JP2897291B2 - Image pattern recognition method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image pattern recognition method applicable to position detection of a work object in an industrial automatic machine.

2. Description of the Related Art In recent years, industrial automatic machines have been required to have a function of measuring and correcting the position of a work object in real time in order to achieve more precise work. Imaging of the work target from an imaging device such as
There is a tendency to use an image pattern recognition technique of extracting an image pattern of a recognition target set in advance on a work target from an obtained image and measuring a displacement of the work target based on the position information.

Conventionally, as this type of image pattern recognition technology, a method using a binary image as an input image and a method using a grayscale image have been generally proposed.

In the method using a binary image, recognition symmetry is generally recognized by the processing procedure shown in FIG. An image input step 1 is a step of inputting an image of an appropriate area including a recognition target from an imaging device. Reference numeral 2 denotes an input image binarization step in which the luminance of each pixel of the input image is converted into high luminance and low luminance based on a specified threshold value to obtain a binary image. Reference numeral 3 denotes an image pattern extracting step in which a boundary between a high-luminance pixel group and a low-luminance pixel group, that is, a pattern contour is searched from the binary image obtained in the input image binary process to extract an image pattern. 4 is a pattern feature amount extraction step for calculating pattern feature amounts such as a pattern area and a second moment of area based on the contour of the pattern extracted in the image pattern extraction step, and 5 is a recognition pattern identification step. The correlation between the pattern contours and the pattern features obtained in the image pattern extraction step and the pattern feature quantity extraction step and the contour and the feature quantity of the recognition target given in advance is identified to identify an image pattern corresponding to the recognition target. It is a process. Reference numeral 6 denotes a recognition result output step in which the center position of the image pattern selected in the recognition pattern identification step is output as the actual measurement position of the recognition target.

In the method using the grayscale image, the recognition target is generally recognized by the processing procedure shown in FIG. 7 is an image input step. Reference numeral 8 denotes a filtering step. As shown in FIG. 8, a 2 × 3 pixel two-dimensional filtering operator 11 such as a Laplacian filter is applied to all the pixels 10 of the input image 9 to change only the luminance change points on the image. This is a step of detecting an edge and extracting an edge. Reference numeral 12 denotes a segmentation step in which a plurality of edges obtained in the filtering step are connected / line-divided using Haugh transform or the like, and then segmented to extract an outline of an image pattern.
Subsequent steps are a pattern feature amount extraction step 13, a recognition pattern identification step 14, and a recognition result output step 15, similarly to the method using the binary image.

Problems to be Solved by the Invention In the case of the above-mentioned conventional image pattern recognition technology, a method using a binary image can be expected to have high processing speed,
When the contrast between the image pattern of the recognition target appearing on the input image and the background image is low, it is difficult to set a threshold value for always separating the image pattern and the background image in the input image binarization process, and it corresponds to the recognition target. It is difficult to extract image patterns clearly. On the other hand, the method using the grayscale image can perform good image pattern recognition even when the contrast of the input image is low, but includes a filtering process for all the pixels of the input image as can be understood from the details of the processing procedure described above. Therefore, it is difficult to realize high-speed image pattern recognition.

SUMMARY OF THE INVENTION An object of the present invention is to realize an image pattern recognition method in which accurate recognition is performed even when the contrast of an input image is low and the processing is fast.

Means for Solving the Problems The present invention extracts luminance information on a specified scanning line on an input image, and performs image quality improvement processing such as noise removal and image enhancement on the extracted luminance information. The start and end points of the pattern are extracted from the change point of, and all combinations of the start point and the end point are generated as a candidate pattern group of an image pattern corresponding to a recognition target. Identify the image pattern corresponding to the recognition target based on the similarity between the position and the given width of the recognition target and the position, and measure the position of the recognition target by measuring the position of the image pattern. It is a feature.

Operation The present invention enables high-speed grayscale image processing by the above-described configuration, and realizes an accurate and high-speed image pattern recognition method even for an input image having low contrast between a pattern and a background image.

Embodiment Hereinafter, as an embodiment of the present invention, an example of recognizing a mark position drawn on a circuit board in order to correct a position of a circuit board which is a work object in an automatic circuit forming machine will be described with reference to FIGS. This will be described with reference to FIG. The purpose of the embodiment is to mark 2
In order to obtain a dimensional position, in the following description, only a part for obtaining a position in one axis direction will be described. Therefore, the two-dimensional position can be obtained by performing the following processing once for the X coordinate axis and the Y coordinate axis.

In FIG. 1, reference numeral 16 denotes an image input step, and as shown in FIG.
An image 20 of an appropriate area including the mark 19 to be recognized is input. Here, FIG. 2 shows the contents of the image 20, wherein the hatched portion in the figure represents a high luminance area of the input image, and 21 represents a noise pattern.

Reference numeral 22 denotes a luminance extraction step in which a scanning line 23 parallel to the coordinate axes is designated on the input image 20 obtained in the image input step 16, and the luminance 24 of each pixel on the scanning line 23 is extracted.
The number of designated scanning lines need not be one, but a plurality of parallel scanning lines may be designated, and the luminance 24 may be an average value of the luminance of the corresponding points on each scanning line.

Reference numeral 25 denotes an image quality improving step which first performs a smoothing process on the luminance information extracted in the luminance extracting step to remove a noise component. Next, a luminance histogram 26 is taken, a luminance band 27 to be emphasized is determined based on the distribution state, a luminance conversion table 28 is created, and luminance is emphasized. Generally, the luminance of an input image tends to be unevenly distributed between a low luminance portion representing a background image region and a high luminance portion representing a pattern region. The central part of the luminance distribution area is adopted.

29 is a pattern start point and end point extraction process,
Is scanned in one direction, and a change point from low luminance to high luminance is extracted as a pattern start point, and a change point from high luminance to low luminance is extracted as a pattern end point. FIG. 2b is an example of scanning the change of the luminance 24 from left to right, and A, C,
Four points E and G are extracted, and three points B, D and F are extracted as pattern end points.

Reference numeral 30 denotes a candidate pattern group generation step which generates all combinations of the pattern start point and the end point obtained in the pattern start point and end point extraction step 29, and converts these combinations into image pattern candidate patterns corresponding to the mark 19. This is the step of forming the group 31. In the case of FIG. 2, pattern a (AB), pattern b
(AD), pattern c (AF), pattern d (C-
Six candidate patterns of D), pattern e (CF), and pattern f (EF) are generated. Note that the pattern start point G has no corresponding pattern end point and is ignored.

Reference numeral 32 denotes a pattern feature amount calculation step in which a pattern width Di and a center position Pi are obtained for each candidate pattern obtained in the candidate pattern group generation step 30.

33 is a recognition pattern identification process, which is a pattern width Di of the candidate pattern, a center position Pi, and a width of the mark 19 to be recognized.
D, position P, and an evaluation formula G1 × | Di−D | + G2 × | Pi−P | using the evaluation weighting factors G1 and G2 (both are positive values), and the best candidate that minimizes the value of this evaluation formula One pattern is selected from the candidate pattern group 31. The above-mentioned evaluation formula has a characteristic that the calculated value becomes smaller as the pattern width Di becomes equal to the width D of the mark 19 and the pattern center position Pi becomes equal to the position P of the mark 19. Therefore, it can be determined that the best candidate pattern that minimizes the value of this evaluation expression is an image pattern corresponding to the mark 19 to be recognized. In the case of FIG. 4, the candidate pattern e is selected as the best candidate pattern.

34 is a recognition result output step in which the center position Pi of the best candidate pattern selected in the recognition pattern identification step 33 is output as the actually measured position of the mark 19 to be recognized. In FIG. 4, the center position Pe of the candidate pattern e is the mark 19
Is output as the position of.

According to the present invention, an efficient grayscale image processing method is obtained, and an accurate and high-speed image pattern recognition method is realized even for an input image having a low contrast between a recognition pattern = background image.

[Brief description of the drawings]

FIG. 1 is a flowchart of an embodiment of the present invention, FIG. 2 is an explanatory diagram of an image in an image pattern recognition method to which the present embodiment is applied, FIG. 3 is an explanatory diagram of a histogram in an image quality improvement step, FIG. 4 is an explanatory view of an image pattern candidate pattern, FIG. 5 is a perspective view of a television camera illustrating an image input process,
6 is a flowchart of a conventional image pattern recognition method using a binary image, FIG. 7 is a flowchart of a conventional image pattern recognition method using a gray image, and FIG. 8 is a conventional gray image pattern recognition method. FIG. 3 is an explanatory diagram of a two-dimensional filter used in FIG. 19 ... mark, 20 ... input image, 23 ... scanning straight line, 24 ...
... luminance change, 26 ... luminance histogram, 28 ... luminance conversion table, 31 ... candidate pattern.

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Masamichi Morimoto 1006 Kazuma Kadoma, Osaka Prefecture Matsushita Electric Industrial Co., Ltd. (56) References JP-A-58-56429 (JP, A) JP-A-58- 21107 (JP, A) JP-A-57-2539 (JP, A) JP-A-52-91331 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) G01B 11/00-11 / 30 102 G06T 7/00

Claims (1)

(57) [Claims] A first step of picking up an image of a recognition target by an image pickup means; a second step of specifying a scanning line on the image picked up in the first step and extracting luminance information of a pixel on the scanning line;
A third step of improving the image quality by removing noise and enhancing the image of the extracted luminance information, and all the images traversed by the scanning line specified in the second step in the previous step focusing on the change point of the luminance information after the image quality improvement A fourth step of extracting a start point and an end point of the pattern, a fifth step of generating all combinations of the extracted pattern start point and end point and setting them as a candidate pattern group for a recognition target, and a fifth step. A sixth step of obtaining the pattern width of each candidate pattern obtained, a seventh step of obtaining the center position of each candidate pattern obtained in the fifth step, and a candidate pattern obtained in the sixth step and the seventh step Eighth step of selecting a candidate pattern that most closely matches the width and center position of the recognition target given in advance based on the pattern width and the center position of the group, and recognizing the center position of the pattern candidate selected in the eighth step Image pattern recognition method which is characterized in that a ninth step of outputting a measured position.