JPH0581423A - Pattern recognizing method - Google Patents

Abstract

(57) [Abstract] [Purpose] It is an object of the present invention to provide a pattern recognition method for accurately determining the contour of an object image by linear approximation. [Structure] (a) The boundary point detection means finds all boundary points 10 of the image contour in the area 8, (b) the area dividing means divides the area 8 into small areas in a predetermined direction, and (c) the area The center of gravity 7 of the boundary point in the small area is obtained by the center of gravity calculation means, and (d) the polygonal line connecting the center of gravity of the adjacent small areas with a straight line is obtained by the noise point removal means, and the boundary point on the predetermined direction side (hatched portion) Is removed, and (e) an approximate straight line is obtained at the boundary points remaining by the approximate straight line detecting means by the method of least squares and is set as a straight line corresponding to the contour.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention uses an approximate straight line obtained from the boundary point between the object and the background in the image picked up by the image pickup means in order to detect the position and inclination of the surface mounting electronic component. The present invention relates to a pattern recognition method for detecting the contour of a pattern.

[0002]

2. Description of the Related Art In recent years, devices utilizing pattern recognition have been widely used, but their recognition accuracy is a problem.

The conventional pattern recognition method will be described below.
An example will be described in which the contour of the object (QFP) is detected by linear approximation.

FIG. 3 is a pattern diagram showing the steps of a conventional pattern recognition method. In the image 11 shown in FIG. 3A, the coarse center detection area 12 is set in the horizontal direction and the vertical direction. Both ends of the horizontal coarse center detection area 12 are searched to extract both ends of the object in the horizontal direction to obtain the center in the horizontal direction. Further, the rough center 13 of the object can be detected by searching both ends of the vertical direction rough center detection area 12, extracting both ends of the object in the vertical direction, and obtaining the center in the vertical direction. From the obtained coarse center 13 and the size of the target object, as shown in FIG. 3B, the detection regions 8 including the leads 9 of the target object are set on four sides of the target object (QFP). Next, by detecting in the detection area 8 from the background toward the tip of the QFP lead 9, the boundary point 10 of the lead 9 can be extracted as shown in FIG. By applying the method of least squares on the boundary point 10 obtained above, FIG.
A straight line can be obtained as shown in (d). When the distance from the obtained straight line to each boundary point is more than a predetermined reference distance, the boundary point is regarded as a noise point and removed, and the least squares method is applied to the boundary points within the reference distance. By doing so, it is possible to detect the approximate straight line 14 as shown in FIG. By performing the above processing on the four sides of the object (QFP), the approximate straight lines 14 on the four sides can be detected as shown in (f) of FIG. The center position of the object can be detected by obtaining the intersection of adjacent straight lines of the four approximate straight lines 14 and obtaining the center of gravity of the intersection.

[0005]

In such a conventional pattern recognition method, it becomes difficult to detect an accurate approximate straight line when there are many noise points that are not the original boundary points. For example, in the case of an image as shown in FIG. 4A, the conventional method produces an approximate straight line 14 as shown in FIG.
The straight line connecting the tip positions of the QPF leads 9 is not detected.

An object of the present invention is to solve the above problems, and an object thereof is to provide a pattern recognition method capable of accurately detecting a contour even in a shape having many noise points.

[0007]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention relates to an image obtained by picking up an image of an object by an image pick-up means, in the area determined in advance by a boundary point detection means. Find all boundary points with the background,
The area dividing means divides the area into a plurality of small areas in a predetermined direction, the in-area center of gravity calculating means obtains the center of gravity of the boundary point in each of the small areas, and the noise point removing means calculates the adjacent small areas. Boundary points in a predetermined direction are removed from the polygonal line connecting the centers of gravity with straight lines as noise points, and an approximate straight line is obtained by the least-squares method for all boundary points remaining by the approximate straight line detecting means, and a plurality of approximate straight lines are obtained. The pattern recognition method is such that the contour of the object is obtained from the approximate straight line.

[0008]

According to the present invention, in the above structure, a point greatly deviated from a desired position is removed as a noise point, and an approximate straight line is detected only by an effective point sequence.

[0009]

【Example】

(Embodiment 1) A pattern detecting method of the present invention will be described below with reference to the drawings. FIG. 1 is a flowchart showing a process of a pattern detection method according to an embodiment of the present invention, FIG.
FIG. 6 is a pattern diagram showing the operation of the pattern recognition method of the present invention.

First, as the processing before the start of step 1,
The coarse center of the object is obtained by the same method as the conventional method, the coarse detection area 8 is set from the coarse center as shown in FIG. 2A, and the approximate straight line showing the contour of the object is determined in the detection area 8. To do.

Next, in step 1, scanning is performed from the front end direction of the QFP lead 9 in the detection area 8 to determine the boundary point 1
Detect 0. Next, in step 2, the detection area 8
Is divided into four equal parts in a fixed direction as shown in FIG. Next, in step 3, as shown in FIG. 2C, the center of gravity 7 is calculated by calculating the center of gravity from the coordinates of the boundary points 10 in each of the divided areas. Next, in step 4, the centers of gravity of adjacent areas are connected as shown in FIG. Now, the required boundary point is the tip portion of the QFP lead 9. Therefore, the boundary points existing in the shaded area shown in FIG. 2D are deleted as noise points. Next, in step 5, the least squares method is applied to the remaining boundary points to detect an approximate straight line as shown in (e) of FIG.

As described above, according to the pattern recognition method of the present invention, an image obtained by picking up an image of an object by the image pickup means is used as a boundary point between the object and the background within a predetermined area. A first step of detecting a boundary point, a second step of dividing a predetermined area into a plurality of areas in a certain direction, and a third step of obtaining a center of gravity of boundary point coordinates in one area of the plurality of divided areas. A fourth step of removing a boundary point in a predetermined direction as a noise point from the straight line connecting the centers of gravity of the areas rubbed in close proximity, and a fifth step of obtaining an approximate straight line from the remaining boundary points by the least square method By adopting the pattern recognition method, it is possible to remove a point greatly deviated from the contour as a noise point and detect an accurate approximate straight line using only an accurate point sequence.

[0013]

As is apparent from the above embodiments, the present invention relates to an image obtained by picking up an image of an object by the image pickup means, and the background and the object within a predetermined area by the boundary point detection means. All the boundary points are detected, the area dividing means divides the area into a plurality of small areas in a certain direction, and the center-of-area centroid calculating means finds the center of gravity of each boundary area to eliminate noise points. Means removes, as noise points, boundary points existing in a predetermined direction from the polygonal line connecting the centers of gravity of the adjacent small areas to each other with a straight line, and all the boundary points remaining by the approximate straight line detecting means are set to a minimum of 2 By using a pattern recognition method in which an approximate straight line is obtained by a multiplicative method, and the contour of the object is obtained by a plurality of the approximate straight lines, points greatly deviated from the contour are removed as noise points. And, it is possible to detect an accurate approximate line using only exact sequence of points.

[Brief description of drawings]

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

FIG. 2 is a pattern diagram showing a procedure for detecting an approximate straight line by a pattern recognition method according to an embodiment of the present invention.

FIG. 3 is a pattern diagram showing a procedure of a conventional pattern recognition method.

FIG. 4 is a pattern diagram showing an extraction result of an approximate straight line by the method of the present invention and the conventional method.

[Explanation of symbols]

1 operation of detecting boundary points of contour by boundary point detecting means 2 operation of dividing area into small areas by area detecting means 3 operation of calculating center of gravity of boundary points in small area by area center of gravity calculating means 4 noise point removal Operation for removing noise points by means 5 Operation for detecting approximate straight line from remaining boundary points by approximate straight line detecting means

Front page continuation (72) Inventor Takashi Shimizu 1006 Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd.

Claims (1)

[Claims] 1. An image obtained by picking up an image of an object by an image pickup means, all boundary points between the object and the background are detected by a boundary point detection means within a predetermined area, and the area division means is used. The area is divided into a plurality of small areas in a certain direction, the center of gravity of the boundary points in each small area is obtained by the area center of gravity calculating means, and the center of gravity of each adjacent small area is made straight by the noise point removing means. Boundary points in a predetermined direction are removed as noise points from the polygonal line connected by, and an approximate straight line is obtained by the least squares method for all boundary points remaining by the approximate straight line detecting means,
A pattern recognition method in which the contour of the object is obtained from a plurality of the approximate straight lines.