SIGNAL PROCESSING

• US 20170364740A1
• Filed: 06/17/2016
• Published: 12/21/2017
• Est. Priority Date: 06/17/2016
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Aas, Cecilia J.
• Granted Patent: US 9,928,408 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06V 10/513 Sparse representations
• CPC Class Codes: G06V 10/761 Proximity, similarity or dissimilarity measures
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

A computer-implemented method is provided for classifying an input signal against a set of pre-classified signals. A computer system may calculate, for each of one or more signals of the set of pre-classified signals, a parallelism value indicating a level of the parallelism between that signal and the input signal. The computer system may calculate, for a first subset of the set of pre-classified signals, a sparse vector, wherein each element of the sparse vector serves as a coefficient for a corresponding signal of the first subset. The computer system may determine, for each of the signals in the set of pre-classified signals, a similarity value indicating a level of similarity between that signal and the input signal.

Claims

1. A computer-implemented method for performing facial recognition, the method comprising:

calculating, for one or more signals of a set of pre-classified signals, a parallelism value indicating a level of parallelism between that signal and an input signal, wherein each signal in the set of pre-classified signals is an image of a known person, and wherein the input signal is an image of a person to be identified;

calculating, for a first subset of the set of pre-classified signals, a sparse vector, wherein each element of the sparse vector serves as a coefficient for a corresponding signal of the first subset;

determining, for each of the signals in the set of pre-classified signals, a similarity value indicating a level of similarity between that signal and the input signal,

wherein the determination of the similarity value for at least one of the signals in the first subset is based, at least in part, on the value of the corresponding element of the sparse vector, and

wherein the determination of the similarity value for at least one of the one or more signals is based, at least in part, on the calculated parallelism value for that signal;

determining which signals in the first subset of signals correspond to elements in the sparse vector that have a positive value;

identifying a second subset of signals, the second subset of signals including the signals that correspond to elements of the sparse vector having a positive value,

wherein the one or more signals for which parallelism values are calculated comprise all of the signals in the set of pre-classified signals that are not in the second subset, and

wherein the determination of the similarity value for each of the signals that are not in the second subset is based, at least in part, on the calculated parallelism value for that signal; and

identifying the known person that corresponds to the input signal.

2. The method of claim 1, wherein the one or more signals for which parallelism values are calculated comprises all of the signals in the set of pre-classified signals.

3. The method of claim 2, further comprising:

identifying the first subset of signals from the set of pre-classified signals by selecting signals which are most parallel to the input signal according to the calculated parallelism values.

4. The method of claim 3, wherein the first subset is selected such that a size of the first subset is a predetermined proportion of the set of pre-classified signals.

6. The method of claim 1, wherein the similarity value for each of the pre-classified signals in the second subset is determined based solely on the value of the corresponding element of the sparse vector.

7. The method of claim 1, further comprising:

identifying a minimum positive value of the elements of the sparse vector; and

scaling the parallelism values for each of the pre-classified signals that are not in the second subset into a range between zero and the minimum positive value,

wherein the similarity value for each of the signals in the set of pre-classified signals that are not in the second subset is determined based, at least in part, on the scaled parallelism value for that signal.

8. The method of claim 7, wherein the similarity value for each of the pre-classified signals that are not in the second subset is determined based solely on the scaled parallelism value for that signal.

9. The method of claim 1, wherein the level of parallelism for each of the one or more signals is based on a dot product between a vector representation of the input signal and a vector representation of that signal.

5. (canceled)

10. (canceled)

11. (canceled)

12. A system for performing facial recognition, the system comprising a processor configured to:

calculate, for each of one or more signals of a set of pre-classified signals, a parallelism value indicating a measure of parallelism between that signal and an input signal, wherein each signal in the set of pre-classified signals is an image of a known person, and wherein the input signal is an image of a person to be identified;

calculate, for a first subset of the set of pre-classified signals, a sparse vector for use in a linear combination with the first subset of signals to provide a sparse approximation of the input signal, wherein each element of the sparse vector serves as a coefficient for a corresponding signal of the first subset in the linear combination;

determine, for each of the signals in the set of pre-classified signals, a similarity value indicating a measure of similarity between that signal and the input signal,

wherein the determination of the similarity value for at least one of the signals in the first subset is based, at least in part, on the value of the corresponding element of the sparse vector, and

wherein the determination of the similarity value for at least one of the one or more signals is based, at least in part, on the calculated parallelism value for that signal;

determine which signals in the first subset of signals correspond to elements in the sparse vector that have a positive value;

identify a second subset of signals, the second subset of signals including the signals that correspond to elements of the sparse vector having a positive value,

wherein the one or more signals for which parallelism values are calculated comprise all of the signals in the set of pre-classified signals that are not in the second subset, and

wherein the determination of the similarity value for each of the signals that are not in the second subset is based, at least in part, on the calculated parallelism value for that signal; and

identify the known person that corresponds to the input signal.

13. The system of claim 12, wherein the one or more signals for which the processor is configured to calculate parallelism values comprises all of the signals in the set of pre-classified signals, and wherein the processor is further configured to:

identify the first subset of signals by selecting, from the set of pre-classified signals and using the calculated parallelism values, a predetermined proportion of signals which are most parallel to the input signal.

15. The system of claim 12, wherein the processor is further configured to:

identify a minimum positive value of the elements of the sparse vector; and

scale the parallelism values for each of the pre-classified signals that are not in the second subset into a range between zero and the minimum positive value,

wherein, the similarity value for each of the signals in the set of pre-classified signals that are not in the second subset is determined based, at least in part, on the scaled parallelism value for that signal.

14. (canceled)

16. (canceled)

17. (canceled)

18. A computer program product for performing facial recognition, the computer program product comprising a computer readable storage medium having stored thereon:

first program instructions executable by a processor to cause the processor to calculate, for each of one or more signals of a set of pre-classified signals, a parallelism value indicating a measure of parallelism between that signal and an input signal, wherein each signal in the set of pre-classified signals is an image of a known person, and wherein the input signal is an image of a person to be identified;

second program instructions executable by the processor to calculate, for a first subset of the set of pre-classified signals, a sparse vector for use in a linear combination with the first subset of signals to provide a sparse approximation of the input signal, wherein each element of the sparse vector serves as a coefficient for a corresponding signal of the first subset in the linear combination;

third program instructions executable by the processor to cause the processor to determine, for each of the signals in the set of pre-classified signals, a similarity value indicating a measure of similarity between that signal and the input signal,

wherein the determination of the similarity value for at least one of the signals in the first subset is based, at least in part, on the value of the corresponding element of the sparse vector, and

wherein the determination of the similarity value for at least one of the one or more signals is based, at least in part, on the calculated parallelism value for that signal;

fifth program instructions executable by the processor to cause the processor to identify a second subset of signals from the first subset by identifying those signals which correspond to elements of the sparse vector having a positive value,

wherein the one or more signals which the first program instructions cause the processor to calculate parallelism values for comprise all of the signals in the set of pre-classified signals that are not in the second subset, and

wherein the third program instructions cause the processor to determine the similarity value for each of the signals that are not in the second subset is based, at least in part, on the calculated parallelism value for that signal;

sixth program instructions executable by the processor to cause the processor to identify a minimum positive value of the elements of the sparse vector;

seventh program instructions executable by the processor to cause the processor to scale the parallelism values for each of the pre-classified signals that are not in the second subset into a range between zero and the minimum positive value,

wherein, the third program instructions cause the processor to determine the similarity value for each of the signals in the set of pre-classified signals that are not in the second subset based, at least in part, on the scaled parallelism value for that signal; and

eighth program instructions executable by the processor to cause the processor to identify the known person that corresponds to the input signal.

19. The computer program product of claim 18, wherein the one or more signals for which the first program instructions cause the processor to calculate parallelism values comprise all of the signals in the set of pre-classified signals, and wherein the computer program product further comprises:

ninth program instructions executable by the processor to cause the processor to identify the first subset of signals by selecting, from the set of pre-classified signals, a predetermined proportion of signals which are most parallel to the input signal according to the calculated parallelism values.

20. (canceled)