Hand-based gender classification

• US 8,655,084 B2
• Filed: 06/23/2010
• Issued: 02/18/2014
• Est. Priority Date: 06/23/2009
• Status: Expired due to Fees

• Current Assignee: Board of Regents Nevada System of Higher Education (State of Nevada)
• Original Assignee: Board of Regents Nevada System of Higher Education (State of Nevada)
• Inventors: Bebis, George, Amayeh, Gholamreza
• Primary Examiner(s): Le, Vu
• Assistant Examiner(s): WOLDEMARIAM, AKLILU K
• Application Number:
  US12/821,948
• Publication Number:
  US 20100322486A1
• Time in Patent Office: 1,336 Days
• Field of Search: 382/224, 382/124, 382/115, 382/130
• US Class Current: 382/224
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data
• CPC Class Codes: G06V 10/809 of classification results, e.g. where the classifiers operate on the same input data
• CPC Class Codes: G06V 40/107 Static hand or arm

Abstract

For each of at least one digitally-imaged hand part, where each of the at least one digitally-imaged hand part corresponds to one of a plurality of hand parts, a set of feature parameters representing a geometry of the digitally-imaged hand part is computed. The set(s) of feature parameters for a set of one or more of the digitally-imaged hand parts is/are used to compute distances of the set of digitally-imaged hand parts from each of i) a first eigenspace corresponding to a male class, and ii) a second eigenspace corresponding to a female class. The computed distances are used to classify the gender of a hand as belonging to the male class or the female class.

Claims

1. A method of classifying the gender of a hand having a plurality of hand parts, the method comprising:

for each of at least two digitally-imaged hand parts, where each of the at least two digitally-imaged hand part corresponds to one of the plurality of hand parts, computing a set of feature parameters representing a geometry of the digitally-imaged hand part;

using the sets of feature parameters for a set of two or more of the digitally-imaged hand parts, (a) combining sets of feature parameters for different hand parts to form sets of feature parameters for multiple hand parts, wherein a first set of feature parameters corresponds to a hand'"'"'s fingers and a second set of feature parameters corresponds to the hand'"'"'s palm and (b) computing a similarity of the set of two or more digitally-imaged hand parts to each of i) a first model eigenspace corresponding to a male class, and ii) a second model eigenspace corresponding to a female class, wherein the computed similarities are distances to each of the first and second eigenspaces; and

using the computed distances to classify the gender of the hand as belonging to the male class or the female class comprising;

forming a distance vector from the computed distances;

comparing the distance vector to each of k-nearest neighbor distance vectors in a set of training data; and

classifying the gender of the hand as belonging to the class corresponding to a majority of the k-nearest neighbor distance vectors in the set of training data.

2. The method of claim 1, further comprising:

acquiring a digital image of the hand; and

segmenting the digital image of the hand into the digitally-imaged hand parts, the digitally-imaged hand parts including at least one finger part and a palm part.

3. The method of claim 1, wherein computing the set of feature parameters representing the geometry of a particular digitally-imaged hand part comprises computing a set of contour-based feature parameters for the particular digitally-imaged hand part.

4. The method of claim 1, wherein computing the set of feature parameters representing the geometry of a particular digitally-imaged hand part comprises computing a set of Zernike moments for the particular digitally-imaged hand part.

5. The method of claim 1, wherein computing the set of feature parameters representing the geometry of the particular digitally-imaged hand part comprises computing a set of region-based feature parameters for the particular digitally-imaged hand part.

6. The method of claim 1, wherein computing the set of feature parameters representing the geometry of the particular digitally-imaged hand part comprises computing a set of Fourier descriptors for the particular digitally-imaged hand part.

7. The method of claim 1, further comprising, prior to computing the similarities of the set of digitally-imaged hand parts from each of the first and second eigenspaces, applying Principal Component Analysis to the set(s) of feature parameters for the set of one or more digitally-imaged hand parts.

8. The method of claim 1, wherein using the computed distances to classify the gender of the hand as belonging to the male class or the female class comprises:

classifying the gender of the hand as belonging to the class to which the set of digitally-imaged hand parts is closest.

9. The method of claim 1, wherein using the computed distances to classify the gender of the hand as belonging to the male class or the female class comprises:

combining the distances into a classification score; and

classifying the gender of the hand by comparing i) the classification score to ii) a threshold computed from a set of training data.

10. The method of claim 1, further comprising:

separately performing said computation of distances for each of at least two sets of one or more of the digitally-imaged hand parts, each performance of the computation of distances being undertaken with respect to a different set of eigenspaces, and each of the different sets of eigenspaces corresponding to a particular set of digitally-imaged hand parts;

wherein using the computed distances to classify the gender of the hand as belonging to the male class or the female class includes,

for each particular set of digitally-imaged hand parts, combining the distances of the particular set into a classification score;

in accordance with a weighting function, assigning weights to the classification scores;

combining the weighted classification scores into an overall score for the hand; and

classifying the gender of the hand by comparing i) the overall score for the hand to ii) a threshold computed from a set of training data.

11. The method of claim 10, wherein the overall score is a sum of the weighted classification scores.

12. The method of claim 1, wherein the digitally-imaged hand parts include five digitally-imaged finger parts and one digitally-imaged palm part, the method further comprising:

separately performing said computation of distances for each of the digitally-imaged hand parts, each performance of the computation of distances being undertaken with respect to a different set of eigenspaces, and each of the different sets of eigenspaces corresponding to a particular type of digitally-imaged hand part;

wherein using the computed distances to classify the gender of the hand as belonging to the male class or the female class includes,

for each particular digitally-imaged hand part, combining the distances of the particular digitally-imaged part into a classification score;

in accordance with a weighting function, assigning weights to the classification scores;

combining the weighted classification scores into an overall score for the hand; and

classifying the gender of the hand by comparing i) the overall score for the hand to ii) a threshold computed from a set of training data.

13. The method of claim 1, further comprising:

separately performing said computation of distances for each of at least two sets of one or more of the digitally-imaged hand parts, each performance of the computation of distances being undertaken with respect to a different set of eigenspaces, and each of the different sets of eigenspaces corresponding to a particular set of digitally-imaged hand parts;

wherein using the computed distances to classify the gender of the hand as belonging to the male class or the female class includes,

for each particular set of digitally-imaged hand parts, using the computed distances for the particular set to classify the gender of the particular set; and

classifying the gender of the hand based on the classifications of the at least two sets of digitally-imaged hand parts.

14. The method of claim 13, wherein classifying the gender of the hand based on the classifications of the at least two sets of digitally-imaged hand parts comprises:

classifying the gender of the hand based on a majority vote of the classifications of the at least two sets of digitally-imaged hand parts.

23. The method of claim 1, wherein computing the set of feature parameters representing the geometry of a particular digitally-imaged hand part comprises computing a set of Zernike moments for the particular digitally-imaged hand part.

24. The method of claim 1, wherein computing the set of feature parameters representing the geometry of the particular digitally-imaged hand part comprises computing a set of region-based feature parameters for the particular digitally-imaged hand part.

25. The method of claim 1, wherein computing the set of feature parameters representing the geometry of the particular digitally-imaged hand part comprises computing a set of Fourier descriptors for the particular digitally-imaged hand part.

26. The method of claim 1, further comprising, prior to computing the similarities of the set of digitally-imaged hand parts from each of the first and second eigenspaces, applying Principal Component Analysis to the set(s) of feature parameters for the set of one or more digitally-imaged hand parts.

27. The method of claim 1, wherein using the computed distances to classify the gender of the hand as belonging to the male class or the female class comprises:

classifying the gender of the hand as belonging to the class to which the set of digitally-imaged hand parts is closest.

28. The method of claim 1, wherein using the computed distances to classify the gender of the hand as belonging to the male class or the female class comprises:

forming a distance vector from the computed distances;

comparing the distance vector to each of k-nearest neighbor distance vectors in a set of training data; and

classifying the gender of the hand as belonging to the class corresponding to a majority of the k-nearest neighbor distance vectors in the set of training data.

29. The method of claim 1, wherein using the computed distances to classify the gender of the hand as belonging to the male class or the female class comprises:

combining the distances into a classification score; and

classifying the gender of the hand by comparing i) the classification score to ii) a threshold computed from a set of training data.

30. The method of claim 1, wherein the overall score is a sum of the weighted classification scores.

31. The method of claim 1, wherein the digitally-imaged hand parts include five digitally-imaged finger parts and one digitally-imaged palm part.

15. A gender classification computing system comprising:

at least one processing unit;

at least one memory; and

computer-executable instructions stored in the at least one memory, the computer-executable instructions, when executed by the at least one processing unit to,

for each of at least two digitally-imaged hand parts, where each of the at least two digitally-imaged hand parts corresponds to one of a plurality of hand parts, compute a set of feature parameters representing a geometry of the digitally-imaged hand part;

using the sets of feature parameters for a set of two or more of the digitally-imaged hand parts, (a) combining sets of feature parameters for different hand parts to form sets of feature parameters for multiple hand parts, wherein a first set of feature parameters corresponds to a hand'"'"'s fingers and a second set of feature parameters corresponds to the hand'"'"'s palm and (b) compute distances of the set of digitally-imaged hand parts from each of i) a first eigenspace corresponding to a male class, and ii) a second eigenspace corresponding to a female class; and

use the computed distances to classify the gender of the hand as belonging to the male class or the female class comprising;

forming a distance vector from the computed distances;

comparing the distance vector to each of k-nearest neighbor distance vectors in a set of training data; and

classifying the gender of the hand as belonging to the class corresponding to a majority of the k-nearest neighbor distance vectors in the set of training data.

16. The gender classification computing system of claim 15, further comprising a camera, wherein the computer-executable instructions, when executed by the at least one processing unit, further cause the at least one processing unit to,

acquire a digital image of the hand using the camera;

and

segment the digital image of the hand into the digitally-imaged hand parts, the digitally-imaged hand parts including finger parts and a palm part.

17. The gender classification computing system of claim 15, further comprising a scanner, wherein the computer-executable instructions, when executed by the at least one processing unit, further cause the at least one processing unit to,

acquire a digital image of the hand using the scanner;

and

segment the digital image of the hand into the digitally-imaged hand parts, the digitally-imaged hand parts including finger parts and a palm part.

18. The gender classification computing system of claim 15, wherein the computer-executable instructions, when executed by the at least one processing unit, cause the at least one processing unit to use the computed distances to classify the gender of the hand as belonging to the class to which the set of digitally-imaged hand parts is closest.

19. The gender classification computing system of claim 15, wherein the computer-executable instructions, when executed by the at least one processing unit, cause the at least one processing unit to,

combine the distances into a classification score;

and

classify the gender of the hand by comparing i) the classification score to ii) a threshold computed from a set of training data.

20. A method of classifying the gender of a hand having a plurality of hand parts, the method comprising:

for each of at least two digitally-imaged hand parts, where each of the at least two digitally-imaged hand part corresponds to one of the plurality of hand parts, computing a set of feature parameters representing a geometry of the digitally-imaged hand part;

using the sets of feature parameters for a set of two or more of the digitally-imaged hand parts, (a) combining sets of feature parameters for different hand parts to form sets of feature parameters for multiple hand parts, wherein a first set of feature parameters corresponds to a hand'"'"'s fingers and a second set of feature parameters corresponds to the hand'"'"'s palm and (b) computing a distance of each of at least two sets of two or more digitally-imaged hand parts to each of (i) a first model eigenspace corresponding to first particular set of digitally-imaged hand parts of a male class, and (ii) a second model eigenspace corresponding to a second particular set of digitally-imaged hand parts of a female class;

using the computed distances to classify the gender of the hand as belonging to the male class or the female class comprises;

for each particular set of digitally-imaged hand parts, combining the distances of the particular set into a classification score;

in accordance with a weighting function, assigning weights to the classification scores;

combining the weighted classification scores into an overall score for the hand; and

classifying the gender of the hand by comparing i) the overall score for the hand to ii) a threshold computed from a set of training data.

21. The method of claim 20, further comprising:

acquiring a digital image of the hand; and

segmenting the digital image of the hand into the digitally-imaged hand parts, the digitally-imaged hand parts including at least one finger part and a palm part.

22. The method of claim 20, wherein computing the set of feature parameters representing the geometry of a particular digitally-imaged hand part comprises computing a set of contour-based feature parameters for the particular digitally-imaged hand part.

32. A gender classification computing system, comprising:

at least one processing unit;

at least one memory; and

computer-executable instructions stored in the at least one memory, the computer-executable instructions, when executed by the at least one processing unit to,

for each of at least two digitally-imaged hand parts, where each of the at least two digitally-imaged hand parts corresponds to one of a plurality of hand parts, compute a set of feature parameters representing a geometry of the digitally-imaged hand part;

using the set(s) of feature parameters for a set of two or more of the digitally-imaged hand parts, (a) combining sets of feature parameters for different hand parts to form sets of feature parameters for multiple hand parts, wherein a first set of feature parameters corresponds to a hand'"'"'s fingers and a second set of feature parameters corresponds to the hand'"'"'s palm and (b) computing a distance of each of at least two sets of two or more digitally-imaged hand parts to each of (i) a first model eigenspace corresponding to first particular set of digitally-imaged hand parts of a male class, and (ii) a second model eigenspace corresponding to a second particular set of digitally-imaged hand parts of a female class;

using the computed distances to classify the gender of the hand as belonging to the male class or the female class comprises;

for each particular set of digitally-imaged hand parts, combining the distances of the particular set into a classification score;

in accordance with a weighting function, assigning weights to the classification scores;

combining the weighted classification scores into an overall score for the hand; and

classifying the gender of the hand by comparing i) the overall score for the hand to ii) a threshold computed from a set of training data.

33. The gender classification computing system of claim 32, further comprising a camera, wherein the computer-executable instructions, when executed by the at least one processing unit, further cause the at least one processing unit to,

acquire a digital image of the hand using the camera;

and

segment the digital image of the hand into the digitally-imaged hand parts, the digitally-imaged hand parts including finger parts and a palm part.

34. The gender classification computing system of claim 32, further comprising a scanner, wherein the computer-executable instructions, when executed by the at least one processing unit, further cause the at least one processing unit to,

acquire a digital image of the hand using the scanner;

and

segment the digital image of the hand into the digitally-imaged hand parts, the digitally-imaged hand parts including finger parts and a palm part.

35. The gender classification computing system of claim 32, wherein the computer-executable instructions, when executed by the at least one processing unit, cause the at least one processing unit to use the computed distances to classify the gender of the hand as belonging to the class to which the set of digitally-imaged hand parts is closest.

36. The gender classification computing system of claim 32, wherein the computer-executable instructions, when executed by the at least one processing unit, cause the at least one processing unit to,

form a distance vector from the computed distances;

compare the distance vector to each of k-nearest neighbor distance vectors in a set of training data; and

classify the gender of the hand as belonging to the class corresponding to a majority of the k-nearest neighbor distance vectors in the set of training data.

37. The gender classification computing system of claim 32, wherein the computer-executable instructions, when executed by the at least one processing unit, cause the at least one processing unit to,

combine the distances into a classification score;

and

classify the gender of the hand by comparing i) the classification score to ii) a threshold computed from a set of training data.