Face recognition using evolutionary algorithms

• US 20040001615A1
• Filed: 06/27/2002
• Published: 01/01/2004
• Est. Priority Date: 06/27/2002
• Status: Active Grant

• Current Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Original Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Inventors: Trajkovic, Miroslav, Gutta, Srinivas, Philomin, Vasanth
• Granted Patent: US 7,139,738 B2
• Time in Patent Office: Days
• US Class Current: 382/118
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

A system and method wherein the parameters used for face recognition are determined via an evolutionary algorithm, such as a genetic algorithm. Candidate sets of parameters are evaluated for face recognition performance in a given environment. The most effective sets of parameters are used to generate offspring sets of parameters, and these offspring sets are subsequently evaluated for face recognition performance. This process is repeated, with each generation of offspring sets inheriting attributes of the most successful sets, until the system converges on a set of parameters that exhibits superior performance for the given environment. In like manner, other sets of parameters are determined for other environments. When the face recognition system is deployed, the appropriate set of parameters is used, dependent upon the environment in which the target images originate. In this manner, the system exhibits superior performance in each of the environments in which it is deployed.

Claims

1. A face recognition system, comprising:

a face recognizer that is configured to identify a face in a target image, based on a set of characterizing parameters, and

a parameter set selector, operably coupled to the face recognizer, that is configured to select the set of characterizing parameters from among a plurality of sets of characterizing parameters, for use by the face recognizer, based on an identifier of an environment from which the target image is obtained.

2. The face recognition system of claim 1, wherein

the plurality of sets of characterizing parameters are determined via an evolutionary algorithm applied to a recognition of test target images in a plurality of environments, and

the environment from which the target image is obtained corresponds to one of the plurality of environments.

3. The face recognition system of claim 1, wherein

the identifier of the environment corresponds to an identifier of a camera of a plurality of cameras.

4. The face recognition system of claim 1, wherein

the identifier of the environment includes at least one of:

a time of day, and

a measure of illumination.

5. The face recognition system of claim 1, wherein

the face in the target image is identified as corresponding to one of a plurality of reference images.

6. A system for determining a preferred set of characterization parameters for face recognition, comprising:

a face recognizer that is configured to provide an identifier of a face in a target image, based on a set of characterization parameters,

an evaluator, operably coupled to the face recognizer, that is configured to determine a measure of effectiveness of the set of characterization parameters, based on an accuracy of the identifier, and

an evolutionary algorithm, operably coupled to the evaluator, that is configured to provide alternative sets of characterization parameters to the face recognizer for subsequent determinations of measures of effectiveness for these alternative sets of characterization parameters by the evaluator,

wherein;

the alternative sets of characterization parameters are based on measures of effectiveness of prior sets of characterization parameters, and

the evolutionary algorithm is further configured to identify the preferred set of characterization parameters, based on the measures of effectiveness of each set of characterization parameters of the alternative sets and prior sets of characterization parameters.

7. The system of claim 6, wherein

each set of characterization parameters includes a subset of characterization parameters of a plurality of available characterization parameters.

8. The system of claim 7, wherein

the measure of effectiveness is further based on a size of the subset of characterization parameters.

9. The system of claim 6, wherein

the evolutionary algorithm is configured to provide the alternative sets of characterization parameters by:

determining a fitness of each prior set of the prior sets of characterization parameters, based at least in part on the measures of effectiveness of each prior set,

identifying high-performing sets of the prior sets of characterization parameters, based on the fitness of each prior set,

generating at least one set of the alternative sets of characterization parameters by combining attributes of two or more high-performing sets.

10. The system of claim 9, wherein

the evolutionary algorithm is further configured to

generate at least one other set of the alternative sets of characterization parameters via a random process.

11. The system of claim 6, wherein

the accuracy of the identifier is based on an identification of one of a plurality of reference images.

12. The system of claim 11, wherein

the measure of effectiveness is increased by a first amount for a correct identification of the one of the plurality of reference images corresponding to the target image, and decreased by a second amount for an incorrect identification, the second amount being larger than the first amount.

13. A method of determining a preferred set of characterization parameters for face recognition, comprising:

generating a first population of sets of characterization parameters,

evaluating each member set of the first population of sets, to provide a measure of effectiveness of each member set of the first population of sets, based on an accuracy of identifying each of a plurality of test target images,

generating a second population of sets of characterization parameters, based on the measure of effectiveness of each member set of the first population of sets,

evaluating each member set of the second population of sets, to provide a measure of effectiveness of each member set of the second population of sets, based on a subsequent accuracy of identifying each of the plurality of target images, and

determining the preferred set of characterization parameters, based at least in part on the measure of effectiveness of each member set of the first and second populations of sets.

14. The method of claim 13, wherein

generating the second population of sets includes:

identifying high-performing member sets of the first population of sets, based at least in part on the measure of effectiveness of each member set of the first population of sets,

generating at least one member set of the second population of sets by combining attributes of two or more high-performing member sets of the first population of sets.

15. The method of claim 13, further including

obtaining a plurality of pluralities of test images, each plurality of test images corresponding to each environment of a plurality of environments,

generating and evaluating each member set as detailed in claim 13 for each plurality of test images to determine a preferred set of characterization parameters for each environment.

16. The method of claim 13, wherein

the accuracy of identifying each of the plurality of target images is based on an identification of one of a plurality of reference images as corresponding to the target image.

17. The method of claim 16, further including

obtaining a plurality of pluralities of reference images, each plurality of reference images corresponding to each environment of a plurality of environments,

generating and evaluating each member set as detailed in claim 13 for each plurality of reference images, to determine a preferred set of characterization parameters for each environment.

18. The method of claim 13, wherein

each member set of the first and second populations includes a subset of characterization parameters of a plurality of available characterization parameters.