Optical pattern recognition using multiple reference images

• US 5,175,775 A
• Filed: 07/22/1991
• Issued: 12/29/1992
• Est. Priority Date: 07/27/1990
• Status: Expired due to Fees

• Current Assignee: Seiko Instruments Incorporated (Seiko Group)
• Original Assignee: Seiko Instruments Incorporated (Seiko Group)
• Inventors: Mitsuoka, Yasuyuki, Iwaki, Tadao
• Primary Examiner(s): Razavi, Michael
• Assistant Examiner(s): KLOCINSKI, STEVEN
• Application Number:
  US07/734,275
• Time in Patent Office: 526 Days
• Field of Search: 382/31, 382/43, 382/37, 382/42, 250/550, 359/561, 359/559, 364/822
• US Class Current: 382/211
• CPC Class Codes: G06V 10/92 using spatial domain filters, e.g. joint transform correlators

Abstract

An optical pattern recognition method which can be applied to a joint transform correlator operative to effect correlation processing between a plurality of reference images and at least a single object image to produce correlation peaks corresponding to the respective reference images. The joint transform correlator is provided with a feedback system effective to regulate the intensity of the respective reference images according to the corresponding correlation peak value so as to noise-shape the correlation peaks to discriminate most-correlated reference images. The pattern recognition method is carried out in the correlator sequentially by first preparing a plurality of initial groups each containing a given number of individual reference images and carrying out the pattern recognition of the object image with respect to each of the initial groups to discriminate each of the most-associated reference images having a maximum correlation coefficient in each initial group. Next, the discriminated most-associated reference images are regrouped into new groups, the number of groups is less than that of the initial groups, and the pattern recognition of the object image with respect to the new groups is again carried out so as to enable next regrouping. Then, the regrouping is repeatedly carried out so as to reduce the number of the groups to thereby form a final group and carrying out the pattern recognition of the object image using the final group of reference images to thereby determine a particular reference image corresponding to the object image.

Claims

1. A pattern recognition method for an optical pattern recognition apparatus which is comprised of means for converting a composite input image composed of a group of reference images and at lest one object image into a composite coherent image, means for Fourier-transforming the composite coherent image, means for Fourier-transforming the composite coherent image to produce a joint Fourier-transformed image of the reference images and the object image, means including an optical spatial modulator for recording an optical intensity distribution of the joint Fourier-transformed image on the optical spatial modulator, means for reading the recorded optical intensity distribution with a coherent light to reproduce a corresponding coherent image, means for Fourier-transforming the reproduced coherent image to produce a correlation image, means including a photodetector for detecting the correlation image to produce a corresponding correlation signal, means for processing the correlated signal to determine two-dimensional correlation coefficients between the object image and each respective one of the reference images, and means including a regulating spatial modulator disposed in an optical path extending through the reference images for regulating brightness of each respective one of the reference images according to the corresponding correlation coefficients so as to discriminate a most-associated reference image relative to the object image to thereby effect pattern recognition of the object image, the pattern recognition method comprising the steps of:

preparing a plurality of initial groups each containing a given number of individual reference images and carrying out the pattern recognition of the object image with respect to each of the initial groups to discriminate each of the most-associated reference images having a maximum correlation coefficient in each initial group;

regrouping all of the discriminated most-associated reference images into new groups the number of which is less tan that of the initial groups, and again carrying out the pattern recognition of the object image with respect to the new groups so as to enable a next regrouping; and

repeatedly carrying out the regrouping so as to reduce the number of the groups to thereby form a final group and carrying out the pattern recognition of the object image using the final group of the reference images to thereby determine a particular reference image corresponding to the object image.

2. A pattern recognition method according to claim 1;

wherein each respective one of the most-associated reference images is discriminated in a reference image group by regulating a brightness of each one of the respective reference images in a nonlinear relationship relative to their corresponding detected correlation coefficients.

3. A pattern recognition method according to claim 1;

wherein the pattern recognition of the object image is effected based on a binarized joint Fourier-transformed image which is binarized in the optical spatial modulator.

4. A pattern recognition method according to claim 1;

wherein the respective reference images are variably gray-scaled in a predetermined manner so as to equalize their contribution in the joint Fourier-transformed image.

5. A pattern recognition method for use with an optical pattern recognition apparatus, comprising the steps of:

providing a plurality of initial groups of individual reference images;

performing pattern recognition of an object image with respect to the individual reference images of each respective initial group so as to determine a most-associated reference image for each initial group, wherein the most-associated reference image of each initial group has a maximum correlation with the object image;

forming a plurality of new groups of individual reference images from eh most-associated reference images, wherein the number of new groups is less than the number of initial groups;

performing pattern recognition of the object image with respect to the individual reference images of each respective new group so as to determine a most-associated reference image for each new group;

repeatedly forming a plurality of new groups of individual reference images from the previously determined most-associated reference images, wherein the number of each successive new groups is less than the number of each previous new groups so as to reduce the number of new groups to form a final group of individual reference images; and

carrying out pattern recognition of the object image using the final group to thereby determine a particular reference image corresponding to the object image.

6. A pattern recognition method according to claim 5;

wherein each most-associated reference image is determined by regulating a brightness of each individual reference image in a nonlinear relationship relative to their corresponding correlation with the object image.

7. A pattern recognition method according to claim 5;

wherein the optical pattern recognition apparatus includes an optical spatial modulator; and

the pattern recognition of the object image is effected based on a binarized joint Fourier-transformed image which is binarized in the optical spatial modulator.

8. A pattern recognition method according to claim 5;

wherein the optical pattern recognition apparatus includes means for forming a joint Fourier-transformed image of the object image and the individual reference images; and

each individual reference image is variably grad-scaled in a predetermined manner so as to equalize their contribution to the joint Fourier-transformed image.