SPATIAL-SPECTRAL FINGERPRINT SPOOF DETECTION

• US 20080025579A1
• Filed: 07/31/2006
• Published: 01/31/2008
• Est. Priority Date: 07/31/2006
• Status: Active Grant

• Current Assignee: HID Global Corporation (Assa Abloy AB), Limidigm Incorporated
• Original Assignee: Lumidigm Incorporated (Assa Abloy AB)
• Inventors: Rowe, Robert K., Nixon, Kristin A., Sidlauskas, David P.
• Granted Patent: US 7,804,984 B2
• Time in Patent Office: Days
• US Class Current: 382/124
• CPC Class Codes: G06F 18/2132 based on discrimination criteria, e.g. discriminant analysis
• CPC Class Codes: G06V 10/143 Sensing or illuminating at different wavelengths
• CPC Class Codes: G06V 10/58 relating to hyperspectral data
• CPC Class Codes: G06V 40/1318 using electro-optical elements or layers, e.g. electroluminescent sensing
• CPC Class Codes: G06V 40/45 Detection of the body part being alive

Abstract

Methods and apparatus are provided of deriving a discrimination feature set for use in identifying biometric spoofs. True skin sites are illuminated under distinct optical conditions and light reflected from each of the true skin sites is received. True-skin feature values are derived to characterize the true skin sites. Biometric spoofs is similarly illuminated under the distinct optical conditions and light reflected from the spoofs is received. Spoof feature values are derived to characterize the biometric spoofs. The derived true-skin feature values are compared with the derived spoof feature values to select a subset of the features to define the discrimination feature set.

Claims

1. A method of deriving a discrimination feature set for use in identifying biometric spoofs, the method comprising:

illuminating each of a plurality of true skin sites under a plurality of distinct optical conditions;

receiving first light reflected from each of the true skin sites;

deriving true-skin feature values for each of a plurality of features from the received first light to characterize the true skin sites;

illuminating each of a plurality of biometric spoofs under the plurality of distinct optical conditions;

receiving second light reflected from each of the biometric spoofs;

deriving spoof feature values for each of the plurality of features from the received second light to characterize the biometric spoofs; and

comparing the derived true-skin feature values with the derived spoof feature values to select a subset of the features to define the discrimination feature set.

2. The method recited in claim 1 wherein:

deriving the true-skin feature values comprises extracting a plurality of true-skin images from the received first light for each of the true skin sites;

deriving the spoof feature values comprises extracting a plurality of spoof images from the received second light for each of the biometric spoofs; and

each of the true-skin images and each of the spoof images corresponds to an image under one of the plurality of distinct optical conditions.

3. The method recited in claim 2 wherein:

deriving the true-skin feature values further comprises decomposing each of the true-skin images into a plurality of different spectral frequency components; and

deriving the spoof feature values further comprises decomposing each of the spoof images into the plurality of different spectral frequency components.

4. The method recited in claim 3 wherein decomposing each of the true-skin images and decomposing each of the spoof images comprises performing a wavelet decomposition.

5. The method recited in claim 3 wherein:

deriving the true-skin feature values further comprises calculating a ratio of a first of the different spatial frequency components for the true-skin images to a second of the different spatial frequency components for the true-skin images; and

deriving the spoof feature values further comprises calculating a ratio of a first of the different spatial frequency components for the spoof images to a second of the different spatial frequency components for the spoof images.

6. The method recited in claim 3 wherein:

deriving the true-skin feature values further comprises calculating an intensity distribution for each of the different spatial frequency components for the true-skin images; and

deriving the spoof feature values further comprises calculating an intensity distribution for each of the different spatial frequency components for the spoof images.

7. The method recited in claim 6 wherein at least one of the features is substantially invariant to illumination intensity.

8. The method recited in claim 7 wherein the at least one of the features comprises a ratio of an intensity at a first predetermined percentile of an intensity distribution to an intensity at a second predetermined percentile of the intensity distribution.

9. The method recited in claim 7 wherein at least a second of the features varies with illumination intensity.

10. The method recited in claim 9 wherein:

the at least one of the features comprises a ratio of an intensity at a first predetermined percentile of an intensity distribution to an intensity at a second predetermined percentile of the intensity distribution; and

the at least a second of the features comprises a difference between the intensity at the first predetermined percentile and the intensity at the second predetermined percentile.

11. The method recited in claim 6 wherein at least one of the features varies with illumination intensity.

12. The method recited in claim 1 wherein:

the true skin sites and the biometric spoofs define separate classes; and

comparing the derived true-skin feature values with the derived spoof feature values comprises calculating ratios of within-class variance to between-class variance for a quantity derived from the features.

13. The method recited in claim 12 wherein the quantity derived from the features comprises a Fisher linear discriminant transform of the features.

14. The method recited in claim 1 wherein comparing the derived true-skin feature values with the derived spoof feature values to select the subset of the features comprises applying a learning algorithm to the features to select the subset of the features.

15. The method recited in claim 14 wherein the learning algorithm comprises a genetic algorithm.

16. A computer-readable storage medium having a computer-readable program embodied therein for directing operation of a computational device to derive a discrimination feature set for use in identifying biometric spoofs, the computational device including a processor in communication with a storage device, the computer-readable program including:

instructions for retrieving, with the processor from the storage device, first data representing properties of first light reflected from each of a plurality of true skin sites under a plurality of distinct optical conditions;

instructions for deriving, with the processor, true-skin feature values for each of a plurality of features from the first data to characterize the true skin sites;

instructions for retrieving, with the processor from the storage device, second data representing properties of second light reflected from each of a plurality of biometric spoofs;

instructions for deriving, with the processor, spoof feature values for each of the plurality of features from the second data to characterize the biometric spoofs; and

instructions for comparing, with the processor, the derived true-skin feature values with the derived spoof feature values to select a subset of the features to define the discrimination feature set.

17. The computer-readable storage medium recited in claim 16 wherein:

the instructions for deriving the true-skin feature values comprise instructions for extracting a plurality of true-skin image representations from the first data for each of the true skin sites;

the instructions for deriving the spoof feature values comprise instructions for extracting a plurality of spoof image representations from the second data for each of the biometric spoofs; and

each of the true-skin image representations and spoof image representations corresponds to an image under one of the plurality of distinct optical conditions.

18. The computer-readable storage medium recited in claim 17 wherein:

the instructions for deriving the true-skin feature values further comprise instructions for decomposing each of the true-skin image representations into a plurality of different spatial frequency components; and

the instructions for deriving the spoof feature values further comprise instructions for decomposing each of the spoof image representations into the plurality of different spatial frequency components.

19. The computer-readable storage medium recited in claim 18 wherein the instructions for decomposing each of the true-skin image representations and the instructions for decomposing each of the spoof image representations comprise instructions for performing a wavelet decomposition.

20. The computer-readable storage medium recited in claim 18 wherein:

the instructions for deriving the true-skin feature values further comprise instructions for calculating a ratio of a first of the different spatial frequency components for the true-skin image representations to a second of the different spatial frequency components for the true-skin image representations; and

the instructions for deriving the spoof feature values further comprise instructions for calculating a ratio of a first of the different spatial frequency components for the spoof image representations to a second of the different spatial frequency components for the spoof image representations.

21. The computer-readable storage medium recited in claim 18 wherein:

the instructions for deriving the true-skin feature values further comprise instructions for calculating an intensity distribution for each of the different spatial frequency components for the true-skin image representations; and

the instructions for deriving the spoof feature values further comprise instructions for calculating an intensity distribution for each of the different spatial frequency components for the spoof image representations.

22. The computer-readable storage medium recited in claim 21 wherein at least one of the features is substantially invariant to illumination intensity.

23. The computer-readable storage medium recited in claim 22 wherein the at least one of the features comprises a ration of an intensity at a first predetermined percentile of an intensity distribution to an intensity at a second predetermined percentile of the intensity distribution.

24. The computer-readable storage medium recited in claim 22 wherein at least a second of the features varies with illumination intensity.

25. The computer-readable storage medium recited in claim 24 wherein:

the at least one of the features comprises a ratio of an intensity at a first predetermined percentile of an intensity distribution to an intensity at a second predetermined percentile of the intensity distribution; and

the at least a second of the features comprises a different between the intensity at the first predetermined percentile and the intensity at the second predetermined percentile.

26. The computer-readable storage medium recited in claim 21 wherein at least one of the features varies with illumination intensity.

27. The computer-readable storage medium recited in claim 16 wherein:

the true skin sites and the biometric spoofs define separate classes; and

the instructions for comparing the derived true-skin feature values with the derived spoof feature values comprise instructions for calculating ratios of within-class variance to between-class variance for a quantity derived from the features.

28. The computer-readable storage medium recited in claim 27 wherein the quantity derived from the features comprises a Fisher linear discriminant transform of the features.

29. The computer-readable storage medium recited in claim 16 wherein the instructions for comparing the derived true-skin feature values with the derived spoof feature values to select the subset of the features comprise instructions for applying a learning algorithm to the features to select the subset of the features.

30. The computer-readable storage medium recited in claim 29 wherein the learning algorithm comprises a genetic algorithm.