Method of recognizing images of faces or general images using fuzzy combination of multiple resolutions

• US 5,842,194 A
• Filed: 07/28/1995
• Issued: 11/24/1998
• Est. Priority Date: 07/28/1995
• Status: Expired due to Fees

• Current Assignee: Government of Japan As Represented By The Ministry of Enconomy Trade and Industry, Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation), Real World Computing Partnership
• Inventors: Arbuckle, Thomas D.
• Primary Examiner(s): Hafiz, Tariq R.
• Assistant Examiner(s): SHAH, SANJIV
• Application Number:
  US08/508,889
• Time in Patent Office: 1,215 Days
• Field of Search: 395/10, 395/2.45, 395/2.65, 395/326, 395/3, 395/24, 382/125, 382/212, 382/156, 382/224, 706/45, 706/52, 706/2, 706/26
• US Class Current: 706/52
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data
• CPC Class Codes: G06V 10/431 Frequency domain transformation; Autocorrelation
• CPC Class Codes: G06V 30/2504 Coarse or fine approaches, e.g. resolution of ambiguities or multiscale approaches
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

A system comprising a neural network, or computer, implementing a feature detection and a statistical procedure, together with fuzzy logic for solving the problem of recognition of faces or other objects) at multiple resolutions is described. A plurality of previously described systems for recognizing faces (or other objects) which use local autocorrelation coefficients and linear discriminant analysis are trained on a data set to recognize facial images each at a particular resolution. In a second training stage, each of the previously described systems is tested on a second training set in which the images presented to the previously described recognition systems have a matching resolution to those of the first training set, the statistical performance of this second training stage being used to train a fuzzy combination technique, that of fuzzy integrals. Finally, in a test stage, the results from the classifiers at the multiple resolutions are combined using fuzzy combination to produce an aggregated system whose performance is higher than that of any of the individual systems and shows very good performance relative to all known face recognitior systems which operate on similar types of training and testing data, this aggregated system, however, not being limited to the recognition of faces and being able to be applied to the recognition of other objects.

Claims

1. A process for recognizing known images of a plurality of objects, the process comprising:

(a) obtaining a plurality of different resolutions for an unknown image;

(b) performing local autocorrelation for each resolution of the unknown image to produce a feature vector for each resolution;

(c) performing linear discriminant analysis for each feature vector, wherein each feature vector from the local autocorrelation is processed employing a transformation matrix, the transformation matrix being calculated with respect to known images required to be recognized, the linear discriminant analysis producing a plurality of outputs;

(d) performing fuzzy integration to merge the outputs from the linear discriminant analysis to produce a merged output for each of the known images, each merged output being indicative of similarity of the unknown image to each known image;

(e) analyzing each merged output to identify a known image corresponding to the unknown image.

2. The process according to claim 1 including calculating the transformation matrix using a neural network.

3. The process according to claim 2 wherein the known images are images of faces.

4. The process according to claim 1 wherein performing fuzzy integration comprises performing Sugeno fuzzy integration.

5. The process according to claim 4 wherein performing fuzzy integration comprises calculating a Sugeno fuzzy measure.

6. The process according to claim 4 wherein the known images are images of faces.

7. The process according to claim 5 wherein the known images are images of faces.

8. The process according to claim 1 wherein obtaining a plurality of different resolutions of an unknown image comprises obtaining multiple copies of the unknown image using an image gathering means.

9. The process according to claim 1 wherein obtaining a plurality of different resolutions of an unknown image comprises applying pixel averaging to the unknown image.

10. A process for evaluating the similarity of a plurality of collections of images and for classifying the collections of images with respect to a known collection of images of objects of the same class, each of the collections containing only images of the same object produced at different orientations and resolutions, the resolutions being known for each image and being the same between the collections of images thereby forming sets of images at differing resolutions, the process comprising:

(a) obtaining a plurality of different resolutions of an image to produce a collection of images to be classified;

(b) providing a plurality of first modules implementing local autocorrelation coefficient feature extraction and providing a plurality of second modules implementing linear discriminant analysis and equal in number to the first modules, each of the first modules providing an output to each of the second modules to form pairs of modules, the second modules storing information characterizing a collection of known images of objects of the same class, the second modules providing information on similarity of images from the collection of images to be classified to the collection of known images, the pairs of local autocorrelation feature analysis modules and linear discriminant analysis modules operating only at resolutions of known and predetermined accuracy for both images of the collection of images to be classified and the collection of known images;

(c) providing a third module for integrating outputs from the pairs of local autocorrelation feature extraction modules and linear discriminant analysis modules, and for merging information from the pairs of modules operating at multiple resolutions, the third module performing fuzzy integration;

(d) training the third module, including measuring the performance of the pairs of local autocorrelation feature extraction modules and linear discriminant analysis modules to provide performance measurement;

(e) combining the output of the pairs of local autocorrelation feature extraction modules and linear discriminant analysis modules based on the performance measurement;

(f) producing an output similar to an output produced by one of the pairs of local autocorrelation coefficient feature extraction modules and linear discriminant analysis modules; and

(g) providing a fourth module for analyzing the output from the third module such that recognition of images from the collection of images to be classified or rejection of the collection of images to be classified is effected relative to the images from the collection of known images, an output of the fourth module being an identification of each of the images from the collection of images to be classified as being similar to images from the collection of known images or as being different from the images in said collection of known images, wherein the pairs of local autocorrelation feature extraction modules and linear discriminant analysis modules in conjunction with the third module and the fourth module permit recognition or rejection of the images from the collection of images to be classified, the recognition or rejection being with respect to the collection of known images, the recognition or rejection having higher performance measurement than a performance measurement of any one of the pairs of local autocorrelation feature extraction modules and linear discriminant analysis modules in conjunction with one of the modules for implementing rejection or recognition.

11. The process according to claim 10 wherein the pairs of local autocorrelation feature extraction modules and linear discriminant analysis modules are implemented by at least one neural network.

12. The process according to claim 10 wherein the collection of known images are images of faces.

13. The process according to claim 11 wherein the collection of known images are images of faces.

14. The process according to claim 10 wherein the third module is implemented by Sugeno integration.

15. The process according to claim 11 wherein the third module is implemented by Sugeno integration.

16. The process according to claim 14 wherein a fuzzy measure employed in the fuzzy integration is a Sugeno fuzzy measure.

17. The process according to claim 15 wherein a fuzzy measure employed in the fuzzy integration is a Sugeno fuzzy measure.

18. The process according to claim 16 wherein the collection of images to be classified are images of faces.

19. The process according to claim 17 wherein the collection of images to be classified are images of faces.

20. A process for recognizing an unknown image with respect to sets of known stored images comprising:

(a) obtaining a plurality of different resolutions of the unknown image;

(b) performing location autocorrelation coefficient feature extraction followed by linear discriminant analysis for each resolution of the unknown image to produce a plurality of outputs;

(c) combining the plurality of outputs into groups for each set of known stored images using fuzzy integration and producing a classifying number for each group;

(d) performing recognition or acceptance of the unknown image by examining the classifying number and comparing the classifying number to a threshold; and

(e) training the step of combining such that the plurality of outputs for the plurality of resolutions are combined in a manner other than averaging.

21. The process according to claim 20 wherein performing local autocorrelation includes using at least one neural network.

22. The process according to claim 21 wherein combining the outputs includes performing Sugeno fuzzy integration using a Sugeno fuzzy measure.

23. The process according to claim 21 wherein the unknown image is an image of a face.

24. The process according to claim 22 wherein the unknown image is an image of a face.

25. The process according to claim 22 wherein the unknown image comprises an image of a machine part.