JP3345123B2 - Pattern recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern recognition device for collating a test image with a reference image.

[0002]

2. Description of the Related Art A pattern recognition apparatus for judging the identity of two images has many fields of application such as seal verification and fingerprint verification, and it is required to improve the accuracy of the recognition and to shorten the time required for the recognition. I have.

[0003] As one of these recognition methods, a method of performing pattern matching on a computer is known. In this method, a test image and a reference image are first input to a computer, the background noise of these images is removed, the S / N ratio is improved by thinning and binarization, and then the position and angle are slightly adjusted. The two patterns are shifted one by one to see the coincidence of the two patterns.

[0004] On the other hand, there have been proposed some methods for obtaining the correlation of an image by using an optical operation. In particular, the spatial light modulator is 2
Transform Correlat (Joint Transform Correlat)
ion: JTC; "Yatakai, Fourier transform in optical information processing, optics 21, 6, 392-399") is excellent in high-speed performance. FIG. 11 shows an optical system of a JTC optical correlator which is a main part of such a pattern recognition device. FIG. 12 shows an image obtained in this optical system.

In this JTC optical correlator, a liquid crystal projector 40
The test image (input image) and the reference image displayed on the
04 and at the same time, SLM409 (Spatial Light Mod
ulator: spatial light modulator) (FIG. 12)
(A)). By irradiating the SLM 409 with laser light via the half mirror 411, these images are converted into coherent images, and this coherent image is
05, a Fourier image on the SLM 410 placed on the focal plane of the lens 405 (FIG. 12B)
To form The Fourier pattern applied to the SLM 410 is read out by irradiating a laser beam through the half mirror 412, and this pattern is again subjected to Fourier transform by the lens 406, thereby obtaining an image (hereinafter, referred to as a correlation value) between the input pattern and the reference pattern. A correlation image is obtained (FIG. 12C). The correlation image is detected by the photodetector 415 disposed on the focal plane of the lens 406, and the detection data is monitored by the image measurement device 500.

[0006] As shown in FIG.
It is composed of zero-order light and ± first-order lights appearing on both sides of the zero-order light. This device determines that the reference image and the test image are the same when the intensity of the + 1st-order light or the -1st-order light indicating the degree of correlation of the correlation image exceeds a predetermined value.

[0007]

In the above-described pattern recognition by the computer, since the position and the direction of the test image and the reference image do not match, the position and the angle of the reference image and the test image are shifted little by little. Each time, it was necessary to perform calculations to determine these identities. For this reason, calculations must be performed for a large number of combinations,
This calculation required enormous processing time.

In the above-described pattern recognition by the joint correlation method, if there is a small difference or lack, the correlation signal is not so affected, and accurate pattern recognition cannot be performed.

SUMMARY OF THE INVENTION The present invention has been made in view of such a problem, and has as its object to provide a pattern recognition apparatus capable of accurately and quickly determining the identity of images.

[0010]

In order to solve such a problem, a pattern recognition apparatus according to the present invention is a pattern recognition apparatus for determining the identity between a test image and a reference image. An image rotation unit for rotating one of the test images, and an operation image to be subjected to a correlation operation on the same plane of the test image and the reference image, one of which is rotated by the image rotation unit Image conversion display means for converting and displaying the image, an optical system for obtaining a correlation image by optically correlating the operation image displayed on the image conversion display means, and a photodetector for detecting the correlation image And, based on the correlation image detected by the photodetector, determine the direction match and positional relationship between the test image and the reference image, perform template matching, the test image and the reference image Arithmetic processing means for determining identity with The arithmetic processing means obtains a positional relationship between the test image and the reference image after the rotation process when it is determined that the directions match, and performs the rotation process based on the positional relationship. It is characterized in that the later-described test image and the reference image are coordinate-transformed, and both are superimposed to perform the template matching.

[0011]

According to the configuration of the pattern recognition device of the present invention, first, either the test image or the reference image is rotated by the image rotation means (for example, a digital storage device and a rotation calculation device). Next, the test image and the reference image are converted by the image conversion display means into a target image to be subjected to a correlation calculation on the same plane. Then, the correlation image of the operated image is detected by the photodetector. Subsequently, based on the signal of the intensity, the arithmetic processing unit determines the degree of coincidence and the positional relationship between the test image and the reference image, and then performs template matching to determine the identity between the test image and the reference image. be able to.

Further, the pattern recognition device of the present invention is provided with a holding unit for registering a plurality of reference image groups, and when a signal is input from outside to the holding unit, a reference image corresponding to the signal is output. And input to the image rotation means, it is possible to determine the identity of the test image and the reference image corresponding to this signal.

Further, the image conversion display means outputs image data of the rotated reference image and the test image as image information on the same plane, and an image for converting this image into an image to be subjected to a correlation operation. Even if it is constituted by the conversion means, the same effect as described above is exerted.

The image rotation means and the image conversion display means include a pre-processing spatial light modulation tube. The reference image or the test image is input to the photocathode and converted into an electronic image. Since the rotation is performed by passing the flow through an electric field or a magnetic field, and writing is performed on an image holding surface that can be read by irradiation of light, a complicated calculation accompanying the rotation of these images is not required.

Further, the arithmetic processing means notifies the image rotating means of a rotation instruction with respect to the reference image or the test image based on the correlation image detected by the photodetector. The image rotating means can be controlled based on information on the direction to the image.

In the case where the image rotating means is controlled by the arithmetic processing means, the arithmetic processing means first sets the intensity of the photodetector at each predetermined rotation angle when the rotation angle of the image rotation means reaches a predetermined value. Is smaller than a predetermined value, it is determined that the test image and the reference image are not the same. Next, at the point in time when the rotation angle reaches a predetermined value, if the intensity signal for each predetermined rotation angle is larger than the predetermined value, the image rotation means is rotated by the rotation signal so that the intensity signal becomes substantially maximum. Control. Then, when the intensity signal reaches a rotation angle at which the signal becomes substantially maximum, template matching between the reference image and the test image is performed. When the overlapping ratio of these images is equal to or less than a predetermined value, it is determined that the reference image and the test image are not the same, and when the overlapping ratio is larger than the predetermined value, it is determined that they are the same.

When the image rotating means is not controlled by the arithmetic processing means, the arithmetic processing means firstly sets the position of the test image or the reference image rotated by the image rotating means to a predetermined rotation angle. When the signal of the intensity for each predetermined rotation angle is equal to or smaller than a predetermined value, it is determined that the reference image and the test image are not the same. Next, when the rotation angle reaches a predetermined value,
When the signal of the intensity for each predetermined rotation angle is larger than the predetermined value, the rotation angle at which the intensity becomes substantially maximum is calculated by interpolation. Then, template matching between the reference image and the test image is performed at a rotation angle at which the signal of this intensity becomes substantially maximum. If the overlapping ratio of the images is equal to or less than a predetermined value, it is determined that the images are non-identical. If the overlapping ratio of the images is greater than a predetermined value, it is determined that the reference image and the test image are the same. I do.

[0018]

An embodiment of the present invention will be described below with reference to the accompanying drawings. In the description, the same elements will be denoted by the same reference symbols, without redundant description.

(First Embodiment) FIG. 1 shows the configuration of a pattern recognition apparatus according to a first embodiment of the present invention. Note that FIG.
(A) to (c) show images obtained in each apparatus, and FIG. 3 shows a flowchart for explaining the procedure (first procedure) of this embodiment. Hereinafter, the configuration of the pattern recognition device according to the first embodiment will be described in detail with reference to this flowchart. Each step in the flowchart is indicated by adding "s" to the beginning of the step number, and "Y" if the condition for determination in each step is satisfied, and "N" if not satisfied. It is shown as follows.

First, signals of a test image and a reference image in pixel units (referred to as a test signal and a reference signal, respectively) are input to a storage area in the image rotating unit 101 (step 20). Next, the reference signal for each frame in the storage area is transmitted to the image rotation unit 10.
The rotation is converted by the rotation calculation unit in 1. Then, the input signal output from the image rotation unit 101 and the rotation-converted reference signal are input to the storage areas a and b in the image presentation unit 102, respectively, and are converted into one-frame video signals. Thereafter, the video signal is applied to the image display device 103 (for example,
The image is input to a liquid crystal display or a CRT, and a test image and a reference image are projected on an image display surface of the image display device 103 (step 21).

Here, the rotation conversion performed by the image rotation unit 101 is a conversion that rotates the reference image projected on the image display device 103 in a predetermined coordinate system. The test image and the reference image are set in advance so as to be displayed at the same magnification.

The image projected on the image display surface of the image display device 103 is subjected to a correlation operation by JTC (step 22). That is, the test image and the reference image displayed on the image display device 103 are placed on the writing surface of the liquid crystal SLM 109 provided opposite to the front surface of the image display device 103, between the optical paths of the image display device 103 and the liquid crystal SLM 109. (FIG. 2 (a)). Here, the laser light emitted from the laser light source 116 is
After passing through the lenses 108 and 107, the liquid crystal SLM 10
9 is illuminated on the reading surface.

The image written on the liquid crystal SLM 109 is read out by irradiating the laser beam on the reading surface of the liquid crystal SLM 109, and is converted into a coherent image.

The converted coherent image is a liquid crystal S
Liquid crystal SLM11 provided opposite to the front surface of LM109
0 is written on the liquid crystal SLM109 and the liquid crystal SLM11.
The image is written as a Fourier image (FIG. 3B) via a lens 105 provided between the optical paths 0 and 0. At this time, the laser light emitted from the laser light source 116 passes through the lens 108 and the lens 107, and then passes through the half mirror 11
4 and is sequentially reflected by the mirror 113 and the half mirror 112 to irradiate the reading surface of the liquid crystal SLM 110. By the irradiation of the laser light, the liquid crystal SLM1
The Fourier image written on the writing surface of No. 10 is read out, again subjected to Fourier transform by the lens 106 provided opposite to the reading surface of the liquid crystal SLM 110, and the Fourier transformed image (hereinafter referred to as a correlation image) (FIG. 3 (c))
Photodetector 11 provided on focal plane 118 of lens 106
5 (for example, a position detection element (PSD) or a photodiode (PD)). The photodetector 115 detects the intensity (hereinafter, correlation value) and the position of the correlation peak of the correlation image.

After the correlation operation by the JTC is performed in this way (step 22), the operation processing section 117 operates as follows.

First, when the correlation value detected by the photodetector 115 is equal to or smaller than a threshold value (for example, 50% of the autocorrelation value) (step 23N), a rotation signal is input to the image rotation unit 101 (L1 in FIG. 1). ), The reference image is allowed to rotate (for example, 6 degrees)
(Step 24). If the rotation angle at this time is not more than 360 degrees (step 25N), the reference image is rotated again by the image rotation unit 101 until the correlation value becomes larger than the threshold value. Note that the initial value of the rotation angle is 0 degree. The correlation value is equal to or less than the threshold value (step 23N);
And when the rotation angle exceeds 360 degrees (Step 25Y)
Determines that the test image and the reference image are non-identical images (step 3
3).

Next, when the correlation value exceeds the threshold value (step 23Y), a rotation signal is input to the image rotation unit 101 (L1 in FIG. 1), and the reference image is reduced in the direction in which the correlation peak intensity increases. Rotate (step 26) and
The correlation calculation by TC is performed (step 27), and until the correlation value becomes smaller than the previous correlation value (step 28Y),
The minute rotation (step 26) is continued.

When the correlation value becomes smaller than the previous correlation value, the position of the correlation peak is detected by the photodetector 115 (step 29). Thereafter, in order to perform template matching based on the signal at this position, coordinate conversion is performed between the reference signal and the input signal for each frame unit so that the test image and the reference image overlap (step 30). If the overlapping ratio (correlation value) between the test image and the reference image is equal to or less than the threshold value (step 31N), the test image and the reference image are determined to be non-identical images (step 33), and If larger (step 31Y), the test image and the reference image are determined to be the same image (step 32).

Here, the principle of template matching (step 30) is the same as that of superposing two films. That is, by superimposing the test image (input pattern) and the reference image (reference pattern), taking an XOR (exclusive OR), and calculating the area of the portion overlapping the same portion, the identity of the two patterns is obtained. Judge. In this case, the magnitude of the correlation value and the threshold value is determined (step 31) based on the ratio of the non-overlapping area to the overlapping area, and the threshold value is determined with required accuracy.

As another example of template matching, there is a method of thinning a reference pattern in order to reduce the influence of noise and blur, and determining the ratio based on the ratio of the image included in the input pattern.

The reference signal and the test signal may be input from the image rotation unit 101 to the arithmetic processing unit 117. Also,
An image rotation unit 101, an image presentation unit 102, an image display device 103,
The arithmetic processing unit 117 can be performed by one computer. As a method of performing the coordinate transformation in the computer, a conventional primary transformation of rotation may be used. The coordinate conversion equation is given by the following equation, where coordinates before conversion are (x, y) and coordinates after conversion are (X, Y). Here, θ is the rotation angle of the image when the center of gravity of the image is the origin.

[0032]

(Equation 1)

In the liquid crystal SLMs 109 and 110, the phase of the laser light is changed according to the intensity of the input light. In the first embodiment, the liquid crystal SLMs 109, 101
However, the same calculation can be performed by using an intensity-modulated SLM. However, this intensity-modulated SLM
Is used, the half mirrors 111 and 110 need to be polarization beam splitters.

When the test image and the reference image are substantially the same and the rotation angles of these images are substantially the same, a correlation image (FIG. 3)
(C)) is composed of zero-order light and ± first-order lights on both sides thereof. By measuring the intensity and position of the ± first-order light (interval between the zero-order light and the first-order light), the rotational angle difference and the position of the test image and the reference image are detected. FIG. 4 shows the correlation image projected on the focal plane 118 of the lens 106 shown in FIG.
It is a top view at the time of installing SD. A primary light image 121 and a −1st order light image 119 are projected on both sides of the 0th order light image 120, and a PSD is arranged on a focal plane of the −1st order light image 119. As described above, since the + 1st-order light and the -1st-order light are at point-symmetric positions with respect to the 0th-order light and have the same intensity, it is not always necessary to measure both.

Next, FIG. 5 shows the relationship between the rotational angle difference between the test image and the reference image and the correlation peak intensity value of ± first order light of the correlation image. As is clear from FIG. 5, the angle tolerance depends on the type of image. For example, in the case of a seal stamp, 50% of the maximum value of the correlation peak intensity value is obtained only when the rotation angle difference is about ± 3.
Up to about a degree shift. That is, this tendency is substantially the same irrespective of the type of image, and when the angle tolerance is about ± 3 degrees, for example, the reference image or the test image is rotated every 6 degrees, and the correlation peak intensity values are obtained. May be compared in order.
If the images are the same, a correlation peak appears at any angle.

As described above, in the first embodiment of the present invention, first, the reference signal of the reference image input to the image rotation unit 101 is rotationally converted, and a correlation image of these images is obtained using JTC. Next, the position and intensity of the correlation peak of the correlation image are detected by the photodetector 115, and based on the detected intensity signal, the arithmetic processing unit 117 converts the rotation signal into an image rotation unit so that the intensity signal becomes maximum. The rotation angle of the two images is adjusted by inputting the information to the input unit 101. Then, by performing template matching based on the signal at the position of the correlation peak so as to superimpose the two images, it is possible to determine their identity.

In this embodiment, by rotating the reference image,
Although the identity of this and the test image has been determined, the test image may be rotated instead of the reference image.

(Second Embodiment) A pattern recognition apparatus according to a second embodiment of the present invention has the same apparatus configuration as the first embodiment shown in FIG. 1, and the processing procedure in the arithmetic processing unit 117 is different. .

FIG. 6 is a flowchart showing a procedure (second procedure) of the pattern recognition apparatus according to the present embodiment. Second
In the procedure, similarly to the first procedure, the rotated rotated test image and reference image are subjected to a correlation operation by JTC and converted into a correlation image.
(Steps 70-74).

Thereafter, the arithmetic processing section 117 operates as follows. First, after the arithmetic processing unit 117 stores the correlation value of the correlation image at each rotation angle (step 75),
If it is determined that the rotation angle of the reference image is not more than 360 degrees (step 76N), a rotation signal is input to the image rotation device 101 to rotate the reference image by an allowable rotation angle (for example, every 6 degrees) ( Step 77).

Next, when it is determined that the rotation angle is greater than 360 degrees (step 76Y), and when it is determined that the maximum correlation value is equal to or smaller than the threshold value (step 78N), the reference image and the test image are not the same. An image is determined (step 86). Also,
It is determined that the rotation angle is greater than 360 degrees (step 76).
Y) If it is determined that the maximum correlation value is larger than the threshold value (step 78Y), either the test image or the reference image is shifted from the angle having the largest correlation value to the angle having the largest correlation value among the angles on both sides thereof. Is slightly rotated (Step 7
9).

That is, for example, when the input or reference image is rotated every 6 degrees, the correlation value (correlation peak intensity value) takes the maximum value at the i-th time (6 × i degrees) and i-
If a large value follows the maximum value at the first time (6 × (i−1) degrees), the micro rotation is the i-th (6 × i degrees).
To the i-1st (6 × (i-1) degrees) from the. At this time, a small rotation angle, determined from the correlation value P _i-1 Metropolitan in the correlation values P _i and one previous angle theta _i-1 in the current angle theta _i (Newton method). Specifically, this value is defined as the rotation index, and the rotation angle index is defined as: (P _i −P _i−1 ) / (θ _i −θ _i−1 ). When the rotation angle index is large, it is far from the maximum value, so a large rotation angle (for example, 1 degree) is adopted. Since the rotation angle index decreases as the peak (maximum value) approaches, the minute rotation angle is reduced accordingly. Then, the minimum value of the minute rotation angle is determined in advance, and a point in time when the minimum value becomes smaller than the minimum value is defined as a peak.

After this minute rotation, the correlation calculation by JTC is performed again (step 80). When the i-th correlation value is larger than the (i-1) -th correlation value (step 81)
Y), a small rotation is performed, and if it is determined that the correlation value is equal to or smaller than the correlation value of the previous angle (step 81N), it is determined that the correlation value at this rotation angle is the maximum value, and the reference image and the reference image are compared. It is determined that there is no angle difference between the test images.

Thereafter, relative positions are detected by the photodetector 115 (step 82), and template matching is performed by parallel movement of the reference image and the test image based on these relative positions (step 83).

Template matching (step 83)
Is performed, it is determined that the overlapping ratio (correlation value) between the test image and the reference image is larger than the threshold value (step 84).
Y), the reference image and the test image are determined to be the same image (step 8)
5) If it is determined that the correlation value is equal to or less than the threshold value (step 84N), the reference image and the test image are determined to be non-identical images (step 86).

As described above, in the second embodiment of the present invention,
Since the control technique such as Newton's method is used to detect the rotation angle that has the maximum value, the rotation angle that has the maximum value can be obtained with high accuracy.

(Third Embodiment) A pattern recognition apparatus according to a third embodiment of the present invention has the same device configuration as the first embodiment shown in FIG. Part 11
7 is different.

FIG. 7 is a flowchart showing a procedure (third procedure) of the pattern recognition apparatus according to this embodiment. Third
In the procedure, similar to the first and second procedures, the rotated rotation of the test image and the reference image is performed by the JTC, and the identity of the test image and the reference image is determined by converting into a correlation image. To obtain the angle at which the maximum value of the intensity of the correlation peak is obtained by interpolation calculation, the arithmetic processing unit 117
There is a difference in that the input of the rotation signal to 01 (L1 in FIG. 1) is not required.

First, in the same manner as in the second procedure, the rotated test image and reference image are correlated by JTC and converted into a correlated image (steps 90 to 94), and the correlation value at each rotation angle is calculated. The reference image is rotated until the rotation angle of the reference image becomes larger than 360 degrees while being stored in the processing unit 117 (steps 95 to 97).

Here, when the maximum value (maximum correlation value) of the correlation peak detected by the photodetector 115 is equal to or less than the threshold value (step 98N), it is determined that the reference image and the test image do not match. If the maximum correlation value is larger than the threshold value (step 98Y), the angle of the maximum correlation value is calculated by interpolation (approximate calculation of the maximum point from the maximum correlation and the correlation values at both sides) (step 96). (Step 99). That is, when the rotation angle φ is a, b, c, and the corresponding correlation value Y is A, B, C, for example,
By substituting these values into a parabolic function represented by Y = K ₁ (φ−K ₂ ) ² −K ₃ , an angle at which the maximum correlation value is approximately obtained is obtained as φ = K ₂ .

At the angle at which the maximum correlation value is obtained, it is determined that there is no angle difference between the reference image and the test image.

Thereafter, the relative peak positions are detected by the photodetector 115 (step 102), and template matching is performed by parallel movement of the reference image and the test image based on the signals at these positions (step 103). . If the correlation value of the image after performing the template matching is larger than the threshold value (step 104Y), the reference image and the test image are determined to be the same image (step 105), and if the correlation value is equal to or smaller than the threshold value. In (Step 104N), the reference image and the test image are determined to be non-identical images (Step 106).

As described above, according to the third embodiment of the present invention, since the angle at which the maximum value of the intensity of the correlation peak is obtained is obtained by the interpolation calculation, the arithmetic processing unit 117 uses the image rotation unit 1.
No input of a rotation signal to L1 (L1 in FIG. 1) is required.

(Fourth Embodiment) FIG. 8 is a block diagram of a pattern recognition apparatus according to a fourth embodiment of the present invention. In this embodiment, in the pattern recognition apparatus shown in FIG. 1, the rotation of an image and the conversion into a coherent image are performed by a pre-processing spatial light modulator (PSLM). That is,
This is an example in which PSLMs 218 and 219 are used instead of the image rotation unit 101, the image presentation device 102, the image display device 103, the lens 104, and the liquid crystal SLM 109 shown in FIG. PSLM
Is an apparatus having an electro-optic crystal on an image display surface and capable of enlarging / reducing or rotating an input image by an electronic lens.

First, the image to be inspected is transmitted through the lens 221 to the PS.
The image is input to the photoelectric surface in the LM 219 and is converted into an electronic image. Thereafter, the electron flow of the electron image is rotated by the magnetic field applied in the PSLM 219, and is written on the SLM provided on the front surface of the PSLM 219. Similarly, the reference image is rotated via lens 220 and written to PSLM 218.

Next, these images are the first image shown in FIG.
As in the embodiment, the light is converted into a coherent image by the laser light from the laser light source 116, Fourier-transformed by the lens 105, and written into the liquid crystal SLM 110. The image written on the liquid crystal SLM 110 is read out by irradiating the reading surface of the liquid crystal SLM 110 with the laser light emitted from the laser light source 116, and the read image passes through the lens 106 again. It is Fourier transformed and forms an image on the focal plane 118 as a correlation image.

Then, the signal of the position and the intensity of the correlation peak of the correlation image is input to the arithmetic processing unit 217, and based on the signal of the intensity, the signal of the intensity becomes the maximum as in the first embodiment. So that the arithmetic processing unit 217
218, 219, the rotation signal is input, and PSLM 218,
219 controls the rotation of any of the images written.

Thereafter, template matching is performed based on the position signal to determine the identity of the test image and the reference image. Further, the arithmetic processing unit 217 controls the PSLMs 218 and 219 such that the test image and the reference image are written in advance in the PSLMs 218 and 219 at the same magnification.

In determining the identity of the images, the same procedure (first to third procedures) as in the first to third embodiments may be used.

As described above, in the fourth embodiment of the present invention, P
Since the rotation of the image is controlled by using the SLMs 218 and 219, the identity of the test image and the reference image can be determined as in the first embodiment. Also, the rotation of this image
Since the PSLMs 218 and 219 are used, there is no need to perform complicated coordinate conversion calculations required for rotation, and therefore, image rotation can be performed at high speed. It is sufficient that at least one of the PSLM 218 and the PSLM 219 has the rotation function.

(Fifth Embodiment) FIG. 9 is a block diagram of a pattern recognition apparatus according to a fifth embodiment of the present invention. The present embodiment is an example in which the pattern recognition device according to the first to third embodiments of the present invention is applied to a seal verification device.

FIG. 9 shows a mark input unit 302 (for example, a mark input unit 302) for inputting a test signal to the image rotation unit 101 and the arithmetic processing unit 117 in FIG. 1 in addition to the device configuration of the first to third embodiments of the present invention.
An image scanner or CCD), a registered image holding unit 301 for inputting a reference signal, and an account number input unit 302 for outputting a reference image to the image rotating unit 101 by inputting an account number. 1 is a semiconductor position detecting element (PSD) 315 that detects the position of a correlation peak.
And a photodiode (PD) 316 for detecting the intensity. FIG. 10 is a flowchart showing the procedure of the fifth embodiment. This procedure differs from the first procedure (FIG. 3) according to the first embodiment in that steps for registering and inputting a reference image and a test image are further added. Hereinafter, the configuration of the fifth embodiment shown in FIG. 9 will be described based on this flowchart.

First, the test image is input from the mark input unit 303 (step 50), converted into a test image signal (test signal) for each pixel, and output. The registered image holding unit 3
01, a reference image input from a mark input unit (for example, a keyboard) 303 and an account number input from an account number input unit 302 are registered in advance.

Next, when an account number is input from the account number input section 302 (step 52), a reference image signal (reference signal) in pixel units is output from the registered image holding section 301 (step 51). . The reference signal and the test signal output in this manner are input to the image rotation unit 101.

These are processed by the same route as in the first embodiment, and the rotation angle difference between the reference image and the test image is made coincident based on the information on the intensity of the correlation peak detected by the PD 316 ( Based on the position signal detected by the PSD 315, template matching is performed so that the test image and the reference image overlap each other, and the identity of these two images is determined (steps 53 to 64).

The arithmetic processing unit 117 may be configured to be divided into a processing unit for performing the above series of processing and a result display unit for displaying a result of the processing. Further, the reference signal and the test signal may be input from the image rotation unit 101 to the arithmetic processing unit 117. Further, the image rotation unit 101, the image presentation unit 102, the image display device 103, the arithmetic processing unit 117, the registered image holding unit 301
Can be performed by one computer.

As described above, in the fifth embodiment of the present invention, first, the reference signal of the reference image is transmitted from the mark input unit 303 to the image rotation unit 1.
01 and the arithmetic processing unit 117, the customer's account number is input from the account number input unit 302, and the reference image corresponding to the account number is stored in the registered image holding unit 301.
Then, by inputting the account number, the reference signal is transmitted to the image rotation unit 1.
01 to determine the identity of the image in the same manner as in the first embodiment, so that when checking the seal at a bank or the like, the user simply inputs the customer's account number and promptly outputs the stamp from the stamp input unit 303. The read test image can be compared with the reference image corresponding to the account number.

In the fifth embodiment, the reference image is rotated by the image rotating unit 101, but this may be performed by using an electron lens such as a PSLM as in the fourth embodiment.

For image rotation, a plurality of rotated reference images may be registered in the registered image holding unit 301 in advance, and the images may be rotated by reading out the reference images.

In the procedure of this embodiment, a step of registering and inputting a reference image and a test image is further added to the first procedure, but this is added to the second and third procedures of the second and third embodiments. It can be realized even by adding such steps.

Note that the present invention is not limited to the above-described embodiment, and various modifications are possible.

For example, in the first to fifth embodiments of the present invention, the image written on the liquid crystal SLM is subjected to the Fourier transform through the lenses 109 and 110, but these may be those for optically Fourier transforming the image. Alternatively, an image on which an intensity-modulated SLM has been written may be read out by irradiating the image with a laser beam, and the image may be Fourier-transformed by passing the image through a lens via a polarizing beam splitter.
Alternatively, Fourier transform may be performed by transmitting a laser beam through a film on which a test image or a reference image is written and then passing the laser beam through a lens.

In the fifth embodiment, the image rotating unit 101
Has rotated the image signal input from the mark input unit 303, but this may be achieved by rotating the marked paper or the like with a stepping motor or the like.

In the first to fifth embodiments, a computer or an electronic lens is used as the image rotating means.
The image may be rotated by an OVE prism.

In the first to fourth embodiments of the present invention, two images of the test image and the reference image are projected on the image display device 103, and in the second embodiment, the projected images are projected on the PSLMs 218 and 219. May be projected onto them.

[0076]

As described above in detail, according to the present invention, the test image and the reference image are rotated by the image rotating means to optically obtain a correlation image of these images, and the The template matching is performed by the arithmetic processing means, which eliminates the need for complicated calculations involved in superimposing the test image and the reference image, unlike the conventional template matching using a computer. The identity of the images can be determined.

Further, since the present invention includes a holding unit for registering a group of reference images, it is possible to manage this signal in association with the reference image, and determine the identity between the reference image and the test image. In this case, simply by inputting this signal, it is possible to easily and quickly quote the reference image and perform pattern recognition.

Further, if the reference image or the test image is rotated by the PSLM, the image can be rotated without performing a complicated calculation accompanying the image rotation. Therefore, the pattern can be recognized at high speed. It is possible.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a configuration of a pattern recognition device according to a first embodiment of the present invention.

FIG. 2 is a view showing an image obtained in the apparatus of FIG. 1;

FIG. 3 is a flowchart when the pattern recognition devices of the first and second embodiments of the present invention determine the identity of a test image and a reference image.

FIG. 4 is an explanatory diagram illustrating a state in which a PSD is installed in a correlation image.

FIG. 5 is a diagram illustrating a relationship between a rotation angle between a test image and a reference image and a relative intensity of a correlation image.

FIG. 6 is a flowchart showing a procedure when the pattern recognition apparatus according to the second embodiment of the present invention determines the identity of a test image and a reference image.

FIG. 7 is a flowchart illustrating a procedure when the pattern recognition apparatus according to the third embodiment of the present invention determines the identity of a test image and a reference image.

FIG. 8 is a diagram showing a configuration of a pattern recognition device according to a fourth embodiment of the present invention.

FIG. 9 is a diagram illustrating a configuration of a pattern recognition device according to a fifth embodiment of the present invention.

FIG. 10 is a flowchart showing a procedure when the pattern recognition apparatus according to the fifth embodiment of the present invention determines the identity of a test image and a reference image.

FIG. 11 is an explanatory diagram illustrating a joint transform correlation method.

FIG. 12 is a diagram showing an image obtained in the apparatus of FIG. 11;

[Explanation of symbols]

101: image rotating unit, 102: image presenting unit, 103: image display device, 104, 105, 106, 107, 108, 22
0, 221, 404, 405, 406 ... lens, 10
9, 110 ... Liquid crystal SLM, 111, 112, 114, 4
11, 412: half mirror, 113: mirror, 11
5,415: photodetector, 116: laser light source, 11
7, 217: arithmetic processing unit, 118: focal plane, 119 ...-
1st order light, 120 ... 0 order light, 121 ... 1st order light, 218, 2
19: PSLM, 301: Registered image holding unit, 302: Account number input unit, 303: Seal input unit, 315: PSD, 3
16 PD, 409, 410 SLM, 403 Liquid crystal projector, 500 Image measuring device.

──────────────────────────────────────────────────続 き Continuing from the front page (72) Inventor Yuji Kobayashi 1126-1, Nomachi, Hamamatsu City, Shizuoka Prefecture Inside Hamamatsu Photonics Co., Ltd. (72) Inventor Tsutomu Hara 1126, Nomachi, Ichinomachi, Hamamatsu City, Shizuoka Prefecture Hamamatsu Photonics Co., Ltd. (56) References JP-A-6-259563 (JP, A) JP-A-4-363617 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06T 7/00-7 / 60 G06E 3/00 G02B 27/00-27/64 G06K 9/62-9/82

Claims (8)

(57) [Claims] 1. A pattern recognition apparatus for judging identity between a test image and a reference image, comprising: an image rotation unit configured to rotate either the reference image or the test image; One of which is subjected to the rotation processing, the image conversion display means for converting the test image and the reference image into a target image to be subjected to a correlation calculation on the same plane and displaying the converted image, and the image conversion display means An optical system for optically calculating the correlation image to obtain a correlation image, a photodetector for detecting the correlation image, and the test image based on the correlation image detected by the photodetector. And determining the coincidence and positional relationship of the direction of the reference image and the reference image, performing template matching, and an arithmetic processing means for determining the identity of the test image and the reference image, the arithmetic processing means, Direction of the test image and the reference image after the rotation processing When it is determined that they match, the positional relationship between the two is obtained, the test image and the reference image after the rotation process are subjected to coordinate conversion based on the positional relationship, and the two are superimposed to perform the template matching. A pattern recognition device characterized by the following. A storage unit for registering a plurality of reference image groups, wherein a signal from the outside is input to the storage unit, and the storage unit outputs the reference image corresponding to the signal; 2. The pattern recognition apparatus according to claim 1, wherein the input is input to an image rotation unit. 3. An image presenting means for outputting, as image information on the same plane, the test image and the reference image rotated by the image rotating means, wherein the image conversion display means outputs the test image and the reference image. 2. The pattern recognition apparatus according to claim 1, further comprising: an image conversion unit that inputs the obtained image information, converts the image information into an image to be subjected to a correlation operation, and displays the image. 4. A digital storage device in which the image rotating means stores digitized data of the test image and data of the reference image, and data of the test image and the reference image in the digital storage device. 2. The pattern recognition device according to claim 1, further comprising: a rotation operation device for performing a rotation operation independently on the data of the pattern. 5. The image rotation unit and the image conversion display unit pass an electron flow generated by inputting the reference image or the test image to a photocathode through an electric field or a magnetic field generated therein. 2. The pattern recognition apparatus according to claim 1, further comprising a pre-processing spatial light modulator for rotating the reference image and writing the image on an image holding surface readable by light irradiation. 6. The arithmetic processing means, wherein the test image or the reference image rotated by the image rotating means is
At the point in time when the rotation angle reaches the predetermined rotation angle, if the signal of the intensity of the photodetector for each predetermined rotation angle is equal to or less than a predetermined value, it is determined that the reference image and the test image are not the same. When the rotation angle reaches a predetermined value, if the signal of the intensity for each predetermined rotation angle is larger than a predetermined value, the rotation angle at which the intensity is substantially maximum is obtained by interpolation calculation, The template matching of the reference image and the test image is performed at the substantially maximum rotation angle, and when the overlapping ratio of the reference image and the test image is equal to or less than a predetermined value, the reference image and the test image are compared. Is determined to be non-identical, and when the overlapping ratio of the reference image and the test image is larger than a predetermined value, it is determined that the reference image and the test image are the same, The pattern recognition device according to claim 1. 7. The image processing device according to claim 1, wherein the arithmetic processing unit notifies the image rotation unit of an instruction to rotate the reference image or the test image based on the correlation image detected by the photodetector. The pattern recognition device according to claim 1. 8. When the signal of the intensity of the photodetector for each predetermined rotation angle is equal to or less than a predetermined value at the time when the rotation angle of the image rotation unit reaches a predetermined value, When it is determined that the test image and the reference image are not the same, and when the rotation angle becomes a predetermined value, the signal of the intensity for each predetermined rotation angle is larger than a predetermined value, The image rotation means is controlled by a rotation signal so that the signal of the intensity is substantially maximum, and the template matching of the reference image and the test image is performed at a rotation angle at which the signal of the intensity is approximately maximum, and the reference is performed. When the overlapping ratio of the image and the test image is equal to or less than a predetermined value, it is determined that the reference image and the test image are not the same, and the overlapping ratio of the reference image and the test image is a predetermined value. If larger, it is determined that the reference image and the test image are the same The pattern recognition device according to claim 7, wherein: