Image feature extractor and an image feature analyzer

• US 6,067,369 A
• Filed: 12/16/1997
• Issued: 05/23/2000
• Est. Priority Date: 12/16/1996
• Status: Expired due to Term

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Kamei, Toshio
• Primary Examiner(s): Au, Amelia
• Assistant Examiner(s): Bali, Vikkram
• Application Number:
  US08/991,304
• Time in Patent Office: 889 Days
• Field of Search: 382/124, 382/125, 382/115, 382/203, 382/204, 382/113, 382/116, 382/127, 382/126, 382/309, 382/190, 382/278, 382/252
• US Class Current: 382/125
• CPC Class Codes: G06V 10/48 by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation
• CPC Class Codes: G06V 40/1365 Matching; Classification

Abstract

For providing an efficient image pattern matching system, an image feature extractor (11) of the invention comprises a feature vector extraction means (101) for extracting a feature vector from the pattern image, a quality index extraction means (106) for extracting quality index information of the pattern image, an error distribution information storing means (105) wherein error distribution information representing correlation between error distribution of the feature vector and the quality index information is prepared, and a confidence attribution means (107) for obtaining confidence estimation information referring to the error distribution information retrieved with the quality index information. The image feature analyzer (51) prepares the error distribution information. An image feature matching system collates feature information of an image pattern extracted by the image feature extractor (11) by comparing it to registered feature information.

Claims

1. An image feature extractor for outputting a feature information data set of a pattern image including data of a feature vector of the pattern image and confidence estimation information of the feature vector attributed to each component of the feature vector, said image feature extractor comprising:

a feature vector extraction means for extracting the feature vector from the pattern image;

at least one quality index extraction means for extracting quality index information of the pattern image;

an error distribution information storing means, each corresponding to each of said at least one quality index extraction means, wherein error distribution information is prepared, said error distribution information including correlation between variance or standard deviation of error distribution of a plurality of feature vectors extracted by said feature vector extraction means and the quality index information extracted from respective one of said at least one quality index extraction means corresponding to said plurality of feature vectors;

a confidence attribution means, each corresponding to each of said error distribution information storing means and obtaining confidence information referring to said error distribution information prepared in respective one of said error distribution information storing means by way of said quality index information extracted from the pattern image by respective one of said at least one quality index extraction means, said confidence information being output as the confidence estimation information when said at least one is one; and

a confidence estimation means provided when said at least one is more than one for obtaining the confidence estimation information by integrating said confidence information obtained by each of said confidence attribution means.

2. The image feature extractor recited in claim 1, said feature vector extraction means comprising a primary feature vector extraction means for extracting the feature vector of the pattern image of a skin pattern by extracting ridge direction information of each sub-region of the pattern image.

3. The image feature extractor recited in claim 1, said feature vector extraction means comprising a primary feature vector extraction means for extracting the feature vector from the pattern image of a fingerprint pattern by extracting information of distances and ridge line numbers between every two of singular points of the fingerprint pattern.

4. The image feature extractor recited in claim 1, said feature vector extraction means comprising a primary feature vector extraction means for extracting the feature vector from the pattern image by sampling discrete Fourier power spectrums in a certain frequency band, said discrete Fourier power spectrums being obtained by performing discrete Fourier transfer of said pattern image.

5. An image feature analyzer for obtaining error distribution information including correlation between variance or standard deviation of error distribution of a plurality of feature vectors and quality index information corresponding to said plurality of feature vectors, said image feature analyzer comprising:

a feature vector extraction means for extracting the plurality of feature vectors from pattern images;

a quality index extraction means for extracting quality index information from said pattern images; and

an error distribution information analyzing means for calculating the error distribution information from distribution among each feature vector set of the plurality of feature vectors and a value of the quality index information of each feature vector belonging to said each feature vector set, said each feature vector set comprising feature vectors extracted more than one pattern images of a same pattern.

6. The image feature analyzer recited in claim 5, said feature vector extraction means extracting each of the plurality of feature vectors from each of the pattern images of skin patterns by extracting ridge direction information of each sub-region of respective each of the pattern images.

7. The image feature analyzer recited in claim 5, said feature vector extraction means extracting each of the plurality of feature vectors from each of the pattern images of fingerprint patterns by extracting information of distances and ridge line numbers between every two of singular points of respective each of the plurality of fingerprint patterns.

8. The image feature analyzer recited in claim 5, said feature vector extraction means extracting each of the plurality of feature vectors from each of the pattern images by sampling discrete Fourier power spectrums in a certain frequency band, said discrete Fourier power spectrums being obtained by performing discrete Fourier transfer of respective each of said pattern images.

9. An image feature extractor for outputting a feature information data set of a pattern image including data of a feature vector of the pattern image and confidence estimation information of the feature vector attributed to each component of the feature vector, said image feature extractor comprising:

a feature vector extraction means for extracting the feature vector from the pattern image;

at least one quality index extraction means for extracting quality index information of the pattern image;

an error distribution information storing means, each corresponding to each of said at least one quality index extraction means, wherein each distribution information is prepared, said error distribution information representing correlation between error distribution of a plurality of feature vectors extracted by said feature vector extraction means and the quality index information extracted from respective one of said at least one quality index extraction means corresponding to said plurality of feature vectors;

a confidence attribution means, each corresponding to each of said error distribution information storing means and obtaining confidence information referring to said error distribution information prepared in respective one of said error distribution information storing means by way of said quality index information extracted from the pattern image by respective one of said at least one quality index extraction means, said confidence information being output as the confidence estimation information when said at least one is one; and

a confidence estimation means provided when said at least one is more than one for obtaining the confidence estimation information by integrating said confidence information obtained by each of said confidence attribution means,

said feature vector extraction means comprising;

a primary feature vector extraction means for extracting the feature vector of the pattern image of a skin pattern by extracting ridge direction information of each sub-region of the pattern image, and

at least one vector transformation means each transforming the feature vector making use of a parameter set.

10. The image feature extractor recited in claim 9, one of said at least one vector transformation means transforming the feature vector making use of a parameter set comprising at least one of primary principal component vectors of distribution of feature vectors each obtained from a data set of a plurality of pattern images in the same way with the feature vector.

11. The image feature extractor recited in claim 9, one of said at least one vector transformation means performing affine transformation of the feature vector making use of each of a plurality of parameter set defined in a certain search range and outputting an affine transformation of the feature vector giving a minimum distance to a reference feature vector.

12. The image feature extractor recited in claim 11, one of said at least on quality index extraction means extracting said quality index information according to said minimum distance.

13. An image feature analyzer for obtaining error distribution information representing correlation between error distribution of a plurality of feature vectors and quality index information corresponding to said plurality of feature vectors, said image feature analyzer comprising:

a feature vector extraction means for extracting the plurality of feature vectors from pattern images;

a quality index extraction means for extracting quality index information from said pattern images; and

an error distribution information analyzing means for calculating the error distribution information from distribution among each feature vector set of the plurality of feature vectors and a value of the quality index information of each feature vector belonging to said each feature vector set, said each feature vector set comprising feature vectors extracted more than one pattern images of a same pattern,

said feature vector extracting means comprising at least one vector transformation means each transforming the plurality of feature vectors making use of a parameter set,

said feature vector extracting means extracting each of the plurality of feature vectors from each of the pattern images of skin patterns by extracting ridge direction information of each sub-region of respective each of the pattern images.

14. The image feature analyzer recited in claim 13, one of said at least one vector transformation means transforms the plurality of feature vectors making use of a parameter set comprising at least one primary principal component vectors of distribution of feature vectors each obtained from a data set of a plurality of pattern images in the same way with the plurality of the feature vector.

15. The image feature analyzer recited in claim 13, one of said at least one vector transformation means performing affine transformation of the plurality of feature vectors making use of each of a plurality of parameter set defined in a certain search range and outputting an affine transformation of each of the plurality of feature vectors giving a minimum distance to a reference feature vector.

16. The image feature analyzer recited in claim 15, said quality index extraction means extracting said quality index information according to said minimum distance.

17. An image feature extractor for outputting feature information data set of a pattern image including data of a feature vector of the pattern image and confidence estimation information of the feature vector attributed to each component of the feature vector, said image feature extractor comprising:

a feature vector extraction means for extracting the feature vector from the pattern image;

at least one quality index extraction means for extracting quality index information of the pattern image;

an error distribution information storing means, each corresponding to each of said at least one quality index extraction means, wherein error distribution information is prepared, said error distribution information representing correlation between error distribution of a plurality of feature vectors extracted by said feature vector extraction means and the quality index information extracted from respective one of said at least one quality index extraction means corresponding to said plurality of feature sectors;

a confidence attribution means, each corresponding to each of said error distribution information string means and obtaining confidence information referring to said error distribution information prepared in respective one of said error distribution information storing means by way of said quality index information extracted from the pattern image by respective one of said at least one quality index extraction means, said confidence information being output as the confidence estimation information when said at least one is one; and

a confidence estimation means provided when said at least one is more than one for obtaining the confidence estimation information by integrating said confidence information obtained by each of said confidence attribution means,

said feature vector extraction means comprising;

a primary feature vector extraction means for extracting the feature vector from the pattern image by sampling discrete Fourier power spectrums in a certain frequency band, said discrete Fourier power spectrums being obtained by performing discrete Fourier transfer of said pattern image, and

a vector transformation means for transforming the feature vector making use of a parameter set comprising at least one of primary principal component vectors distribution of feature vectors each obtained from a data set of a plurality of pattern images in the same way with the feature vector.

18. An image feature extractor for outputting a feature information data set of a pattern image including data of a feature vector of the pattern image and confidence estimation information of the feature vector attributed to each component of the feature vector, said image feature extractor comprising:

a feature vector extraction means for extracting the feature vector from the pattern image;

at least one quality index extraction means for extracting quality index information of the pattern image;

an error distribution information storing means, each corresponding to each of said at least one quality index extraction means, wherein error distribution information is prepared, said error distribution information representing correlation between error distribution of a plurality of feature vectors extracted by said feature vector extraction means and the quality index information extracted from respective one of said at least one quality index extraction means corresponding to said plurality of feature vectors;

a confidence attribution means, each corresponding to each of said error distribution information string means and obtaining confidence information referring to said error distribution information prepared in respective one of said error distribution information storing means by way of said quality index information extracted from the pattern image by respective one of said at least one quality index extraction means, aid confidence information being output as the confidence estimation information when said at least one is one; and

a confidence estimation means provided when said at last one is more than one for obtaining the confidence estimation information by integrating said confidence information obtained by each of said confidence attribution means,

one of said at least one quality index extraction means extracting said quality index information according to a minimum value among distances of the feature vector to a set of reference feature vectors each obtained from a high-quality pattern image in the same way with the feature vector.

19. An image feature extractor for outputting a feature information data set of a pattern image including data of a feature vector of the pattern image and confidence estimation information of the feature vector attributed to each component of the feature vector, said image feature extractor comprising:

a feature vector extraction means for extracting the feature vector from the pattern image;

at least one quality index extraction means for extracting quality index information of the pattern image;

an error distribution information storing means, each corresponding to each of said at least one quality index extraction means, wherein error distribution information is prepared, said error distribution information representing correlation between error distribution of a plurality of feature vectors extracted by said feature vector extraction means an the quality index information extracted from respective one of said at least one quality index extraction means corresponding to said plurality of feature vectors;

a confidence attribution means, each corresponding to each of said error distribution information storing means and obtaining confidence information referring to said error distribution information prepared in respective one of said error distribution information storing means by way of said quality index information extracted from the pattern image by respective one of said at least one quality index extraction means, said confidence information being output as the confidence estimation information when said at least one is one; and

a confidence estimation means provided when said at least one is more than one for obtaining the confidence estimation information by integrating said confidence information obtained by each of said confidence attribution means,

one of said at least one quality index extraction means extracting said quality index information according to confidence of said ridge direction information of each sub-region of the pattern image.

20. An image feature analyzer for obtaining error distribution information representing correlation between error distribution of a plurality of feature vectors and quality index information corresponding to said plurality of feature vectors, said image feature analyzer comprising:

a feature vector extraction means for extracting the plurality of feature vectors from a pattern images;

a quality index extraction means for extracting quality index information from said pattern images; and

an error distribution information analyzing means for calculating the error distribution information from distribution among each feature vector set of the plurality of feature vectors and a value of the quality index information of each feature vector belonging to said each feature vector set, said each feature vector set comprising feature vectors extracted more than one pattern images of a same pattern,

said feature vector extracting means comprising a vector transformation means for transforming the plurality of feature vectors making use of a parameter set comprising at least one of primary principal component vectors of distribution of feature vectors each obtained from a data set of a plurality of pattern images in the same way with the plurality of feature vectors,

said feature vector extracting means extracting each of the plurality of feature vectors from each of the pattern images by sampling discrete Fourier power spectrums in a certain frequency band, said discrete Fourier power spectrums being obtained by performing discrete Fourier transfer of respective each of said pattern images.

21. An image feature analyzer for obtaining error distribution information representing correlation between error distribution of a plurality of feature vectors and quality index information corresponding to said plurality of feature vectors, said image feature analyzer comprising:

a feature vector extraction means for extracting the plurality of feature vectors from pattern images;

a quality index extraction means for extracting quality index information from said pattern images; and

an error distribution information analyzing means for calculating the error distribution information from distribution among each feature vector set of the plurality of feature vectors and a value of the quality index information of each feature vector belonging to said each feature vector set, said each feature vector set comprising feature vectors extracted more than one pattern images of a same pattern,

said quality index extraction means extracting said quality index information according to a minimum value among distances of each of the plurality of feature vectors to a set of reference feature vectors each obtained from a high-quality pattern image in the same way with the plurality of feature vectors.

22. An image feature analyzer for obtaining error distribution information representing correlation between error distribution of a plurality of feature vectors and quality index information corresponding to said plurality of feature vectors, said image feature analyzer comprising:

a feature vector extraction means for extracting the plurality of feature vectors from pattern images;

a quality index extraction means for extracting quality index information from said pattern images; and

an error distribution information analyzing means for calculating the error distribution information from distribution among each feature vector set of the plurality of feature vectors and a value of the quality index information of each feature vector belonging to said each feature vector set, said each feature vector set comprising feature vectors extracted more than one pattern images of a same pattern,

said feature vector extracting means extracting each of the plurality of feature vectors from each of the pattern images of skin patterns by extracting ridge direction information of each sub-region of respective each of the pattern images,

said quality index extraction means extracting said quality index information according to confidence of said ridge direction information of each sub-region of said pattern images.

23. An image feature extractor for outputting a feature information data set of a pattern image including data of a feature vector of the pattern image and confidence estimation information of the feature vector attributed to each component of the feature vector, said image feature extractor comprising:

a feature vector extraction means for extracting the feature vector from the pattern image;

at least one quality index extraction means for extracting quality index information of the pattern image;

an error distribution information storing means, each corresponding to each of said at least one quality index extraction means, wherein error distribution information is prepared, said error distribution information representing correlation between error distribution of a plurality of feature vectors extracted by said feature vector extraction means and the quality index information extracted from respective one of said at least one quality index extraction means corresponding to a said plurality of feature vectors;

a confidence attribution means, each corresponding to each of said error distributing information storing means and obtaining confidence information referring to said error distribution information prepared in respective one of said error distribution information storing means by way of said quality index information extracted from the pattern image by respective one of said at least one quality index extraction means, said confidence information being output as the confidence estimation information when said at least one is one; and

a confidence estimation means provided when said at least one is more than one for obtaining the confidence estimation information by integrating said confidence information obtained by each of said confidence attribution means,

said feature vector extraction means comprising;

a primary feature vector extraction means for extracting the feature vector from the pattern image of a fingerprint pattern by extracting information of distances and ridge line numbers between every two of singular points of the fingerprint pattern, and

at least one vector transformation means each transforming the feature vector making use of a parameter set.

24. An image feature analyzer for obtaining error distribution information representing correlation between error distribution of a plurality of feature vectors and quality index information corresponding to said plurality of feature vectors, said image feature analyzer comprising:

a feature vector extraction means for extracting the plurality of feature vectors from pattern images;

a quality index extraction means for extracting quality index information from said pattern images; and

an error distribution information analyzing means for calculating the error distribution information from distribution among each feature vector set of the plurality of feature vectors and a value of the quality index information of each feature vector belonging to said each feature vector set, said each feature vector set comprising feature vectors extracted more than one pattern images of a same pattern,

said feature vector extracting means comprising at least one vector transformation means each transforming the plurality of feature vectors making use of a parameter set,

said feature vector extracting means extracting each of the plurality of feature vectors from each of the pattern images of fingerprint patterns by extracting information of distances and ridge line numbers between every two of singular points of respective each of the plurality of fingerprint patterns.