. Therefore, an approximate straight line indicating the contour of the target object can be set by detecting the above-mentioned area.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A pattern recognition method according to one embodiment of the present invention will be described below with reference to FIGS.

By the same method as the above-mentioned conventional one, FIG.
As shown in (a), the rough center of the object is obtained, and the detection area 15 is set from the rough center. In this detection area 15, an approximate straight line indicating the contour of the object is set according to the flowchart shown in FIG.

First, boundary point detection 1 in the flowchart
Then, scanning is performed from the leading end of the lead 14 in the detection area 15 to detect the boundary point 16. Next, in the region division 2 of the flowchart, as shown in FIG. 2B, the detection region 15 is divided into three equal parts in a certain direction, and each region is divided into regions,
Region, region.

In the linear approximation 3 in the divided area of the flowchart, first, the minimum 2
When the linear approximation by the multiplication method is performed, the result is as shown in FIG. Excluding the boundary points 16 whose distance between the approximate straight line and each boundary point 16 is equal to or greater than a predetermined reference distance and performing linear approximation on the boundary points 16 within the reference distance, FIG.
Can be detected. By performing this approximation line extraction for regions as shown in FIG. 3B, the approximation lines of the three regions can be detected as shown in FIG. 3C.

Next, in the evaluation value calculation 4 of the flowchart, the approximate line of the region, the region and the boundary point 16 of the region are determined.
Is obtained as an evaluation value. That is, the scatter of the boundary points 16 from one straight line is determined. Then, the approximate line of the region, the region and the boundary point of the region
An evaluation value of 16 and an evaluation value of the approximate straight line of the area and the boundary point 16 of the area and the area are obtained.

Then, in the optimum approximate straight line detection 5 of the flowchart, an approximate straight line having the best evaluation value, that is, the variation of the boundary point 16 from one straight line is minimized. Further, in noise point removal 6 in the flowchart, the detection area 15
In the above, the boundary points 16 in which the distance between the approximate straight line and the boundary points 16 is equal to or longer than a predetermined reference distance are excluded. Then, a straight line approximation by the least squares method is performed using the remaining boundary points 16 to detect an approximate straight line 17 as shown in FIG.

For example, in the case of the prior art, the contour of the object could not be detected from the image of FIG. 8A, but by employing the pattern recognition method of the present invention, FIG. As shown in (a) of FIG. 10 and (b) of FIG. 10, an approximate straight line connecting only the boundary points 16 indicating the outline of the object can be obtained.

[0015]

As described above, according to the pattern recognition method of the present invention, by dividing the detection area into a plurality of areas, the approximate straight lines of some of the divided areas are located at the contour positions of the object. Thus, an approximate straight line indicating the outline of the target object can be set by detecting the above-described area.

[Brief description of the drawings]

FIG. 1 is a flowchart schematically showing a pattern recognition method according to an embodiment of the present invention.

FIGS. 2 (a), (b), and (c) are diagrams each showing a part of a procedure of approximate line detection in the embodiment.

FIGS. 3A, 3B, and 3C are diagrams each showing a part of the procedure of approximate straight line detection in the embodiment.

FIG. 4 is a diagram showing a part of a procedure of approximate line detection in the embodiment.

FIGS. 5A and 5B are diagrams each showing a part of a procedure in a conventional pattern recognition method.

FIGS. 6 (a), (b), and (c) are diagrams each showing a part of a procedure in a conventional pattern recognition method.

FIG. 7 is a diagram showing a part of a procedure in a conventional pattern recognition method.

FIGS. 8A and 8B are diagrams showing a part of a procedure according to a conventional pattern recognition method.

FIG. 9 is a diagram showing an approximate straight line extraction result by a conventional pattern recognition method.

FIG. 10 is a diagram showing an approximate straight line extraction result according to an embodiment of the present invention.

[Explanation of symbols]

 11 Image 12 Coarse center search area 14 QFP lead 15 Detection area 16 Boundary point 17 Approximate line

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-61-29979 (JP, A) JP-A-2-10476 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G06T 5/00 G06T 7/00

Claims (1)

(57) [Claims] In an image obtained by imaging an object by an imaging means, all boundary points are detected using a boundary between the object and the background as a boundary point in a predetermined area, and the predetermined point is detected. The area is divided into a plurality of areas in a certain direction, and within one of the divided areas, an approximate straight line approximating the boundary point is obtained by the least square method, and the approximate straight line is separated from the approximate straight line by a predetermined reference distance or more. Exclude the boundary points as noise points, obtain an approximate straight line again at the remaining boundary points, and from the approximate straight line in the area, the variation of the boundary points existing in other areas as the evaluation value in the area, The evaluation value is detected as a straight line in an area having an evaluation value closest to a predetermined reference value, and an approximate straight line of the area obtained above is set as a straight line in a predetermined area. A pattern recognition method, wherein a boundary point whose distance to a boundary point is larger than a predetermined reference distance is deleted, and an approximate straight line is obtained from the remaining boundary points by a least square method.