JPH0498580A - Method and device for pattern recognition - Google Patents

Abstract

PURPOSE:To shorten processing time by performing the pattern matching of an image to be inspected obtained from an image pickup device with a reference image by pattern collation by using a template. CONSTITUTION:The image to be inspected from the image pickup device 11 is pattern-collated with the reference image of a reference image memory circuit 16. A vertical image data extraction circuit 14 outputs image data to be inspected of seven rows opened uniformly at certain set intervals in the vertical direction of the image to be inspected simultaneously. A horizontal image data extraction circuit 15 connects output from the vertical image data extraction circuit 14, and outputs the point data of 49 binary images simultaneously. An arithmetic and logic circuit 17 performs the pattern comparison of reference image data with horizontal image data, and outputs the coincidence pulse of picture element frequency. A correlation value estimation circuit 18 performs the pattern collation of 49 data, and estimates and stores a correlation value at each coordinate position individually.

Description

[Detailed Description of the Invention] This invention relates to a pattern matching method and device for automatically detecting and processing a binarized image to be inspected obtained from an imaging device in a non-contact manner. The present invention relates to a pattern recognition device using reference image pattern matching, which performs pattern matching of an image to be inspected using a template in order to detect the position of an imaged target workpiece.

Template matching is a process of calculating the degree of approximation and searching for the most similar partial image by counting the matching of corresponding pixels between a partial image in the tray 1 inspected image and a preset standard pattern to be compared. It is also called pattern matching processing. FIG. 12 is a principle diagram for explaining the outline of this pattern matching process, and is a premise of the present invention. This will be explained below.

m rows x n columns (m5M, n
≦N) and the pixels at the corresponding positions of the reference image 7 for pattern matching of the same pixel size are compared and the number of matching pixels is counted. The number of comparisons between the partial image 6 and the reference image 7 at one coordinate is mXn, and the number of input data is 2X (mXn). As an example, the size of the image to be inspected 5 is M = 320 pixels, N = 240 pixels, and the size of the reference image 7 is
When the size of is m = n = 64 pixels, the number of input data is 8192, and when m = n = 128 pixels, the number of input data is 32854, and this amount of processing must be calculated as similarity at one coordinate. . Image to be inspected 5
There are CM-m+1)X (N-n+1) partial images 6 in the image, and mXn comparisons must be made for each of them, so the total metric is Xn (M-m+1)X
(N-n+1) comparison operations are required. That is,
If the size of the image to be inspected 5 is M = 320 pixels, N = 240 pixels, and the size of the reference image 7 is m = n = 64 pixels, a huge amount of calculation requires 186,322,944 comparison operations. becomes.

For example, Japanese Patent Publication No. 60-17152 and Japanese Unexamined Patent Application Publication No. 60-171 disclose a pattern matching process that automatically detects the pattern of a two-dimensional image to be inspected at high speed in a non-contact manner.
Japanese Patent No. 2-210596 describes an apparatus that performs effective pattern matching processing. Special Public Service 1980-1715
The pattern matching method according to Publication No. 2 is generally called a composite partial pattern matching method, and its principle is that two-dimensional patterns of a plurality of characteristic specific parts of the image to be inspected are registered as reference images 7 respectively. Then, partial images 6 are sequentially cut out from the two-dimensional pattern of the photographed image to be inspected 5, and the cut out partial images 6 and the reference image 7 are compared to detect the most matching coordinate position. This is done by setting the size of the partial image 6 to 12×12 pixels. Due to the characteristics of the pattern matching method, in order to perform real-time recognition, the calculation process for the similarity at one coordinate must be completed within one pixel clock period, so it must be performed for 12×12=144 pixels at the same time. Therefore 12X12=14
The reference image data of 4 pixels must have a configuration that can be output simultaneously, and the image data to be inspected must also have a configuration that can similarly output 144 pixels simultaneously. In addition, in order to complete the processing within one image scanning period, the comparison circuit that compares the similarity must be completed within one pixel clock period, so the calculation process for the similarity at one coordinate must be completed within one pixel clock period.
2=144 exclusive OR circuits are required. Since these input and output data are based on similarity at one coordinate, in order to process them simultaneously, 288
Input wiring and 144 output wiring are required.

Also) References Coarse template
e matching JrEEE Trans,,S
MC-7104-107 (1977) and rSequ
ential hierarchy old at 5scene m
Atching J IEEE Trans, ,C-2
According to page 7, 4, 359-3GG (19VS), pattern matching is performed on the image with the highest image resolution (original image), and images with coarse resolution are prepared in advance and the image resolution is gradually increased. A hierarchical pattern matching method is known in which the detection efficiency is increased by Conventionally, in the hierarchical pattern matching method or coarse detection method, which is a recognition method that performs pattern matching on coarse image resolution and pattern matching on images with high image resolution, one pixel at coarse resolution is the luminance data of multiple pixels. It describes grayscale image processing in which pixel data is processed as coarse pixels based on the sum of the values.The detection processing using this hierarchical pattern matching method is performed by software, and it is necessary to perform real-time position detection. has not arrived yet.

As another method using a hierarchical pattern matching method, JP-A-62-55VS1 discloses that when pattern matching is performed at coarse pixel intervals, the image resolution is kept as is, but the entire area of the reference pattern for matching is By performing pattern matching only on partial areas,
Efforts are being made to speed up processing. In any of the above methods using the hierarchical pattern matching method, the coarse pixels are not the reference image data itself registered in the original image;
By compressing or limiting the original image, the amount of reference image data is reduced during comparison processing, thereby speeding up the processing.

A conventional device for pattern matching using a reference pattern of arbitrary shape and size is disclosed in Japanese Patent Application Laid-open No. 62
There is a method according to No.-21059E3. According to this method, a reference pattern (8×12) is placed at an arbitrary position to form a reference pattern of n×m size with an arbitrary size and an arbitrary shape. Pattern matching is executed for each reference pattern in n×m, and the result is shifted to the target pixel coordinates of the n×m reference pattern.

The purpose is achieved by adding the grayscale images obtained in this way after each shift.

However, since the size of the reference image 7 is small, the composite partial pattern matching method that uses a large number of small partial patterns registers three or four partial patterns for one inspection image 5 range. As a result, there were the following problems.

■) In pattern registration, since the size of the reference image 7 is small, it is necessary to register characteristic partial patterns in combination, and there is a problem in the ease of registration work by selecting the quality of the operator. there were.

2) In pattern matching based on parts, since the size of the reference image 7 is small, the recognition rate is low if only one point is recognized, so a partial pattern of three or four points is registered for one image range to be inspected. . For this reason, in addition to the original method of detecting the position of a pattern based only on the magnitude of the correlation value, it is also necessary to check the relative positional relationship (distance, angle) between each partial pattern, which increases processing time and increases speed. position cannot be detected.

3) Image 5 to be inspected because the size of the partial pattern is small
If there are multiple identical patterns larger than the partial pattern size in the pattern detection range, it is impossible to detect the pattern position. For example, if three identical patterns exist in the pattern detection range, the position of the most matching partial pattern is registered as a partial pattern, and by repeating pattern matching, one of the three coordinate positions becomes the candidate point. I can't decide.

4) If the size of the partial pattern is increased, the above problem can be solved, but the original problem of pattern matching occurs, which is that the hardware scale increases significantly.

5) Since the size of the partial pattern is approximately 12×12 pixels, it is easily affected by noise, and the position detection accuracy is correspondingly poor.

Since noise is inherently random, the probability of occurrence of noise per pixel is generally the same regardless of whether the size of the partial pattern is small or large. From this point of view, it is not directly related to the size of the partial pattern, but the actual gray level of the target work pattern is not always a pattern with a clear contrast that makes it easy to perform binarization. The noise referred to in this case is that it is difficult for video signals that cross the threshold level to occur in areas that are far from the binarization threshold level, so it is difficult for noise to appear in the area, but there is a video signal that is almost the same as the binarization threshold level. Noise is likely to occur on a binarized image in a pattern portion that has a relatively large number of levels compared to other locations. Therefore, on the target workpiece pattern, there are parts with clear contrast and parts with less contrast, and there are pattern parts with clear contrast that are less prone to noise when binarized, and patterns that are more distinctive than pattern shapes. Therefore, if a pattern part with unclear contrast but more characteristic than the pattern shape is registered as a partial pattern, it will be more susceptible to noise, and the position detection will be affected accordingly. Accuracy deteriorates. This effect can be reduced by increasing the number of partial patterns, but this increases processing time accordingly.

JP-A-62-55 which is a hierarchical pattern matching method
During rough detection, VS1 thins out the reference image 7 to create a coarse reference pattern using a limited part of the original image and performs pattern matching at a certain coarse pixel interval. will be affected. In addition, when registering the reference image 7, there was a problem in the ease of registration work by selecting the quality of the operator.

In the method of JP-A No. 62-210596, the size of the reference image 7 is basically a small area of 8 x 12 pixels, and it is necessary to obtain pattern matching of large size. is possible, but in order to do so, it is necessary to perform zero processing using a large number of partial 7-inch turns, which increases processing time, and the number of partial patterns to be registered changes depending on the pattern to be detected. The recognition time differs depending on the target work, making time management of the equipment equipped with the recognition device complicated. Furthermore, registering such a large number of partial patterns requires complicated operations and judgments by the operator, and the registration work is not easy.

Therefore, the present invention has been proposed in view of the above-mentioned drawbacks, and it is an object of the present invention to provide a new and improved pattern recognition device.

. - What technical means were adopted to solve the above problems will be described below.

In a pattern recognition method that automatically extracts a specific pattern from a two-dimensional image, the two-dimensional video signal from the imaging means is digitized, an arbitrary size of the characteristic pattern of the image to be inspected is cut out, and it is pre-binarized as a reference image. A reference image storage means capable of storing an image as it is as an original image and outputting image data stored at an arbitrary two-dimensional coordinate position; and a reference image storage means capable of outputting image data stored at an arbitrary two-dimensional coordinate position; Individual inspection of VS rows in which (VS-1) horizontal scanning line holding means capable of holding the number of scanning lines are evenly spaced at k (1≦k≦dy) horizontal scanning line intervals (=pixel intervals) vertical image data extraction means for simultaneously outputting image data to be inspected corresponding to coordinate positions; By connecting (HS-1) pixel shift means in series that can hold pixels up to the number of pixels set by the microcomputer, one horizontal direction of image data to be inspected or one set pixel at the same time (1≦ j≦dx) evenly spaced HS
The above-mentioned (HS-1) is inputted with the image data to be inspected simultaneously outputted from the VS horizontal scanning lines from the vertical extraction means.
Horizontal image data extraction means that is configured to simultaneously output the image data to be inspected corresponding to the individual inspection coordinate positions of the total VS x HS point of the two-dimensional image data to be inspected by providing pixel shifting means in series at the VS stage. and, comparing the patterns of the reference image data from the reference image storage means and the image data to be inspected corresponding to each inspection coordinate position from the horizontal image data extraction means at VS×HS points at the same time, and When VS x HSS x HS congruent, the pattern matching output corresponding to the inspection coordinate position on each side of the point is -
a logic operation means that simultaneously serially outputs data from each portion corresponding to each inspection coordinate position as a data, and simultaneously counts each coincident data output corresponding to each inspection coordinate position from the logic operation means. The first rough detection is performed using a horizontal image. The data extraction means and vertical image data extraction means are set so that HS×VS pieces of image data to be inspected are simultaneously output at the pixel interval of the first pattern matching, and the image data to be inspected is set at the center position of the detection range of the image to be inspected. Among the correlation data of the HS x VS candidate points, the first rough detection was performed by arranging the centers of the inspection positions in H5 rows x VS columns to correspond, and the first pattern matching was performed. Search for the part that holds the maximum correlation value, and if the value is larger than a predetermined rough detection threshold value, use the pattern matching coordinate position corresponding to the part that holds the maximum correlation value,
The center coordinate position of the next pattern matching is determined, and the first
The coordinate position of the highest correlation value obtained in the second rough detection is arranged so that the center of the inspection position in the HS row x VS column corresponds to the coordinate position of the highest correlation value obtained in the second rough detection. Pixel spacing - second pixel spacing
)/2}≦0, and the second pixel interval that satisfies the minimum integer value is arranged so that the center of the HS row × VS column corresponds, and pattern matching is performed to obtain the second HS × VS number. The coordinate position of the highest correlation value of the correlation data of the candidate point is determined, and the center of the HS row x VS column is arranged around the coordinate position of the highest correlation value determined in the second rough detection, and the third Until the set pixel interval in the horizontal and vertical directions becomes 1 pixel interval, the pattern matching after the pattern matching in the horizontal direction is (pixel interval of the Pth pattern matching - pixel interval of the Pth + 1st pattern matching X (
The minimum H at the pixel interval of the P+1st pattern matching, which is an integer value.
The coordinate position of the Pth highest correlation value is placed in the center of 5 rows x VS columns, and the maximum value of the correlation data of the HS x VS candidate points of pattern matching performed at 1 pixel intervals is determined by a predetermined value. This is achieved by a pattern recognition method characterized in that if the value is larger than the detection threshold value, the coordinate position corresponding to the portion where the maximum correlation value is held is set as the detection position.

Another solution is to set the constant to dx=3 in a pattern recognition method that automatically extracts a specific pattern in a two-dimensional image.
2. cN'=32. HS=7. VS=7, and in the first rough detection, both the horizontal image data extraction means and the vertical image data extraction means are set so that the image data to be inspected is output at 8 pixel intervals, and the detection range of the image to be inspected is The center of the inspection position of 7 rows and 7 columns (4th row and 4th column)
Perform the first rough detection by arranging the
If the maximum correlation value among the correlation values of the coordinate positions of the first 49 candidate points is larger than a predetermined coarse detection threshold value, calculate the coordinate position corresponding to the part where the maximum correlation value is held. , a second coordinate position is calculated based on the coordinate position obtained in the first rough detection corresponding to the part where the maximum correlation value is held.
The second coarse detection is performed at 3 pixel intervals, and if the maximum correlation value among the correlation values of the coordinate positions of the second 49 candidate points is larger than a predetermined coarse detection threshold value, the maximum correlation value is determined. Find the coordinate position corresponding to the part where . If the maximum correlation value among the correlation values of the coordinate positions of the candidate points is larger than a predetermined detection value, the coordinate position corresponding to the part holding the maximum correlation value is set as the detection position. This is achieved by a pattern recognition method characterized by:

Still another solution is the solution described in item 1, in which d
x=32. dy=32. HS=7. VS=7, and the detection means is set so that 7×7=49 pieces of image data to be inspected are output simultaneously at 16 pixel intervals for the first pattern matching and at pixel intervals for the first pattern matching. Perform the first rough detection so that the center of the 7 rows x 7 columns of inspection positions (4th row, 4th column) corresponds to the center of the detection range, and if the correlation is higher than the coarse detection threshold value. If a pattern matching position that is a value cannot be obtained, pattern matching is performed at 16 pixel intervals the second time, but the matching position is determined by performing recognition at a position shifted by 8 pixels from the image data generation position of the reference image pattern. The second
The coarse detection threshold value for the first time is the same as the coarse detection threshold value for the first time, and if a candidate point that is larger than the coarse detection threshold value for the second time is found, the second time is set in advance by the microcomputer. Find the coordinate position of the highest correlation value of the correlation data of the 49 candidate points in the second round, and calculate the coordinate position of the highest correlation value found in the second round of rough detection. 3rd centering around the position
Correlation data of 49 candidate points of the third time, which is larger than the coarse detection threshold value for the third time set after the second rough detection by the microcomputer after the second rough detection is performed at 3-pixel intervals. Find the coordinate position of the highest correlation value of
The fourth and subsequent pattern matching is performed at 1 pixel intervals, centering on the highest value position coordinates obtained in the third rough detection,
If the maximum value of the correlation data of the 49 candidate points in the fourth round is larger than a predetermined detection threshold value -, the coordinate position corresponding to the part where the maximum correlation value is held is set as the detection position. This is achieved using a pattern recognition method.

In addition, another solution is the solution described in paragraph 1,
dx=32. dV=32. HS=7. VS=7, and the detection means is set so that 7×7=49 pieces of image data to be inspected are simultaneously output at a pixel interval of 32 pixels for the first pattern matching. The center of the inspection position of 7 rows x 7 columns (4 rows and 4
If a pattern matching position with a correlation value higher than the coarse detection threshold value could not be obtained in the second pattern matching, the second pattern matching position The second time is the same (first at 32 pixel intervals)
Pattern matching is performed using the second coarse detection threshold value, but the matching position is calculated by performing recognition at a position shifted by 16 pixels from the image data generation position of the reference image.
If a pattern matching position with a correlation value higher than the rough detection threshold value could not be obtained, the third time
Pattern matching is performed at 2-pixel intervals, but the matching position is calculated by performing recognition at a position where the image data generation position of the reference image is further shifted by 8 pixels from the second time.
The third rough detection threshold value is the same as the first rough detection threshold value, and the correlation value of the coordinate positions of the 49 candidate points in the third time, which is larger than the first coarse detection threshold value, is Find the coordinate position with the maximum correlation value from among them,
The fourth point is centered around the coordinate position obtained in the third rough detection.
The fourth rough detection is performed at 3-pixel intervals, and the coordinate position with the maximum correlation value is determined from among the correlation values of the coordinate positions of the 49 candidate points in the fourth time, and the coordinate position is determined in the fourth rough detection. The fifth detection is performed at 1 pixel intervals centering on the coordinate position
If the maximum correlation value among the correlation values of the coordinate positions of the 49 candidate points is larger than a predetermined detection threshold value, the coordinate position corresponding to the part holding the maximum correlation value is detected. This is achieved by a pattern recognition method characterized by position.

Still another solution is the solution described in item 1,
By interpolation processing using the correlation data of HS × VS number of cross-turn matching performed at pixel intervals and the inspection coordinate position, the matching coordinate position of 1 pixel or less in both the horizontal and vertical directions is calculated and used as the detection position. Using the characteristic pattern recognition method, we also use the correlation data of pattern matching performed at 1-pixel intervals to determine the pattern matching position in the horizontal direction from the correlation data of the left and right neighboring coordinates including the coordinate position with the highest correlation value. A matching coordinate position of 1 pixel or less in both the horizontal and vertical directions is calculated by spline interpolation using a total of 3 data, including the coordinate position with the highest correlation value in the vertical direction, and the correlation data of the upper and lower neighboring coordinates. and the detected position. <Respectively achieved by turn recognition methods.

The above-mentioned recognition method further determines a candidate point by simultaneously comparing the correlation value data of the correlation value accumulating means with a pattern detection threshold value set in advance by a microcomputer. In order to detect the N11 turn verification state at high speed, the data output of the correlation value accumulation means is compared with the held detection thread and Soshu value, and the magnitude is simultaneously determined between the coordinate position corresponding to each inspection coordinate position. Holding means corresponding to the individual inspection coordinate positions for comparing and determining and holding the results are provided for each of the VS x HS individual inspection coordinate positions, and the microcomputer has one horizontal scanning line HS force point. Vertical pattern matching V
By providing only S number of horizontal candidate point detection means, it is possible to grasp the outline of the state of each pattern matching position of each horizontal scanning line at high speed and easily specify the position corresponding to the calculation. In the first rough detection, the horizontal image data extraction means and the vertical image data extraction means are set so that HS×VS pieces of image data to be inspected are simultaneously output at the pixel interval of the first pattern matching. , the center of the inspection position of the HS row x VS column is arranged so that it corresponds to the center of the detection range of the image to be inspected, and the first rough detection is not performed), and the horizontal candidate point detection means detects the first After the first rough detection, the first
In the second coarse detection thread, it is checked whether there is a candidate point at a detection position exceeding the Sosch value, and if there is, the information of the horizontal candidate point detection means of the horizontal comparison line indicating the existence is read, and the information is read. From the means that holds the correlation value at the pattern matching coordinate position corresponding to z
This corresponds to the part that holds the maximum correlation value among the correlation count values of the plurality of candidate points subjected to a-turn matching. The N6 tanning matching coordinate position is determined as the center coordinate position for the next pattern matching, and the inspection position of the HS row x VS column is determined with the coordinate position of the highest correlation value obtained in the first rough detection as the center. The centers are arranged so that they correspond, and the second rough detection is performed in the horizontal direction by ε first pixel interval - second pixel interval X (HS-
1)/2)≦0, and the vertical direction is (first pixel interval - second pixel interval)
The pixel spacing X(VS-1)/2)≦0 is arranged so that the centers of H5 rows x VS columns correspond to each other at a second pixel spacing satisfying X(VS-1)/2)≦0. The coordinate position of the highest correlation value of the correlation data of the plurality of candidate points in the second time, which is larger than the rough detection value for the second time set after the first rough detection, is determined, and The horizontal direction is (
Pixel interval of Pth pattern matching - Pixel interval of P+1st pattern matching X (HS-1)/2)≦0,
The vertical direction is (pixel interval of Pth pattern matching - pth
Pixel interval of tenth pattern matching X(VS-1)/2
'}≦0 by arranging the center of the HS row x VS column at the pixel interval of the P+1st pattern matching that satisfies A pattern recognition method characterized in that the coordinate position having the maximum value of the correlation data of the single or plural candidate points of the P-thousandth time, which is larger than the detection threshold value for the P+1th time performed, is set as the detection position. This is achieved by

Yet another second solution is a pattern recognition method that automatically extracts a specific pattern from a two-dimensional image, in which the two-dimensional video signal from the imaging means is digitized and any characteristic pattern of the image to be inspected is extracted. A reference image pattern storage means is capable of outputting image data whose size is cut out and stored in advance as a reference image and stored at an arbitrary two-dimensional coordinate position, and an arbitrary horizontal scanning dy line of the two-dimensional image data to be inspected is set by a microcomputer. (VS-1) horizontal scanning line holding means capable of holding (1≦k≦
dy) vertical image data extraction means configured to simultaneously output image data to be inspected corresponding to individual inspection coordinate positions of VS rows evenly spaced at horizontal scanning line intervals (k pixel intervals); By serially connecting HS pixel shift means capable of holding VS simultaneous outputs of image data from the data extraction means in horizontal arbitrary pixels dx for the number of pixels set by the microcomputer, The image data to be inspected in the horizontal direction is located at the selected pixels (1≦j
≦dx) Input image data to be inspected simultaneously output from VS horizontal scanning lines from the vertical extraction means, which are output simultaneously from HS locations evenly spaced at 5 pixel intervals. By providing the above-mentioned HS number of serial pixel shift means at the VS stage, the total VS of the two-dimensional image data to be inspected x the image data to be inspected corresponding to each inspection coordinate position of the HS point can be simultaneously output. an image data extraction means, and a VS×HS exclusive OR for simultaneously pattern-comparing the image data from the reference image storage means and the image data to be inspected corresponding to each inspection coordinate position from the horizontal image data extraction means. logical operation means for outputting pattern matching outputs corresponding to individual inspection coordinate positions consisting of VS x HS simultaneously as congruence pulses from an exclusive OR means corresponding to each inspection coordinate position; By individually and simultaneously counting each matched pulse output corresponding to each inspection coordinate position from the means, VS × HS pattern matching is completed in one image scan, and the portion corresponding to each inspection coordinate position is calculated independently. The pattern matching state at the candidate point is detected at high speed by simultaneously comparing and determining the magnitude of the held correlation value accumulating means and the correlation value data of the correlation value accumulating means with a pattern detection threshold value set in advance by a microcomputer. The data output of the correlation value accumulating means is compared with the held detection threshold value, and the magnitude is determined simultaneously between the coordinate positions corresponding to the individual inspection coordinate positions, and the results are held. VS the holding means corresponding to the inspection coordinate position of
xHS number of inspection coordinate positions are provided, and the microcomputer collects the comparison results of 1 vertical scanning line VS locations, and each pattern of 1 vertical scanning line is read by the microcomputer once. HS that performs pattern matching in the horizontal direction using means that maintains the state so that the outline of the state of the matching position can be grasped at high speed
It consists of a vertical candidate point detection means that can grasp the outline of the state of each vertical scanning line at high speed and easily specify the position corresponding to the calculation by providing only a book. Both the image data extraction means and the vertical image data extraction means are set so that the image data to be inspected is simultaneously output at the first pixel interval, and the center of the inspection position of row HS x column VS is set at the center of the detection range of the image to be inspected. After the first rough detection by the vertical candidate point detection means, the first rough detection threshold value set in advance by the microcomputer is set by the vertical candidate point detection means. Check whether there is a candidate point for the detection position that exceeds the
If it exists, the information of the vertical candidate point detection means of the vertical comparison line indicating the existence is read, and from the means holding the correlation value at the pattern matching coordinate position corresponding to the information, the first The pattern matching coordinate position corresponding to the part that holds the maximum correlation value among the correlation count values of the plurality of candidate points subjected to pattern matching is determined as the center coordinate position of the next multiple putter matching, and Centered on the coordinate position of the highest correlation value obtained by rough detection, the centers of the inspection positions in the HS row x VS column are arranged so that they correspond,
The second coarse detection in the horizontal direction is (first pixel interval - second pixel interval)
pixel spacing × (HS-1)/2)≦0, and in the vertical direction (
First pixel spacing - Second pixel spacing x (VS-1)/2
}≦0 by arranging the centers of the HS row x VS column at a second pixel interval that satisfies The coordinate position of the highest correlation value of the correlation data of the plurality of candidate points in the second time, which is larger than the rough detection value for the second time, is determined, and the coordinate position of the highest value found in the second rough detection is determined. The center of the HS row x VS column is placed as the center, and the horizontal direction is (pixel interval of the second pattern matching - Pth + first pattern matching) until the third and subsequent pattern matching is performed at 1 pixel intervals. Pixel interval x (HS
-1)/2}≦0, and in the vertical direction, (pixel spacing of the second pattern matching - pixel spacing of the P+1st pattern matching X (VS-1)/2)≦0. The P+1 detection threshold value is obtained by arranging the center of the HS row x VS column at a pixel interval for pattern matching, and performing pattern matching by the microcomputer from the vertical candidate point detection means at 1 pixel intervals. This is achieved by a pattern recognition method characterized in that the coordinate position having the maximum value of the correlation data of the P-11th single or plural candidate points, which has become larger, is set as the detection position.

Yet another solution is a pattern recognition method that automatically extracts a specific pattern from a two-dimensional image, by digitizing the two-dimensional video signal from the imaging means and cutting out a characteristic pattern of any size from the image to be inspected. A reference image storage means capable of outputting image data stored in advance as a reference image and stored at an arbitrary two-dimensional coordinate position; Vertical image data extraction means for simultaneously outputting image data to be inspected corresponding to individual inspection coordinate positions in the VS row; By connecting HS pixel shift means in series that can hold up to any dX pixel in the direction by the number of pixels set by the microcomputer, the image data to be inspected in one horizontal direction can be simultaneously transferred to one set pixel (1 ≦J≦5d Horizontal image data extracting means which is configured to simultaneously output the image data to be inspected corresponding to the individual inspection coordinate positions of the total VSXHS point of the two-dimensional image data to be inspected by providing the HS number of serial pixel shift means in the VS stage. and VS×HS exclusive ORs for simultaneously pattern comparing the image data from the reference image pattern storage means and the image pattern data to be inspected corresponding to each inspection coordinate position from the horizontal image data extraction means. a logic operation means for outputting a pattern matching output corresponding to each inspection coordinate position as a VS x HSS x HS congruence rule from an exclusive OR means corresponding to each inspection coordinate position, and from the logic operation means By individually and simultaneously counting each match pulse output corresponding to each inspection coordinate position, VS×HS pattern matching can be performed in one
Correlation value accumulating means that independently calculates and holds correlation values for parts corresponding to individual inspection coordinate positions completed by image scanning; and Correlation value data of the correlation value accumulating means is transferred to a pattern detection threshold set in advance by a microcomputer. The data output of the correlation value accumulating means is compared with the held detection threshold value so that the pattern matching state at the candidate point can be detected at high speed by comparing and determining the magnitude with the value at the same time. The holding means corresponding to the individual inspection coordinate positions for holding the results by simultaneously comparing and determining the size with the coordinate positions corresponding to the VS x HS individual inspection coordinate positions are provided. The microcomputer collects the comparison results of HS locations on one horizontal scanning line and holds the status so that the status of each pattern matching position on one horizontal scanning line can be quickly grasped by reading the microcomputer once. Horizontal candidate point detection means that is capable of quickly grasping the outline of the state of each pattern matching position of each horizontal scanning line and easily specifying the position corresponding to the calculation by providing only VS units for vertical pattern matching. In order to be able to quickly grasp the outline of the state of each pattern matching position of the VS horizontal correlation lines of the horizontal candidate point detection means, the horizontal candidate point detection means can be grasped by one data input by a microcomputer. VS of point detection means
If there is even one state held larger than the rough detection threshold value in each state holding means corresponding to each inspection coordinate position of the book, the output of each state holding means corresponding to each inspection coordinate position of the book is It consists of a candidate point detection means that notifies the microcomputer that there is a horizontal correlation line that exceeds the coarse detection threshold value by holding the state of the line, and the first rough detection is performed by the horizontal image data extraction means and the vertical The image data extraction means is set so that HS x VS pieces of image data to be inspected are simultaneously output at the pixel interval of each first pattern matching, and HS rows x VS columns are placed at the center position of the detection range of the image to be inspected. After the first rough detection by the candidate point detection means, the center of the inspection position of It is checked whether there is a candidate point at the detection position exceeding the detection threshold value, and if there is, the information of the horizontal candidate point detection means of the horizontal comparison line indicating the existence is read, and the pattern matching corresponding to the information is performed. From the means that holds the correlation value at the coordinate position, the pattern matching coordinate position corresponding to the part that holds the maximum correlation value among the correlation count values of the plurality of candidate points for which the first pattern matching was performed is used for the next pattern matching. is determined as the center coordinate position of pattern matching, and arranged so that the center of the inspection position of the HS row × VS column corresponds to the coordinate position of the highest correlation value obtained in the first rough detection, The second rough detection is performed in the horizontal direction (the first
pixel interval - second pixel interval x (H5-1)/2)≦
0, and the vertical direction is (first pixel interval - second pixel interval
(VS-1)/2)≦0 by arranging the centers of rows HS and columns VS at a second pixel interval that satisfies The coordinate position of the highest correlation value of the correlation data of the plurality of second candidate points that has become larger than the second rough detection threshold value set after the rough detection is determined, and in the second rough detection HS line × centered on the obtained highest value position coordinates
Until the center of the VS column is arranged and the third and subsequent pattern matching is performed at 1 pixel intervals, the horizontal direction is (pixel interval of Pth pattern matching - pixel interval of Pth+1st pattern matching x (HS ~1)/2}≦0, and the vertical direction is (first
The center part of the HS row x VS column was arranged at the pixel interval of the second pattern matching + the first pattern matching, and pattern matching was performed by the microcomputer from the horizontal candidate point detection means at one pixel interval. This is achieved by a pattern recognition method characterized in that the coordinate position having the maximum value of the correlation data of the single or plural candidate points of the P+1st time, which is larger than the detection threshold value for the first time, is set as the detection position.

Still another fourth solution is a pattern recognition method that automatically extracts a specific pattern from a two-dimensional image, in which the two-dimensional video signal from the imaging means is digitized, and any characteristic pattern of the image to be inspected is extracted. a reference image pattern storage means capable of outputting image data whose size is cut out and stored in advance as a reference image pattern and stored at an arbitrary two-dimensional coordinate position; - an arbitrary horizontal scanning line dy of the two-dimensional image data to be inspected;
(1≦k
≦dy) Vertical image data extraction means configured to simultaneously output image data to be inspected corresponding to individual inspection coordinate positions of VS rows evenly spaced at horizontal scanning line intervals (k pixel intervals); By serially connecting HS pixel shift means capable of holding VS simultaneous outputs of image data to be inspected from the image data extraction means up to arbitrary dX pixels in the horizontal direction by the number of pixels set by the microcomputer, 1 horizontal direction image data to be inspected or a certain set pixel at the same time (
1≦J≦dx) intervals (3 pixel intervals), which are output simultaneously from the HS force points that are evenly spaced at intervals of 3 pixels, are simultaneously output from the VS horizontal scanning lines from the vertical extraction means. By using the image data as input and providing the HS number of serial pixel shift means in the VS stage, the image data to be inspected corresponding to the individual inspection coordinate positions of the total VS x HS points of the two-dimensional image data to be inspected can be simultaneously output. horizontal image data extraction means configured to perform a VS×HS pattern comparison, which simultaneously performs a pattern comparison between image data from the reference image pattern storage means and image pattern data to be inspected corresponding to each inspection coordinate position from the horizontal image data extraction means. VS×HSS×H
a logic operation means which outputs an S congruent pulse from an exclusive OR means corresponding to each inspection coordinate position; and a logic operation means which outputs each matching pulse corresponding to each inspection coordinate position from the logic operation means individually. Correlation value accumulating means that completes VS×HS pattern matching in one image scan by simultaneously counting and independently calculating and retaining at a portion corresponding to each inspection coordinate position; and correlation value data of the correlation value accumulating means. The data output of the correlation value accumulating means and the held detection threshold value are used to detect the pattern matching state at a candidate point at high speed by simultaneously comparing and determining the magnitude with a pattern detection threshold value set in advance by a microcomputer. The value is compared with the coordinate position corresponding to each inspection coordinate position, and the holding means corresponding to each inspection coordinate position that holds the result is compared and determined at the same time with the coordinate position corresponding to each inspection coordinate position. The microcomputer is equipped with the comparison results of one vertical scanning line vs. several locations, and the microcomputer can quickly obtain an outline of the state of each pattern matching position of one vertical scanning line with one reading of the microcomputer. By providing a means for holding the state so that it can be grasped only in the HS book that performs pattern matching in the horizontal direction, it is possible to quickly grasp the outline of the state of each vertical scanning line and easily specify the position corresponding to the calculation. In order to be able to quickly grasp the outline of the state of the vertical candidate point detecting means and each pattern matching position of the HS vertical correlation lines of the vertical candidate point detecting means, one data input by a microcomputer is required. HS of the vertical candidate point detection means so that it can be grasped.
If the output of each state holding means corresponding to each inspection coordinate position of the book is held in a state larger than the previous 4' packing coarse detection threshold value, at least one state is maintained. The candidate point detection means holds the state of the vertical correlation line and notifies the microcomputer that there is the vertical correlation line exceeding the coarse detection threshold value, and the first rough detection is performed by the horizontal image data extraction means. - Set the vertical image data extraction means to simultaneously output HS x VS pieces of inspected image data at the pixel interval of each first pattern matching, and set the HS row x VS pieces of inspected image data at the center position of the detection range of the inspected image. A first rough detection is performed by arranging the centers of the inspection positions in the VS column to correspond to each other, and after the first rough detection by the vertical candidate point detection means, the first rough detection point is set in advance by the microcomputer. It is checked whether there is a candidate point at the detection position that exceeds the coarse detection threshold value for the second time, and if there is, the information of the vertical candidate point detection means of the vertical comparison line indicating the existence is read, and From the means that maintains the correlation value,
The pattern matching coordinate position corresponding to the part that holds the maximum correlation value among the correlation count values of the plurality of candidate points subjected to the first pattern matching is determined as the center coordinate position of the next pattern matching, and the The coordinate position of the highest correlation value obtained in the first rough detection is arranged so that the center of the inspection position in the HS row x VS column corresponds to the coordinate position of the highest correlation value obtained in the first rough detection, and the second rough detection is pixel spacing − second pixel spacing x(H
S-1)/2)≦0, and the vertical direction is (first pixel interval -
The second pixel interval X (VS-1)/2)≦0 is arranged so that the centers of rows HS and columns VS correspond to each other at second pixel intervals that satisfy the condition X (VS-1)/2)≦0. The second rough detection set by the computer after the first rough detection
Find the coordinate position of the highest correlation value of the correlation data of the plurality of candidate points in the second time, which is larger than the rough detection threshold value for the second time, and center around the coordinate of the highest value position found in the second rough detection. Until the center of the HS row x VS column is arranged and the 3rd and subsequent pattern matching is performed at 1 pixel intervals, the horizontal direction is (pixel interval of the Pth pattern matching - pattern matching of the Pth + Ith pattern matching) pixel spacing x (HS-1),,'&2
}≦0, and the vertical direction is (pixel interval of Pth pattern matching - pixel interval of p+th pattern matching x (V
S-1)/2}≦0 by arranging the center of the HS row x VS column at the pixel interval of the P+1st pattern matching satisfying The detection position is characterized in that the coordinate position having the maximum value of the correlation data of the P+1 single or multiple candidate points, which is larger than the P+1 detection threshold value performed at 1 pixel intervals, is set as the detection position. This is achieved through pattern recognition methods.

Furthermore, in the pattern recognition method having the first solution described above, in the first rough detection, both the horizontal image data extraction means and the vertical image data extraction means output the image data to be inspected at the first pixel interval. The coarse detection threshold value for the first detection is set in advance so that the center of the inspection position in the HS row x VS column corresponds to the center position of the detection range of the image to be inspected. If a candidate point is found, the horizontal candidate point detecting means detects the first single or plural pattern matching candidate points whose value is larger than the first coarse detection threshold set in advance by the microcomputer. The center of gravity position is determined from the XY coordinate position data indicating Yoinashi, and the second rough detection is performed (first pixel interval -
A second pixel interval that satisfies the second pixel interval Find the centroid position of the second single or multiple candidate point position data that is larger than the rough detection threshold value, and calculate the centroid position of the third and subsequent candidate points around the centroid position obtained in the second coarse detection. Until pattern matching is performed at one pixel interval (pixel interval of Pth pattern matching - pixel interval of p+th pattern matching x (HS-1)/2}≦
The P+1th pattern matching is performed at the pixel interval of the P+1st pattern matching that satisfies 0, and is larger than the P+1st detection threshold value set after the second rough detection by the microcomputer from the candidate point detection means. This is achieved by a pattern recognition method characterized in that the maximum value of the position data of a single candidate point or a plurality of candidate points is determined as the detected position.

Further, in the pattern recognition method having the second solving means, in the first rough detection, both the horizontal image data extraction means and the vertical image data extraction means are set so that the image data to be inspected is output at the first pixel interval. Then, a first rough detection is performed by arranging the center of the inspection position of the HS row x VS column so that it corresponds to the center of the detection range of the image to be inspected, and the output from the vertical candidate point high-speed detection means is If a candidate point is found that is larger than the first rough detection threshold value preset by the microcomputer, the vertical candidate point detection means detects the first coarse detection threshold value preset by the microcomputer. The center of gravity position is determined from the XY coordinate position data indicating the presence or absence of a single or multiple pattern matching candidate point in the first round that is larger than the detection threshold value, and the position of the center of gravity determined in the first rough detection is used as the center. The second rough detection (first pixel interval - second pixel interval
(HS-1)/2) ≦ 0 at the second pixel interval satisfying The centroid position of the second single or plural candidate point position data that has become larger is determined, and the second
Until the third and subsequent pattern matching is performed at 1 pixel intervals around the center of gravity position obtained in the first rough detection (pixel interval of the Pth pattern matching - p+) is the pixel interval of the second pattern matching x P+ that satisfies (MS-1)/2)≦0
The P+1st detection threshold is set at the pixel interval of the first pattern matching, and is larger than the detection threshold value for the P101st detection set by the microcomputer from the candidate point detection means after the second rough detection. This is achieved by a pattern recognition method characterized in that the maximum value of the position data of a single or a plurality of candidate points is determined as the detection position.

Here, in the first solution, dx=32. dy=3
2. HS=7. VS=7, and in the first rough detection of the sono detection means, both the horizontal image data extraction means and the vertical image data extraction means are set so that the image data to be inspected is output at 8 pixel intervals, and the detection range of the image to be inspected is The center of the inspection position in 7 rows and 7 columns corresponds to the center of the first inspection position.
If a candidate point is found whose output from the horizontal candidate point high-speed detection means is larger than the first rough detection threshold value set in advance by the microcomputer after performing the first rough detection, the horizontal candidate point The center of gravity position is determined from the high-speed detection means based on the correlation value of the position data of the first single or multiple candidate points, which is larger than the coarse detection threshold value for the first time set in advance by the microcomputer. , a second rough detection is performed at three pixel intervals centering on the center of gravity position obtained in the first rough detection, and the candidate point high-speed detection means detects the first
The centroid position of the position data of the single or multiple candidate points that has become larger than the second rough detection threshold value set after the second rough detection is calculated, and A third round of detection is performed at one pixel intervals centering on the center of gravity position determined by The third one is bigger
This is achieved by a pattern recognition method characterized in that the maximum value of the position data of a single candidate point or a plurality of candidate points is determined as the detected position.

In addition, when there is no coordinate position that reaches a predetermined pattern matching correlation value within the image range to be inspected, this is achieved by a pattern recognition method characterized in that detection of the target position is disabled.

Yet another solution involves a binarization circuit that digitizes a two-dimensional video signal from an imaging device, and a feature that cuts out a characteristic pattern of an image to be inspected to an arbitrary size within the storage capacity of a memory and stores it in advance as a reference image. a reference image storage circuit capable of outputting image data stored at any two-dimensional coordinate position; and a reference image storage circuit capable of outputting image data stored at an arbitrary two-dimensional coordinate position; and a reference image storage circuit capable of outputting image data stored at an arbitrary two-dimensional coordinate position; Horizontal scanning line shift register (VS-1) that can hold a number arbitrarily set according to instructions from a microcomputer.
The horizontal scanning lines (1≦k≦dx) are set in the vertical direction of the binarized image by the (VS-1) horizontal scanning line shift registers and the binarization circuit.
a vertical image data extraction circuit configured to simultaneously output image data to be inspected corresponding to individual inspection coordinate positions of VS rows evenly spaced at intervals (=pixel intervals); and from the vertical image data extraction circuit. Variable-length pixel/shift register that can hold VS simultaneous output of binary image data to be inspected from 1 pixel to 6 pixels in the horizontal direction, arbitrarily set according to instructions from the microcomputer. By connecting H5 pieces in series, one horizontal direction binarized image data to be inspected is simultaneously output from the HS force points that are evenly spaced at a set 3-pixel interval (1≦j≦dy). By providing the VS stage of the HS series variable length pixel/shift register which inputs the binarized image data to be inspected which are simultaneously output from the VS horizontal scanning lines from the vertical extraction circuit. a horizontal image data extraction circuit configured to simultaneously output binarized inspected image data corresponding to individual inspection coordinate positions of the total VS×HS point of the binarized image; and a reference from the reference image storage circuit. VS×H for simultaneously comparing patterns of image data and image data to be inspected corresponding to individual inspection coordinate positions from the horizontal image data extraction circuit;
Consisting of S exclusive OR circuits, the pattern matching output corresponding to each inspection coordinate position is output as a VS x HSS x HS congruence signal from an exclusive OR section corresponding to each inspection coordinate position. VS
It is equipped with a correlation value integration circuit that completes S pattern matching in one image scanning period and independently calculates and stores the parts corresponding to each inspection coordinate position. Both image data extraction units are set so that the image data to be inspected is output simultaneously at the first pixel interval, so that the center of the inspection position in the HS row x VS column corresponds to the center position of the detection range of the image to be inspected. The coordinate position of the highest correlation value of the correlation data of the first VS×HS candidate points is determined by performing the first rough detection, and the highest correlation value determined by the first coarse detection is calculated. The horizontal direction of the second rough detection centered on the coordinate position is (1st pixel interval - 2nd pixel interval x (HS-1)/2
)≦0, and in the vertical direction, the second pixel spacing is the smallest integer value that satisfies (first pixel spacing - second pixel spacing X (VS-1)/2)≦0. Find the coordinate position of the highest correlation value of the correlation data of the 5 candidate points of the VSXH,
The third and subsequent pattern matching is performed centering on the position coordinate having the highest correlation value obtained in the second rough detection, until the set pixel interval in the horizontal and vertical directions becomes 1 pixel interval. Pixel interval of the 1st pattern matching - Pth + pixel interval of the 1st pattern matching x (HS - 1) / 2) ≦ 0,
The vertical direction is (pixel interval of Pth pattern matching - Pth
+ pixel interval of first pattern matching x (VS-1)/
2}≦0, and the maximum value of the correlation data of the HS×VS candidate points of the pattern matching performed at 1 pixel interval is performed at the pixel interval of the P+1st pattern matching that satisfies the minimum integer value. This is achieved by a pattern recognition apparatus characterized in that if the value is larger than a predetermined detection threshold value, the coordinate position corresponding to the portion where the maximum correlation value is held is set as the detection position.

On the other hand, a pattern recognition device that automatically extracts a specific pattern from a two-dimensional image includes a binarization circuit that digitizes the two-dimensional video signal from the imaging device, and a memory that stores the characteristic patterns of the image to be inspected. a reference image storage circuit capable of outputting image data cut out in an arbitrary size within a capacity range as a reference image and stored in an arbitrary two-dimensional coordinate position; and a binarized image to be inspected from the binarization circuit. It is composed of horizontal scanning line shift registers (VS-1) that can hold data from one horizontal scanning line to dx horizontal scanning lines arbitrarily set according to instructions from a microcomputer.
) horizontal scanning line shift registers and the binarization circuit, the binarized image is evenly spaced at horizontal scanning line (1≦k≦dx) intervals (=pixel interval) in the vertical direction of the binarized image. A vertical image data extraction circuit configured to simultaneously output image data to be inspected corresponding to each inspection coordinate position of an open VS row, and binarized image data to be inspected from the vertical image data extraction circuit. By serially connecting HS variable-length pixel/shift registers that can hold VS simultaneous outputs from 1 to 4 pixels in the horizontal direction for pixels arbitrarily set according to instructions from a microcomputer, one horizontal At the same time, the binarized inspected image data of the direction is set at a certain pixel (
The binary values simultaneously output from the VS horizontal scanning lines from the vertical extraction circuit are output simultaneously from the HS power points equally spaced at intervals of 1≦j≦dV (1 pixel interval). The HS is inputted with image data to be inspected.
A horizontal image in which binarized image data to be inspected corresponding to each inspection coordinate position of the total VS x HS point of the binarized image is output simultaneously by providing variable length pixel O shift registers arranged in series at the VS stage. a data extraction circuit, and VS×HS exclusive logics for simultaneously pattern-comparing the image data from the reference image pattern storage circuit and the image pattern data to be inspected corresponding to each inspection coordinate position from the horizontal image data extraction circuit. The pattern matching output corresponding to each inspection coordinate position consisting of a sum circuit is VS×HSS
×HS congruence pulse output from an exclusive OR section corresponding to each inspection coordinate position; A correlation value accumulating circuit that completes VS×HS pattern matching in one field by counting at the same time, and independently calculates and holds the portion corresponding to each inspection coordinate position; The output of the correlation value integration circuit corresponds to each inspection coordinate position so that the pattern matching state at the candidate point can be detected at high speed by simultaneously comparing and determining the magnitude with the pattern detection threshold value set in advance by the microcomputer. A comparator is installed to hold the detection threshold value, and the latch corresponding to each inspection coordinate position is compared and judged simultaneously with the latch corresponding to each inspection coordinate position, and the result is held. The microcomputer collects the comparison results of each horizontal scanning line HS, and one reading of the microcomputer provides an outline of the state of each pattern matching position of one horizontal scanning line. By providing a latch section for only the VS for vertical pattern matching, it is possible to quickly grasp the outline of the state of each pattern matching position in each horizontal scanning line and easily identify the position corresponding to the calculation. In order to be able to quickly grasp the outline of the state of the horizontal candidate point detection circuit and each pattern matching position of the VS horizontal scanning lines of the horizontal candidate point detection circuit, one-time data input by a microcomputer is required. The output of each latch section corresponding to each inspection coordinate position of the VS book of the horizontal candidate point detection circuit is threaded in each latch section corresponding to each inspection coordinate position so that it can be grasped.
The candidate point detection circuit is equipped with a candidate point detection circuit that holds the state of the horizontal scanning line if there is even one launch that is larger than the Sosch value, thereby notifying the microcomputer that there is a horizontal correlation line that exceeds the correlation threshold value. For the first rough detection, both the horizontal image data extraction section and the vertical image data extraction section are set so that the image data to be inspected is output at the same time at the first pixel interval, and the HS row The first rough detection is performed by arranging the centers of the inspection positions in the VS column to correspond, and the output from the candidate point detection circuit or the first rough detection threshold set in advance by the microcomputer is detected. If a candidate point larger than the value is found, the horizontal candidate point detection circuit calculates the correlation between the plurality of first candidate points that are larger than the first coarse detection threshold value set in advance by the microcomputer. The coordinate position of the highest correlation value of the data is determined, and the second rough detection is performed in the horizontal direction centering on the coordinate position of the highest correlation value determined in the first rough detection.
pixel spacing - second pixel spacing X (HS-1)/2)≦
0, and the vertical direction is (first pixel interval - second pixel interval ×
(VS-1)/2}≦0 is performed at the second pixel interval that satisfies Find the coordinate position of the highest correlation value of the correlation data of the plurality of candidate points that has become larger, and
The horizontal direction is (pixel interval of Pth pattern matching - pixel interval of Pth Pixel interval for pattern matching x (HS-1)/2}≦0
, the vertical direction is (pixel interval of Pth pattern matching -
Pixel interval X of P+1st pattern matching (VS-1
)/2) ≦ 0 is performed at the pixel interval of the P+1st pattern matching, and pattern matching is performed by the microcomputer from the horizontal candidate point detection circuit at 1 pixel intervals. This is achieved by a pattern recognition device characterized in that the coordinate position having the maximum value of the correlation data of the P+1 single or plural candidate points that is larger than the threshold value is set as the detection position.

According to the present invention, even when the image pattern shape in the image to be inspected is monotonous and the same pattern continues repeatedly, the two-dimensional image data to be inspected from the imaging device is output as continuous serial data. Since it is difficult to specify the position by position detection unless the reference image area of the two-dimensional image is made large, the correlation value at a certain coordinate position cannot be calculated unless a large number of pixels are processed simultaneously. However, if we focus on one coordinate position, the correlation value at that position is the serial data (= reference image data) that is compared with the reference serial data (= reference image data).
image data to be inspected) are simultaneously output within one scanning period,
By synchronizing the inspection image data at each corresponding position with the scanning period of the video signal and performing correlation calculation processing pixel by pixel, even two-dimensional area data can be processed as temporally serial data, and two-dimensional Data correlation calculation 1
This is intended to simplify the calculation of correlation values at each coordinate position, similar to processing using dimensional correlation calculations. In other words, instead of completing the correlation value at each coordinate position within one pixel clock period, a large number of parts are provided horizontally and vertically where the correlation values are calculated as serial data within one scanning period and the results are integrated as serial data. Correlation processing is performed at the coordinate positions of . Therefore, the size of the reference image does not affect the physical scale of the processing means, so by compressing or limiting the original image, the amount of reference image data can be reduced during comparison processing, eliminating the need for processing and reducing the size of the reference image. Since it is possible to increase the size of the pattern, even if it is difficult to register a simple and characteristic pattern shape, by processing the image data of the original image as it is, it is possible to detect the position of the pattern without having to accumulate know-how associated with registration.

Embodiments of the pattern recognition device according to the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a block circuit of a pattern recognition device. As shown in the figure, the pattern recognition device includes a binarization circuit 12 that digitizes a video signal from an imaging device 11, and a characteristic pattern in which a binarized image to be inspected from the binarization circuit 12 is stored in advance. A reference image storage circuit 16 for pattern matching with the reference image of , and can hold binarized image data to be inspected from 1st line to 16th line in the horizontal direction as arbitrarily set according to instructions from the microcomputer 13. It has a variable length horizontal scanning line shift register and is connected to a binarization circuit 12.
A vertical image data extraction circuit 14 outputs 7 lines of inspected image data evenly spaced at set intervals in the vertical direction of the digitized image at the same time, and a vertical image data extraction circuit 14 that outputs the digitized inspected image data horizontally. By connecting seven variable-length pixel/shift registers in series that can hold 1 to 166 pixels in the direction arbitrarily set according to instructions from the microcomputer 13, one horizontal binarized image can be created simultaneously. Binarized image data to be inspected in 7 lines output from the vertical image data extraction circuit 14 are connected so that they are output from 7 equally spaced locations at certain set intervals.
A horizontal image data extraction circuit 15 configured to simultaneously output data at points 7×7), and 49 horizontal image data extraction circuits for pattern comparing the image data from the reference image storage circuit 1B and the image data to be inspected from the horizontal image data extraction circuit 15. The pattern matching output of the exclusive OR means is ANDed with the clock pulse of the pixel frequency, and if the patterns match as a result of pattern matching, matching is detected by generating a pulse corresponding to each pixel. A logic operation circuit 17 that outputs matching pulses of the pixel frequency; and 49 pattern matching is performed by individually counting each matching pulse output from the logic operation circuit 17, and a correlation value at each coordinate position is calculated. and a correlation value integration circuit 18 that integrates and holds the values individually.

Next, the operation of each component circuit section will be explained in detail. In an actual design example, the field of view of the imaging device 11 is quantized into a grid of 320 pixels horizontally and 240 pixels vertically, and the size of the reference image can be any rectangular range within 128×128 pixels. The configuration of the reference image storage circuit 16 includes a total of four counters each in the horizontal and vertical directions for generating timings for starting and completing operations.
A timing signal for reading or writing a reference image is generated, and a counter that counts clocks corresponding to a pixel is operated during a period in which reading or writing of the reference image is enabled by the read timing signal, and image data in the storage circuit is input/output. It is something. This makes it possible to arbitrarily change the pattern size of the reference image by changing the set values for the counters that generate the timing signals for starting and completing the operation of the reference image pattern storage circuit 16, respectively.

In this example, the size of the reference image is 128×12
Although we used a rectangular range of 8 pixels or less, it is of course possible to perform pattern matching using a larger reference image as long as the storage capacity of the memory allows.
Can be designed arbitrarily.

Furthermore, as long as a memory with a large storage capacity is used, it is possible to change the size that can be registered per reference image without increasing the circuit scale, and even if a design using a large number of memories is performed, the circuit scale will be reduced. The increase will not be significant.

The memory of the reference image storage circuit 16 is 258kJl'bit.
When using a RAM in which each pixel corresponds to one address, the size of the rectangular range in which pattern matching is performed can be set in the horizontal and vertical directions of the pattern in units of one pixel. Furthermore, since the storage capacity per pattern is 16kbit, the pattern setting for the reference image is 128X12.
It is possible to use it not only as a pattern range of 3 pixels or less, but also as a storage capacity of 16 kbit per pattern. For example, 200x80 pixels or 100x1
A pattern range of 50 pixels or the like is possible as required within the range allowed by the standard image storage capacity of a rectangle whose length exceeds 128 pixels in either the horizontal or vertical direction.

If the memory of the reference image storage circuit 16 is configured with an X4bit or X8bit RAM that does not have one pixel and one address, for example, if it is a 32kX8bit RAM, it will be 8 pixels in parallel input/output, and the horizontal pattern setting will be 8. The unit is pixel. Either horizontal or vertical direction is 12
It is possible to set a rectangular reference image with a length exceeding 8 pixels if necessary, such as 256k x 1bit.
This is the same as when designing using a RAM that corresponds to one pixel address.

Further, by changing the set values for the counters that generate the timing signals for starting and completing the operation of the reference image storage circuit 16, it is possible to move the pattern generation of the reference image to an arbitrary coordinate position. This means that if the coordinate position where the pattern is stored is from the 100th pixel to the 200th pixel in the horizontal direction, the start counter is 100.
All that is required is to set the completion counter 200 and write the image data to the memory. Then, the value to be set may be changed and read while leaving the difference between the start counter value and the completion counter value unchanged. In other words, if you set the start counter = 50 and the completion counter = 150, the image data will be output at a position 500 pixels to the left in the horizontal direction, and the start counter = 16.
If 0 and completion counter=260 are set, the reference image data will be output at a position 600 pixels to the right in the horizontal direction, and the reference image data stored at an arbitrary coordinate position will be output.

Thereby, the image data to be inspected can be input in real time, and pattern matching is performed by matching the output timing of the reference image data to this real time input. As described above, the function of moving the pattern generation of the reference image to an arbitrary coordinate position is used to achieve the purpose of performing pattern matching in an arbitrary inspection image range of the image to be inspected.

Next, another embodiment of the present invention will be described with reference to the drawings. FIG. 2 is a block diagram showing a block circuit of a pattern recognition device according to a second embodiment of the present invention. As shown in FIG. 1, this block diagram includes a binarization circuit 12 that digitizes the video signal from the imaging device 11, and a characteristic pattern of the binarized image from the binarization circuit 12 that is preliminarily set as a reference image. A reference image storage circuit 16 for storing and pattern matching with the image to be inspected, and two from the binarization circuit 12.
It consists of six variable-length horizontal scanning line shift registers that can hold the digitized image data to be inspected from the 1st line to the 16th line in the horizontal direction as arbitrarily set according to instructions from the microcomputer 13. The variable-length horizontal scanning line shift register 20 and the binarization circuit 12 form a line 7 that is evenly spaced at a set interval in the vertical direction of the binarized image.
The vertical image data extraction circuit 14 outputs the image data to be inspected for each row at the same time, and the binarized image data to be inspected from the binarization circuit 12 is extracted from 1 pixel in the horizontal direction to 166 pixels.
By connecting seven variable-length pixel shift registers in series that can hold the number of pixels arbitrarily set according to instructions from the microinputer 13, one horizontal binarized image can be created at a set interval at the same time. Vertical image data extraction circuit 1 outputs data from seven equally spaced locations.
By providing 7 stages to connect the binarized inspected image data of the I stroke output from 4, the total of the binarized images is 7×7 (=
49) Horizontal image data extraction circuit 15 that simultaneously outputs binarized image data to be inspected at a point, image data from the reference image storage circuit 16, and horizontal image data extraction image pattern to be inspected from the mosquito track 15. The pattern matching output from the 49 exclusive OR means that compares data patterns is ANDed with the clock pulse of the same frequency as the pixel, and if the patterns match as a result of pattern matching, it is determined that each pixel matches. a logic operation circuit 17 which outputs a matching pulse having the same frequency as that of the pixel, and a logic operation circuit 1
a correlation value accumulating circuit 18 which individually counts each matched pulse output from 7 to accumulate and hold correlation values at each coordinate position where each of the 49 patterns has been matched; In order to detect candidate points at high speed by comparing the correlation value data with a detection threshold value set in advance by the microcomputer 13, a comparator corresponding to each output of the correlation value integration circuit 18 is provided for detection. 4 latches that compare and judge the size with the latch that holds the Surenyu value and hold the result.
The microcomputer 13 has a latch section that collects the comparison results of one horizontal scanning line so that they can be grasped in one read by the microcomputer 13.
It is constituted by a horizontal candidate point detection circuit 19 which easily specifies the calculation corresponding position by providing only a book.

The operation of each circuit section configured as described above will be explained in detail. In an actual design example, the field of view of the imaging device 11 is quantized into a grid of 320 pixels horizontally and 240 pixels vertically, and the size of the reference image can be any rectangular range within 128×128 pixels. Reference image storage circuit 1
The configuration of 6 is provided with a total of four counters in the horizontal and vertical directions to generate timings for starting and completing operations, generates a timing signal for reading or writing the reference image, and uses the readout timing signal to read or write the reference image. Image data in the reference image storage circuit is input/output by operating a counter that counts clocks corresponding to pixels during a writing period. This makes it possible to arbitrarily change the pattern size of the reference image by changing the set values for the counters that generate the timing signals for starting and completing the operation of the reference image storage circuit 16, respectively.

In this example as well, the size of the reference image is 128x12.
Although a rectangular range of 8 pixels or less is used, it can be designed arbitrarily.

Second, the vertical image data extraction circuit 14 simultaneously converts the image to be inspected into a video signal from the imaging device 11 and outputs the image to be inspected, which is binarized by the binarization circuit 12. This is the part for output. The vertical image data extraction circuit 14 will be explained using the block circuit diagram shown in FIG.

If the coordinates for pattern matching in one scanning period are 7 points in the horizontal direction and 7 points in the vertical direction, the variable length horizontal scanning line can hold image data to be inspected up to 16 horizontal scanning lines, which can be set in units of 1 horizontal scanning line. Six shift registers 20 are required. The six variable length horizontal scanning line shift registers 20 are connected in series as shown in FIG.
From 0 onwards, image data to be inspected is output. In order to configure image data of any length from 1 horizontal scanning line to 16 horizontal scanning lines as one shift register that can hold the number of horizontal scanning lines arbitrarily set according to instructions from the microcomputer 13, its storage capacity is required. is 320x16 if there are 320 pixels in one horizontal scanning line.
=5120 pixels (=5kbit) are required, so
Generally, a RAM with a capacity of 8 kbit or more may be used.

To hold image data to be inspected for horizontal scanning lines of arbitrary length, the lower 9 bits of the RAM address should be connected directly to the RAM as a counter that simply counts the basic clock of the pixel, and the remaining upper addresses should be used as a counter that counts the vertical scanning lines. Alternatively, a cyclic counter may be used that clears the count value at an arbitrary set number of times (1≦k≦16) and starts counting again from the beginning. In this way, the seven points VIDT, V2DT, V3DT, and V4D of the two-dimensional inspected image data are
Image data to be inspected is simultaneously output from T, V5DT, VE3DT, and V7DT.

Next, 7 points VIDT, V2DT, V3DT.

The image data to be inspected from V4DT, V5DT, V8DT, and V7DT is input to the horizontal image data extraction circuit 15. FIG. 4 is a block diagram showing the configuration of the horizontal image data extraction circuit 15. Pixel data to be inspected of an arbitrary length up to 16 pixels on a horizontal scanning line is divided into pixel data arbitrarily set according to instructions from the microcomputer 13 (1≦j≦1).
6) of variable length pixel/shift registers 21 that can hold 6) in series with respect to the input of image data to be inspected in the horizontal direction.
Make this connection. Since there are six variable length pixel/shift registers 21, the output is six, but the input image data V
IDT-V7DT can also be used as an output, so including this image data, the output from one horizontal line is 7
Output VIDT, VIHl, VIH2, VIHS, VIH
4, VIH5° VIH6. 7 vertical points of image data to be inspected VIDT, V2DT, V3DT, V4D
T.

V5DT, V8DT, V7DT17) Since seven of these six variable length pixel shift registers 21 are independently required for each input voltage, a total of 42 variable length pixel shift registers 21 are used.

NPC's 8 is used as the variable length pixel/shift register 21.
M5828 is used. This IC allows the input of image data to be inspected from up to eight vertical image data extraction circuits 14, so by outputting the output from seven 8M5828s connected in series to the outside, one pixel at a time in the horizontal direction can be input. 49 points of image data to be inspected, which are shifted by pixels in the vertical direction, are output at the same time.

The logic operation circuit 17 extracts the reference image data in the reference image storage circuit 16 and the 49 inspected image data VIDT, VIHI at each coordinate position from the horizontal image data extraction circuit 15.
, VIH2, VIHS--VIH6, V2DT, V2H
1-@-- This is the part that performs an exclusive OR operation on the two data of V7H6 at the same time and outputs the result as a matching pulse. The logic operation circuit 17 as shown in the block diagram shown in FIG. 5 performs an exclusive OR operation simultaneously and independently on 49 types of serially input individual 0 image data VIDT, VIHI, VIH2, VIHS...V
IH6, V2DT, V2H1...V7H6 and reference image data are compared and calculated, and if the result of the comparison is the respective two manual image data, if they match, it is outputted as a pulse output to the outside. Matching pulse pH , Pl2.
Pl 3 ・ ・ Pl 7. P21. P22
・ ・ P27 Fist ・P37... Consists of 49 logic operation units 22 that calculate P77 for each serially inputted image data and serially output it.

The configuration of each logical operation unit 22 will be explained using the block diagram of FIG. 6. The logic operation unit 22 performs a two-man exclusive OR operation on the image data to be inspected and the reference image data, and the match/mismatch output from the exclusive OR unit 23 is converted into one pixel of the image data. The logical product unit 24 is configured to logically AND the pixel clocks corresponding to the input image data and output a serial signal as a matching pulse to inform the outside if the input image data match.

In the embodiment of the present invention, the operation of the logical operation unit 22 is configured such that if the two human-powered image data match, it outputs a matching pulse that is outputted to the outside as a pulse output, but when they do not match, Of course, it is also possible to perform pattern matching by searching for a position where the number of pulses is small with a configuration in which mismatched pulses are outputted to the outside as pulse outputs.

The correlation value integration circuit 18 is configured by the block diagram shown in FIG. /matching pulse P11°Pl2. which is a real signal output. Pl3...Pl7. P21
．． Forty-nine independent counters 25 are provided in a section that integrates P22...P27...P37...P77 as a correlation value. Since the size of the reference image is within 128x128 pixels, in this case, the maximum correlation value is 16384, so 49 binary counters of 14 bits or more may be used.

In addition, the horizontal candidate point detection circuit 19 in the second embodiment simultaneously compares and determines the magnitude of each correlation value of the correlation value integration circuit 18 with a detection threshold value set in advance by a microcomputer. This is provided so that the state of the correlation values at the seven coordinate positions of the scanning line can be read by the microcomputer 13 once. 9 is a block diagram showing the configuration of the horizontal candidate point detection circuit 19. FIG. The horizontal candidate point detection circuit 19 will be described below with reference to FIG.

The horizontal candidate point detection circuit 19 includes counters 25 . ．． A detection threshold holding latch 26 holds the comparison value by writing data from the microcomputer 13 to determine whether the value of 25 is larger than a certain correlation value. Outputs D11, D12. D13...D17. D21. D
22...D27...D37...D77 independently compare the magnitude with the detection threshold value held in the detection threshold holding lunch 26, and hold the results, and the microcomputer 13 grasps the status. It is composed of 49 detection circuits 27 capable of When the 49 detection circuits 27 are grouped into 7 horizontal comparison lines,
This means that one comparison line holds the results of pattern matching at seven points. 1 of the results of these 7 points
Since each result is the result of comparing the magnitude with the detection threshold value, each result can be regarded as 1 bit of information, and one horizontal comparison line can provide 7 bits of information. By assigning this 7-bit information as one I10 address, one Ilo address of the microcomputer 13
It is possible to understand the pattern matching status at seven points by reading the data. The configuration of each detection circuit 27 will be explained using the block diagram of FIG. The detection circuit 27 is comprised of a digital comparator 28 that performs two-way data comparison between a detection threshold value and a correlation value, and a result holding latch 29 that holds the result of the magnitude comparison output from the digital comparator 28.

If the horizontal candidate point detection circuit 19 is not built in, it is necessary to read the correlation value data seven times to understand the state of the correlation value at the seven coordinate positions, and compare the magnitude with the coarsely detected thread value by the microcomputer 13. However, by incorporating the horizontal candidate point detection circuit 19, it is equivalent to reading the I10 data once and performing the above processing contents, so that the processing time can be shortened accordingly. In order to provide this horizontal candidate point detection means for each of the seven horizontal scanning lines, the state of the horizontal candidate point detection circuit 19 is read seven times, the correlation value data is read 49 times, and the microcomputer 13 reads the correlation value data 49 times. The comparison with the coarse detection threshold value is performed on behalf of the user, and the pattern matching position can be identified at a faster processing speed by comparing the magnitude with the detection threshold value.

As mentioned earlier, when reading the state by the microcomputer 13, it is possible to read the state of correlation values at seven coordinate positions of one horizontal scanning line in one read.
When using a 16-bit microcomputer 13, it is possible to grasp the state of the correlation value at 7 coordinate positions of 2 horizontal scanning lines in one data read, or when using a 32-bit microcomputer 13, it is possible to grasp the state of correlation values at 7 coordinate positions in one data read. It is possible to grasp the state of the correlation values at the seven coordinate positions of the horizontal scanning line. In this way, by designing the microcomputer 13 used for control and its connection configuration,
Of course, the number of correlation value states that can be determined in one data read is a plurality of horizontal scanning lines. In addition, if the number of pattern matching points in one horizontal scanning line is increased, if a 16-bit microcomputer 13 is used, it is possible to grasp the state of correlation values up to 16 coordinate positions with one data read, whereas a 32-bit microcomputer 13 can grasp the state of correlation values up to 16 coordinate positions. When the computer 13 is used, it is possible to grasp the state of correlation values at up to 32 coordinate positions by reading data once. When a 16-bit or 32-bit microcomputer 13 is used, the recognition accuracy φ test range of the recognition target is determined by the number of horizontal and vertical pattern matching points than the total number of pattern matching points. It is also possible to do this in accordance with various conditions such as.

In the apparatus according to these embodiments, the total number of horizontal and vertical pattern matching is 7×7=49 points, or 5 points.
Of course, it is possible to set it to any design value such as Obviously, it would be better to provide a larger number of pattern matching points. Increasing the number of pattern matching points is advantageous because it becomes possible to narrow the pixel interval for matching and increases the probability of completing pattern matching with a smaller number of pattern matchings.

The recognition device of the present invention is one of the rough and fine detection methods, and the method for determining the pixel spacing when performing rough detection will be described below to explain how to move from rough detection to fine detection. In the first rough detection, both the horizontal image data extraction circuit 15 and the vertical image data extraction circuit 14 perform inspection image data VIDT, VIHI, VIH2, V at 8 pixel intervals.
IHS・VIH6, V2DT, V2H1・Fist・V7
HS is set to be output, and the center position of each inspection position in 7 rows and 7 columns (the position in the 4th row and 4th column) corresponds to the center position of the detection range of the image to be inspected. Then, based on the correlation values of the position data of the 49 candidate points from the first time, a second detection is performed using the coordinate position corresponding to the part that holds the maximum correlation value among the correlation count values. The coordinate position to be performed. The matching pixel spacing for the second and subsequent pattern matching is determined by selecting the next detection pixel spacing so that the next inspection image range includes the pixel spacing of the rough detection performed first in both the horizontal and vertical directions. Assuming that the inspection range is gradually narrowed, the coordinate position with the maximum correlation value will not be missed even though it is included in the inspection image range.

Therefore, in general, the pixel interval for p-th pattern matching in the horizontal direction is dkp, the pixel interval for p+1-th pattern matching is dk (1) + 1), and the number of points for pattern matching in one horizontal scanning line is an odd value HS. Then, the pixel interval dk (p+1
) is dKp-dk(p+I)X(H5-1)/2≦O...
... Selecting the smallest value among the values that satisfy (1) is an efficient selection method in order to converge the inspection image range with fewer rough detection processing times.

The same applies to the efficient selection of pixel spacing for vertical pattern matching.

Therefore, in the case of this embodiment, dkl=8. Since HS=7, the matching pixel interval dk2 for performing the second pattern matching is 8-dk2X(7-1)/from equation (1).
2≦0・・・(2) Transform this for dk2, dk2 ≧8X2/(7-1)=16/2:2.7・-
...-(3) Therefore, dk2=3 is obtained from the minimum integer value that satisfies the above equation (3). The matching pixel interval dk3 for performing the third pattern matching is dk2=3°HS
= 7, so by substituting into equation (1), d
Obtain k3=1. The second pattern matching pixel interval dj2 in the vertical direction is djl=8°VS=7, which is the same as in the horizontal direction, so dj2=3, and the matching pixel interval dj3 for the third pattern matching is: dj2=3. VS
=7, so dj3=1 is similarly obtained from equation (1).The detection operation will be described below.

Horizontal image data extraction circuit 15 for the first rough detection
And the data is set so that the microcomputer 13 outputs the image data to be inspected to the vertical image data extraction circuit 14 at intervals of 8 pixels in both the horizontal and vertical directions. Also,
The output position of the reference image data is controlled by the microcomputer 13 so that the center position (4th row, 4th column) of each inspection position for pattern matching in 7 rows and 7 columns is placed at the center position of the inspection range of the image to be inspected. . Controlling the output position of the reference image data so that the center position (4th row, 4th column) of each inspection position for pattern matching is arranged specifically means the variable length of the 4th row, 4th column shown in Figure 3. In the image data to be inspected output from the pixel/shift register 21, assuming that the partial image 6 shown in FIG. 7 is placed at the center position to be inspected, the first row and the This means that the image data of the first row and first column of the reference image 7 is output in accordance with the scanning timing at which the image data of the first column is output. Therefore, the same m rows and n columns of partial images 6 are placed at intervals of 8 pixels above, below, and to the left and right of the placed partial images 6, respectively.

Next, all the count values of the 49 correlation value integration circuits 18 are cleared and pattern matching is started. After the first pattern matching, the correlation data of each counter 25 in the correlation value integration circuit 18 is read and the correlation data having the maximum value is searched for, and the pattern matching coordinate position that held the maximum correlation value is searched. is calculated, and if the value is larger than the coarse detection threshold value at 8-pixel intervals, the pattern matching coordinate position that held the maximum correlation value is set as the coordinate position where the second detection is performed. Next, in preparation for the second pattern matching, the microcomputer 13 sends the horizontal image data to the horizontal image data extraction circuit 15 and vertical image data extraction circuit 14.
The data is set so that the image data to be inspected is output at 3-pixel intervals in both the vertical direction, and the reference image data is set so that the pattern matching is performed centering on the coordinate position calculated in the first pattern matching process. The output position of each test position is controlled so that the center position (4th row, 4th column) is located, the detection threshold value is set, and the 49 correlation value integration circuits 18
All count values are cleared and the second pattern matching is started.

After the second pattern matching, the correlation data is read and the data having the maximum value is searched for, and the pattern matching coordinate position that held the maximum correlation value is calculated.

Next, in preparation for the third pattern matching, the microcomputer 13 sends data to the horizontal image data extraction circuit 15 and the vertical image data extraction circuit 14 in both the horizontal and vertical directions.
The data is set so that the image data to be inspected is output at pixel intervals, and the output position of the reference image data is set so that pattern matching is performed centering on the coordinate position calculated in the second pattern matching process. Center position of each inspection position (4 rows 4
column), the detection threshold value is set, all the count values of the 49 correlation value integration circuits 18 are cleared, and the third pattern matching is started.

After the third pattern matching, the data in the correlation value integration circuit 18 is read by the microcomputer 13, and the data having the maximum value is searched for, and the maximum correlation value is larger than the predetermined detection value. If it is a value, the pattern matching coordinate position corresponding to the portion holding the maximum correlation value is calculated, and the coordinate position is set as the detection position.

In addition, in the second embodiment, 49 correlation value integration circuits 1
Clear all 8 count values, set the coarse detection threshold value at 8 pixel intervals in the detection threshold holding latch 26 of the horizontal candidate point detection circuit 19, clear all 49 result holding latch 29, and perform pattern matching. Start.

After the first pattern matching, the horizontal candidate point detection circuit 19
The contents of the seven result holding latches 29 are read, and the counters 25 in the correlation value integration circuit 18 that hold a correlation value larger than the coarse detection threshold value at 8-pixel intervals are identified, and if there are multiple counters 25, they are The correlation data of the counter 25 is read and the correlation data with the maximum value is searched for, and the coordinate position corresponding to the pattern matching coordinate position that held the maximum correlation value is calculated, and the coordinates for the second detection are calculated. position. Next, in preparation for the second pattern matching, the horizontal image data extraction circuit 15 and the vertical image data extraction circuit 14 are sent to the horizontal image data extraction circuit 15 and the vertical image data extraction circuit 14 by the microcomputer 13 to be inspected at 3 pixel intervals in both the horizontal and vertical directions according to the pixel interval determination method described above. Set the data so that the image data is output, and perform pattern matching in 7 rows and 7 columns centered around the coordinate position calculated in the first pattern matching process.
The output position of the reference image data is set to the center position of each inspection position (4
(row, 4th column), all the count values of the 49 correlation value integration circuits 18 are cleared, and the detection threshold holding latch 26 of the horizontal candidate point detection circuit 19 performs rough detection at 3-pixel intervals. The second pattern matching is started by setting the threshold value and clearing all 49 result holding latches 29.

After the second pattern matching, the horizontal candidate point detection circuit 19
The contents of the seven result holding latches 29 are read, and the counters 25 in the correlation value integration circuit 18 that hold a correlation value larger than the coarse detection threshold value at 3-pixel intervals are identified. The correlation data of the counter 25 is read, the data with the maximum value is searched for, and the coordinate position corresponding to the pattern matching coordinate position that held the maximum correlation value is calculated, and the coordinate position is used for the third detection. shall be. Next, in preparation for the third pattern matching, image data to be inspected is output from the microcomputer 13 to the horizontal image data extraction circuit 15 and the vertical image data extraction circuit 14 at one pixel intervals in both the horizontal and vertical directions. The output position of the reference image data is set at the center position (4 4th column)), and set the detection threshold value at 3 pixel intervals in the detection threshold holding latch 26 of the horizontal candidate point detection circuit 19.
All the count values of the nine correlation value integration circuits 18 are cleared and the third pattern matching is started.

After the third pattern matching, the horizontal candidate point detection circuit 19
The contents of the seven result holding latches 2θ are read, and the counter 25 in the correlation value integration circuit 18 that holds a correlation value larger than the coarse detection threshold value at 3-pixel intervals is identified. The correlation data of the counter 25 is read and the data having the maximum value is searched for, and the coordinate position corresponding to the pattern matching coordinate position that held the maximum correlation value is calculated, and the coordinate position is set as the detection position.

Next, a third embodiment of the present invention will be described using FIG. 8.

In the first embodiment, data from the horizontal candidate point detection circuit 19 is read by the microcomputer 13 to grasp the state at the pattern matching position and search for the maximum correlation value. FIG. 8 shows a block diagram of the recognition device in this embodiment.

The second embodiment differs from the first embodiment shown in FIG. 1 only in that a candidate point detection circuit 10 is further added, and other components are the same.

Therefore, the binarization circuit 12 configuring the device of this embodiment. Vertical image data extraction circuit 14, horizontal image data extraction circuit 1
5. Reference image pattern storage circuit 16. Logic operation circuit 17
．． A description of the correlation value integration circuit 18 is omitted since it overlaps with the description in the first embodiment.

The candidate point detection circuit 10 detects data for each of the seven horizontal comparison lines of the horizontal candidate point detection circuit 19 so that the state of correlation value data at multiple positions of the horizontal candidate point detection circuit 19 can be grasped by one data input. If at least one result holding latch 29 has a value larger than the value set in the detection threshold holding latch 26 in the output of the summarized result holding latch 29 among the seven horizontal comparison lines, that state is recorded for each horizontal comparison line. By retaining the information on the comparison line, the microcomputer 13 is informed that the pattern matching coordinate position that exceeds the detection threshold value is among the 49 points. It was created to understand the overall status as more consolidated information. The block diagram in FIG. 11 shows the configuration of the candidate point detection circuit 10. The outputs of the result holding latch 29 from each of the seven horizontal comparison lines Qll, Q12 . Q13...Q17. Q
21. Q22...Q27...Q37...Q77 determines whether or not there is a correlation value that exceeds the detection threshold value in each horizontal comparison line by ORing the 7 bits in each horizontal comparison line as one group. The state of the seven horizontal comparison lines is held in the seven line detection circuits 30 by using 1-bit information as information on one horizontal comparison line regarding the existence or nonexistence of the line. That is, as shown in Figure 11 (
Qll, Q12. Q13. , Q17), (Q21.
Q22. Q23. , Q27), (Q31゜Q32.Q3
3. ,Q37),,,(Q71.Q72,Q73.,Q
The 7-bit inputs of 77) are logically summed, and the results are stored in the seven line detection circuits 30 to hold information on one horizontal comparison line as 1-bit information. In this embodiment, since 7 horizontal comparison lines are provided, the information on each horizontal comparison line can be 7 bits, and the detection threshold value at all pattern matching positions can be calculated by reading the data of Ilo by the microcomputer 13 once. It becomes possible to know whether there is a coordinate position that can be exceeded.

The candidate point detection circuit 10 is a horizontal candidate point detection circuit 19.
By further consolidating the understanding of the state of the correlation values at the coordinate positions of 9 points, 49 points in the inspection range are being subjected to pattern matching.
The state of the correlation value of all the pieces is checked once by the microcomputer 1.
This is provided to speed up the comparison and judgment process, which can be grasped by data reading by the microcomputer 13 and performed by software by the microcomputer 13. As described in , when a 16-bit microcomputer 13 is used, it is possible to grasp the state of the correlation value at 7 coordinate positions of 2 horizontal scanning lines by reading the data once, using 32 bits.
When the microcomputer 13 is used, it is possible to grasp the state of correlation values at seven coordinate positions of four horizontal scanning lines by reading data once. Due to the design of the microcomputer 13 used for control and its connection configuration, it goes without saying that the number of correlation value states that can be grasped in one data read is a plurality of horizontal scanning lines.

In the device according to this embodiment, the total number of horizontal and vertical pattern matching was 7X7=49 points, but 9X9,1
Of course, it is possible to use any design value such as 1X7.5X15, and to complete the pattern matching process in the image to be inspected faster within one image scanning period, it is possible to use a larger number of patterns. It is obvious that it would be better to provide a comparison point. Increasing the number of pattern matching points is advantageous because it increases the probability of completing pattern matching with fewer pattern matchings.

In the present invention, how to move from coarse detection to precise detection, and the method of determining the pixel interval when performing rough detection is the same as in the first embodiment, so a description of this matter will be omitted. .

The detection operation will be explained below. For the first rough detection, the horizontal image data extraction circuit 15 and the vertical image data extraction circuit 14 are
The data is set so that the image data to be inspected is output at 8-pixel intervals in both the vertical direction. In addition, the output position of the reference image data is set by the microcomputer 13 so that the center position of each inspection position (4th row, 4th column) for pattern matching in 7 rows and 7 columns is located at the center position of the inspection range of the image to be inspected 5. Controlled by Controlling the output position of the reference image data so that the center position (4th row, 4th column) of each inspection position for pattern matching is arranged means, specifically, the variable length of the 4th row, 4th column shown in FIG. In the image data to be inspected output from the pixel shift register 21, assuming that the partial image 6 in FIG.
This means that the image data in the first row and first column of the reference image data 7 is output in synchronization with the scanning timing at which the image data in the first row and first column of the eight-column partial image 6 is output. Therefore, the same 12
The partial images 6 of 8 rows and 128 columns are arranged at 8 pixel intervals. Next, the count values of the 49 correlation value integration circuits 18 are completely cleared, and the coarse detection threshold values at 8-pixel intervals are set in the detection threshold holding latch 26 of the horizontal candidate point detection circuit 19. 7bi of the candidate point detection circuit 10.
t Clear all line detection circuits 30 and start pattern matching.

After the first pattern matching, the contents of the line detection circuit 30 in the candidate point detection circuit 10 are transferred to the microcomputer 13.
The horizontal candidate point detection circuit 19 for the horizontal comparison line checks to see if there is a candidate point at a detection position that exceeds the first rough detection threshold value set in advance. the corresponding result holding latch 29 in
The counter 25 that reads the information of The pattern matching coordinate position corresponding to the portion holding the maximum correlation value among the correlation value counters 25 holding the largest correlation value is set as the center position of the next pattern matching. Next, in preparation for the second pattern matching, the horizontal image data extraction circuit 15
And the data is set so that the microcomputer 13 outputs the image data to be inspected to the vertical image data extraction circuit 14 at intervals of 3 pixels in both the horizontal and vertical directions, and the coordinate position calculated in the first pattern matching process is set. In order to perform pattern matching around Clear all the count values of the value integration circuit 18, set the coarse detection threshold value at 3 pixel intervals in the detection threshold holding latch 26 of the horizontal candidate point detection circuit 19, clear all 49 result holding latches 29, and select the candidate point. Point detection circuit 1
Clear all 007 bit line detection circuits 30,
Start the second pattern matching.

After the second pattern matching, the microcomputer 13 reads the contents of the candidate point detection circuit 10 and checks whether there is a candidate point at a detection position that exceeds the first coarse detection threshold set in advance. The information in the corresponding result holding latch 29 in the horizontal candidate point detection circuit 19 of the horizontal comparison line indicating the existence of the horizontal comparison line is read, and the correlation value at the pattern matching coordinate position corresponding to that information is held. The counter 25 holds the maximum correlation value among the correlation values of the plurality of candidate points for which the second pattern matching was performed, which holds a correlation value larger than the coarse detection threshold value. The pattern matching coordinate position corresponding to the part is set as the center position for the next pattern matching. Next, in preparation for the third pattern matching, image data to be inspected is output from the microcomputer 13 to the horizontal image data extraction circuit 15 and the vertical image data extraction circuit 14 at intervals of 3 pixels in both the horizontal and vertical directions. Set the output position of the reference image data so that the center position (4th row, 4th column) of each inspection position is set so that pattern matching is performed centering on the coordinate position calculated in the first pattern matching process. The detection threshold value is set, the count value of the 49 correlation value integration circuits 18 is completely cleared, and the detection threshold holding latch 26 of the horizontal candidate point detection circuit 19 is set at one pixel interval. Cent the thread value for rough detection and hold all 49 results latch 2
9 is cleared, all 7 bits of the line detection circuit 30 of the candidate point detection circuit 10 are cleared, and the third pattern matching is started.

After the third pattern matching, the microcomputer 13 reads the contents of the candidate point detection circuit 10 and checks whether there is a candidate point at a detection position that exceeds the first rough detection threshold value set in advance. The information in the corresponding result holding latch 29 in the horizontal candidate point detection circuit 19 of the horizontal comparison line indicating the existence of the horizontal comparison line is read, and the correlation value at the pattern matching coordinate position corresponding to that information is held. The counter 25 holds the maximum correlation value among the correlation values of the plurality of candidate points for which the third pattern matching was performed, which holds a correlation value larger than the rough detection thread value. The coordinate position corresponding to the pattern matching coordinate position that was detected is calculated, and the coordinate position is set as the detection position.

As explained above, according to the pattern recognition method or device of the present invention, the size of the reference image pattern can be arbitrarily set, and the position detection of the recognition target, which was nearly impossible in the past, becomes possible. Ta. Furthermore, since it is now possible to set a reference image pattern of any size, it simplifies the registration process, and even when pattern matching is performed in a large image area to be inspected, the processing time is not increased and the positioning can be performed quickly. Detection has become possible.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a first embodiment of a pattern recognition device according to the present invention, FIG. 2 is a block diagram similarly showing the configuration of the second embodiment, and FIG. FIG. 4 is a block diagram illustrating the configuration of the vertical image data extraction circuit in the embodiment; FIG. 4 is a block diagram illustrating the configuration of the horizontal image data extraction circuit; FIG. 5 is a block diagram illustrating the configuration of the logic operation circuit; FIG. 6 is a block diagram illustrating the configuration of each logic operation unit of the logic operation circuit,
FIG. 7 is a block diagram illustrating the configuration of the correlation value integration circuit, FIG. 8 is a block diagram illustrating the configuration of a third embodiment of the present invention, and FIG. 9 is a horizontal candidate in the embodiment of FIG. 8. FIG. 10 is a block diagram illustrating the configuration of the point detection circuit; FIG. 10 is a block diagram illustrating the configuration of each detection circuit of the horizontal candidate point detection circuit; FIG. 11 is a block diagram illustrating the configuration of the candidate point detection circuit; FIG. 12 is a schematic diagram of the pattern matching process that is the premise of the present invention. 10... Candidate point detection circuit, 11... Imaging device, 12... Binarization circuit, 13... Microcomputer, 14...
...Vertical image data extraction circuit, 15...Horizontal image data extraction circuit, 16...Reference image pattern storage circuit, 17...Logic operation circuit, 18... ...Correlation value integration circuit, 19...Horizontal candidate point detection circuit. Figure 3: Total) 1. $ Lead m Brown diagram Figure 12 Image to be inspected Figure

Claims (4)

[Claims] (1) The video signal of the imaging means that captures the image to be inspected is digitized, the characteristic pattern is cut out in any size, and the image data can be output as a binary image as a reference image and stored at any two-dimensional coordinate position. a standard image storage means having a 2D image data to be inspected; and a number (VS-1) of horizontal scanning line holding means capable of holding an arbitrary number dy of horizontal scanning lines of the two-dimensional image data to be inspected for the number of horizontal scanning lines set by the microcomputer. Vertical image data extraction means configured to simultaneously output image data to be inspected corresponding to inspection coordinate positions of multiple lines evenly spaced at horizontal scanning line intervals k (1≦k≦dy); By connecting (HS-1) pixel shift means in series that can hold the simultaneous output of the image data to be inspected from the image data extraction means up to an arbitrary pixel dx in the horizontal direction by the number of pixels set by the microcomputer. , 1 set pixel j with horizontal image data to be inspected at the same time
(1≦j≦dx) The image data to be inspected is outputted simultaneously from HS locations equally spaced at intervals of (1≦j≦dx), and the image data to be inspected simultaneously outputted from VS horizontal scanning lines from the vertical extraction means is input. By providing the (HS-1) serial pixel shift means at the VS stage, the image data to be inspected corresponding to the individual inspection coordinate positions of the total VS x HS point of the two-dimensional image data to be inspected can be simultaneously output. horizontal image data extracting means, which simultaneously stores reference image data from the reference image storage means and image data to be inspected corresponding to each inspection coordinate position from the horizontal image data extraction means, VS×H.
Compare patterns at S points, and output pattern matching corresponding to the inspection coordinate position of each point of VS×HS as VS×HS
logical operation means that simultaneously serially outputs matching data from each part corresponding to each inspection coordinate position; and a logic operation means that simultaneously serially outputs each matching data corresponding to each inspection coordinate position from the logical operation means. The first coarse detection consists of a correlation value accumulation means that completes VS×HS pattern matching in one image scan by counting, and calculates and holds correlation values independently for the portion corresponding to each inspection coordinate position. The horizontal image data extraction means and the vertical image data extraction means are set so that HS×VS pieces of image data to be inspected are output simultaneously at the pixel interval of the first pattern matching, and the detection range of the image to be inspected is Correlation data of HS × VS candidate points that were arranged so that the center of the inspection position of HS row × VS column corresponds to the center position, performed the first rough detection, and performed the first pattern matching. Search for the part that holds the maximum correlation value in
The center coordinate position of the next pattern matching is determined, and the first
The coordinate position of the highest correlation value found in the second rough detection is arranged so that the center of the inspection position in the HS row x VS column corresponds to the coordinate position of the highest correlation value found in the second rough detection. Pixel spacing - second pixel spacing x (HS-1/2)}≦0, and in the vertical direction (first pixel spacing - second pixel spacing x (VS-1)
/2}≦0, and the second pixel interval that satisfies the minimum integer value is arranged so that the center of the HS row × VS column corresponds, and pattern matching is performed. Find the coordinate position of the highest correlation value of the correlation data of the candidate point, and
The center of the HS row x VS column is placed around the coordinate position of the highest correlation value found in the first rough detection, and the pixel spacing in the horizontal and vertical directions is set to 1 pixel interval for pattern matching from the third time onwards. The horizontal direction is {pixel interval of Pth pattern matching - pixel interval of Pth + 1st pattern matching
-1)/2}≦0, and in the vertical direction, (pixel spacing of Pth pattern matching - pixel spacing of Pth+1st pattern matching x (VS-1)/2}≦0, and the minimum integer The coordinate position of the Pth highest correlation value is placed in the center of the HS row x VS column at the pixel interval of the Pth + 1st pattern matching, which is a numerical value, and the HS of the pattern matching performed at 1 pixel intervals. Pattern recognition characterized in that if the maximum value of the correlation data of ×VS candidate points is larger than a predetermined detection threshold value, the coordinate position corresponding to the part holding the maximum correlation value is set as the detection position. Method. (2) Digitize the two-dimensional video signal from the imaging means,
A reference image storage means capable of cutting out an arbitrary size of a characteristic pattern of an image to be inspected and storing it in advance as a reference image and outputting the image data stored at an arbitrary two-dimensional coordinate position; A VS in which (VS-1) horizontal scanning line holding means capable of holding horizontal scanning lines set by a microcomputer are equally spaced at horizontal scanning line intervals k (1≦k≦dy).
vertical image data extraction means for simultaneously outputting image data to be inspected corresponding to individual inspection coordinate positions of a row; and VS simultaneous output of image data to be inspected from the vertical image data extraction means in a horizontal direction. By serially connecting HS pixel shift means capable of holding up to an arbitrary pixel dx by the number of pixels set by the microcomputer, one horizontal direction image data to be inspected can be simultaneously transferred to a set pixel j (1≦ j≦dx), the image data to be inspected that is simultaneously output from VS horizontal scanning lines from the vertical extraction means is outputted simultaneously from HS locations equally spaced at intervals of Horizontal image data extraction means that is configured to simultaneously output the image data to be inspected corresponding to the individual inspection coordinate positions of the total VS x HS point of the two-dimensional image data to be inspected by providing pixel shifting means in series at the VS stage. and VS×HS exclusive ORs for simultaneously pattern comparing the image data from the reference image pattern storage means and the image pattern data to be inspected corresponding to each inspection coordinate position from the horizontal image data extraction means. a logic operation means for simultaneously outputting pattern matching outputs corresponding to each inspection coordinate position as VS×HS matching pulses from an exclusive OR means corresponding to each inspection coordinate position; By simultaneously counting each matching pulse output corresponding to each inspection coordinate position, VS x HS pattern matching is completed in one image scan, and the correlation is calculated and maintained independently in the part corresponding to each inspection coordinate position. The pattern matching state at the candidate point can be detected at high speed by simultaneously comparing and determining the magnitude of the correlation value data of the value accumulating means and the correlation value accumulating means with a pattern detection threshold value set in advance by a microcomputer. , the data output of the correlation value accumulating means is compared with the held detection threshold value, and the magnitude is simultaneously determined between the coordinate positions corresponding to the individual inspection coordinate positions, and the results are held for each inspection coordinate. VS the holding means corresponding to the position
xHS number of inspection coordinate positions are provided, and the microcomputer collects the comparison results of HS locations on one horizontal scanning line, and each pattern of one horizontal scanning line is read by the microcomputer once. In order to quickly grasp the outline of the state of the matching position, by providing a means for holding the state only for the vertical pattern matching VS, it is possible to quickly grasp the outline of the state of each pattern matching position of each horizontal scanning line. It is composed of horizontal candidate point detection means that can easily specify the corresponding position for calculation, and the first rough detection uses the horizontal image data extraction means and vertical image data extraction means to detect the pixels of the first pattern matching. It is set so that HS x VS pieces of image data to be inspected are output simultaneously at intervals, and arranged so that the center of the inspection position in rows HS x columns VS corresponds to the center position of the detection range of the image to be inspected. Perform the first rough detection, and after the first rough detection by the horizontal candidate point detection means, determine whether there is a candidate point at a detection position that exceeds the first rough detection threshold value set in advance by the microcomputer. is checked, and if it exists, reads the information of the horizontal candidate point detection means of the horizontal comparison line indicating the existence, and from the means holding the correlation value at the pattern matching coordinate position corresponding to the information, the The pattern matching coordinate position corresponding to the part that holds the maximum correlation value among the correlation count values of the plurality of candidate points subjected to the first pattern matching is determined as the center coordinate position of the next pattern matching, and the first The coordinate position of the highest correlation value obtained in the second rough detection is arranged so that the center of the inspection position in the HS row x VS column corresponds to the coordinate position of the highest correlation value obtained in the second rough detection. Pixel spacing - second pixel spacing x (HS-1)/2)≦0, and in the vertical direction {first pixel spacing - second pixel spacing x (VS-1)/2}≦0;
The horizontal candidate point detection means is arranged so that the centers of the HS row and VS column correspond to each other at the second pixel interval that satisfies the second pixel interval, and the second pixel is set after the first rough detection by the microcomputer. Find the coordinate position of the highest correlation value of the correlation data of the plurality of candidate points in the second round, which is larger than the coarse detection threshold value, and perform the HS row centering on the coordinate of the highest value position found in the second round of rough detection. × Until the center of the VS column is arranged and the third and subsequent pattern matching is performed at 1 pixel intervals, the horizontal direction is {pixel interval of Pth pattern matching - Pth
+ Pixel interval of first pattern matching × (HS-1)/2
}≦0, and the vertical direction is {pixel interval of Pth pattern matching - pixel interval of Pth+1st pattern matching × (VS
-1)/2}≦0 by arranging the center of the HS row x VS column at the pixel interval of the P+1st pattern matching that satisfies A pattern characterized in that the detection position is the coordinate position having the maximum value of the correlation data of the P+1 single or multiple candidate points, which is larger than the P+1 detection threshold value performed at pixel intervals. Recognition method. (3) A binarization circuit that digitizes the two-dimensional video signal of the imaging device that captures the image to be inspected, and a characteristic pattern that is cut out at an arbitrary size within the memory storage capacity range and set in advance at an arbitrary two-dimensional coordinate position as a reference image. a reference image storage circuit that outputs the image data stored in the image data; and a reference image storage circuit that outputs the image data stored in the binarization circuit, and can hold the binarized image data to be inspected from the binarization circuit for any horizontal scanning line set arbitrarily by instructions from the microcomputer. The image data to be inspected corresponds to inspection coordinate positions evenly spaced at arbitrary horizontal scanning line intervals set in the vertical direction of the binarized image with the binarization circuit. and a vertical image data extraction circuit that simultaneously outputs the output of the binary image data to be inspected from the vertical image data extraction circuit, and retains the output of the binarized image data to be inspected from the vertical image data extraction circuit for arbitrary pixels in the horizontal direction set arbitrarily according to instructions from the microcomputer. By connecting multiple possible variable-length pixels and shift registers in series, one horizontal direction of binarized image data to be inspected can be simultaneously output from multiple locations evenly spaced at a set pixel interval. The serial variable length pixel/shift register is provided with the binarized image data to be inspected simultaneously outputted from the horizontal scanning line from the vertical extraction circuit as an input, and corresponds to the inspection coordinate position of each point of the binarized image. A horizontal image data extraction circuit configured to output binarized image data to be inspected at the same time; and a horizontal image data extraction circuit configured to simultaneously output binary image data to be inspected; Consisting of an exclusive OR circuit that simultaneously compares patterns of inspection image pattern data, the pattern matching output corresponding to each inspection coordinate position is simultaneously output as a matching pulse from an exclusive OR circuit corresponding to each inspection coordinate position. Multiple pattern matching can be completed in one field by individually and simultaneously counting each match pulse output corresponding to each inspection coordinate position from the logic operation circuit and the logic operation circuit, and corresponding to each inspection coordinate position. What is claimed is: 1. A pattern recognition device comprising: a correlation value integration circuit that independently calculates and holds a correlation value in a two-dimensional image; and automatically extracts a specific pattern in a two-dimensional image and uses the coordinate position as a detection position. (4) A binarization circuit that digitizes the two-dimensional video signal from the imaging device; A reference image storage circuit capable of outputting image data stored in dimensional coordinate positions, and binarized image data to be inspected from the binarization circuit up to an arbitrary horizontal scanning line can be arbitrarily processed according to instructions from a microcomputer. Horizontal scanning line shift register (VS
-1) horizontal scanning lines (1≦k≦d
x) VS that are evenly spaced at intervals (k is the pixel interval)
A vertical image data extraction circuit configured to simultaneously output image data to be inspected corresponding to each inspection coordinate position of a row, and a VS number of binary image data to be inspected from the vertical image data extraction circuit. By connecting HS variable-length pixel shift registers in series that can hold simultaneous output up to any pixel in the horizontal direction for the number of pixels arbitrarily set according to instructions from the microcomputer, one horizontal direction can be binarized. The set pixels (1≦j
≦dy) intervals (where j is the pixel interval), the binarized signals output simultaneously from the VS horizontal scanning lines from the vertical extraction circuit are The above-mentioned HS receives the inspection image data as input.
By providing variable length pixel shift registers in series at the VS stage, the binarized image data to be inspected corresponding to the individual inspection coordinate positions of the total VS x HS point of the binarized image can be simultaneously output. an image data extraction circuit, and a VS×HS exclusive OR for simultaneously pattern-comparing the image data from the reference image storage circuit and the image data to be inspected corresponding to each inspection coordinate position from the horizontal image data extraction circuit. a logic operation circuit configured to simultaneously output VS×HS pattern matching outputs corresponding to each inspection coordinate position of the circuit from an exclusive OR section corresponding to each inspection coordinate position; and the logic operation circuit. By simultaneously counting each match pulse output corresponding to each inspection coordinate position from the circuit, VS × HS pattern matching is completed in one field, and calculations and retention are performed independently for the parts corresponding to each inspection coordinate position. The pattern matching state at the candidate point can be detected at high speed by simultaneously comparing and determining the magnitude of the correlation value data of the correlation value accumulating circuit and the pattern detection threshold value set in advance by the microcomputer. In this way, a comparator corresponding to each inspection coordinate position is provided for the output of the correlation value integration circuit to hold the detection threshold value, and the magnitude is simultaneously compared and judged with a latch corresponding to each inspection coordinate position, and the result is held. The number of latches corresponding to the individual inspection coordinate positions to be inspected is provided as many as the latches corresponding to the individual inspection coordinate positions of VS x HS. The state of each pattern matching position of each horizontal scanning line can be grasped quickly by reading the data once.By providing latch parts for only VS of vertical pattern matching, the state of each pattern matching position of each horizontal scanning line can be grasped quickly. A horizontal candidate point detection circuit that can quickly grasp the outline of the calculation and easily specify the position corresponding to the calculation, and an outline of the state of each pattern matching position of the VS horizontal scanning lines of the horizontal candidate point detection circuit. In order to enable high-speed grasping, the output of each latch section corresponding to the individual inspection coordinate position of the VS book of the horizontal candidate point detection circuit is recorded as an individual inspection coordinate so that it can be grasped with one data input by a microcomputer. If there is even one latch that is larger than the threshold value in each latch section corresponding to the position, the state of that horizontal scanning line is maintained and the candidate is notified to the microcomputer that there is a horizontal correlation line that exceeds the correlation threshold value. 1. A pattern recognition device comprising a point detection circuit, wherein a coordinate position having the maximum value of correlation data of a single candidate point or a plurality of candidate points is set as a detection position.