System for recognizing disguised face using gabor feature and SVM classifier and method thereof

• US 8,913,798 B2
• Filed: 08/02/2012
• Issued: 12/16/2014
• Est. Priority Date: 12/21/2011
• Status: Active Grant

• Current Assignee: Electronics and Telecommunications Research Institute
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Kim, Kye Kyung, Lee, Jae Yeon, Yoon, Ho Sub, Kim, Jae Hong, Sohn, Joo Chan
• Primary Examiner(s): Bali, Vikkram
• Assistant Examiner(s): Woldemariam, Aklilu K.
• Application Number:
  US13/565,022
• Publication Number:
  US 20130163829A1
• Time in Patent Office: 866 Days
• Field of Search: 382/118, 382/159, 382/224, 382/264, 382/279
• US Class Current: 382/118
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06V 40/172 Classification, e.g. identification
• CPC Class Codes: G06V 40/40 Spoof detection, e.g. liveness detection

Abstract

Disclosed are a system and a method for recognizing a disguised face using a Gabor feature and a support vector machine (SVM) classifier according to the present invention.

The system for recognizing a disguised face includes: a graph generation means to generate a single standard face graph from a plurality of facial image samples; a support vector machine (SVM) learning means to determine an optimal classification plane for discriminating a disguised face from the plurality of facial image samples and disguised facial image samples; and a facial recognition means to determine whether an input facial image is disguised using the standard face graph and the optimal classification plane when the facial image to be recognized is input.

Claims

1. A system for recognizing a disguised face, the system comprising:

a graph generation means configured to generate a single standard face graph from a plurality of facial image samples;

a support vector machine (SVM) learning means configured to determine an optimal classification plane for discriminating a disguised face from the plurality of facial image samples and disguised facial image samples,

wherein the SVM learning means generates an initial face graph by adjusting a size of the single standard face graph based on position points of both eyes within a rectangular facial area,

and generates an optimal face graph by comparing a similarity between a Gabor feature value at each node of the initial face graph and a standard Gabor feature value at the each node of the single standard face graph and repeatedly modifying the initial face graph using a particle swarm optimization (PSO) algorithm based on a result of the comparison;

wherein the PSO algorithm is an evolutionary calculation used to obtain an optimal solution from a complex function by exchanging information with a personal particle and a particle within a swarm, using variable parameters such as a center between both eyes, a size scaling parameter of the entire single standard face graph, an upper size scaling parameter of the both eyes, or a lower size scaling parameter of the both eyes; and

wherein the personal particle is a first data point from the single standard face graph, and the particle within the swarm is a second data point from the swarm, and the swarm is a collection of data points;

a facial recognition means configured to determine whether an input facial image is disguised using the standard face graph and the optimal classification plane, when the facial image to be recognized is input.

2. The system of claim 1, wherein the graph generation means detects the rectangular facial area with a predetermined size from each of the input facial image samples using an Adaboost algorithm, normalizes each of the detected rectangular facial areas to a predetermined size, extracts a facial feature point from each of the normalized rectangular

facial areas, and generates the standard face graph using the extracted facial feature points.

3. The system of claim 2, wherein the size of the normalized rectangular facial area is set to be greater than or equal to a size of the detected rectangular facial area, and a detector detects the rectangular facial area while gradually reducing a size of the input facial image sample.

4. The system of claim 1, wherein the SVM learning means detects a rectangular facial area from each of the input facial image samples and the disguised facial image samples using an Adaboost algorithm, normalizes each of the detected rectangular facial areas to a predetermined size, generates the optimal face graph using the normalized rectangular facial areas, extracts the Gabor feature value from the generated optimal face graph, and determines the optimal classification plane for discriminating the disguised face using

the extracted Gabor feature value of the optimal face graph and a position value.

5. The system of claim 1, wherein the facial recognition means detects the rectangular facial area from the input facial image using an Adaboost algorithm, normalizes the detected rectangular facial area to a predetermined size, generates the optimal face graph using the normalized rectangular facial area, and determines whether the input facial image is disguised using the Gabor feature value and the position value that are extracted from the generated optimal face graph, and the generated optimal classification plane.

6. The system of claim 5, wherein the facial recognition means generates the optimal face graph using the normalized rectangular facial area, extracts the Gabor feature value and the position value from the generated optimal face graph, and provides the extracted Gabor feature value and position value of each of nodes of the optimal face graph as an input value of an SVM classifier and as a result, obtains an output value of the SVM classifier, and determines whether the input facial image is disguised using the output value and the generated optimal classification plane.

7. A system for recognizing a disguised face, the system comprising:

a detector configured to detect a rectangular facial area from an input facial image using an Adaboost algorithm;

a normalization unit configured to normalize the detected rectangular facial area to a predetermined size;

an extractor configured to generate an optimal face graph using the normalized rectangular facial area, and to extract a Gabor feature value and a position value from the generated optimal face graph,

wherein the generating an optimal face graph includes comparing a similarity between the Gabor feature value at each node of the initial face graph and a standard Gabor feature value at the each node of the standard face graph and repeatedly modifying the initial face graph using a particle swarm optimization (PSO) algorithm based on a result of the comparison, and

wherein the PSO algorithm is an evolutionary calculation used to obtain an optimal solution from a complex function by exchanging information with a personal particle and a particle within a swarm, using variable parameters such as a center between both eyes, a size scaling parameter of the entire graph, an upper size scaling parameter of the both eyes, or a lower size scaling parameter of the both eyes; and

wherein the personal particle is a first data point from the single standard face graph, and the particle within the swarm is a second data point from the swarm, and the swarm is a collection of data points; and

a determining unit configured to determine whether the input facial image is disguised using the extracted Gabor feature value and the position value, and a pre-generated optimal classification plane for recognizing a disguised face, wherein the determining unit determines the input facial image by adjusting a size of the standard face graph based on position points of the both eyes within a the rectangular facial area.

8. The system of claim 7, wherein the determining unit provides the extracted Gabor feature value and the position value of each of nodes of the optimal face graph as an input value of an SVM classifier and as a result, obtains an output value of the SVM classifier, and determines whether the input facial image is disguised using the output value and the optimal classification plane.

9. A method of recognizing a disguised face, the method comprising:

generating, using a processor, a single standard face graph from a plurality of facial image samples;

generating, using the processor, an optimal classification plane for recognizing the disguised face from the plurality of facial image samples and disguised facial image samples, wherein the generating of the optimal classification plane generates an initial face graph by adjusting a size of the single standard face graph based on position points of both eyes within a rectangular facial area and generates an optimal face graph by comparing a similarity between a Gabor feature value at each node of the initial face graph and a standard Gabor feature value at the each node of the standard face graph and repeatedly modifying the initial face graph using a particle swarm optimization (PSO) algorithm based on a result of the comparison, and

wherein the PSO algorithm is an evolutionary calculation used to obtain an optimal solution from a complex function by exchanging information with a personal particle and a particle within a swarm, using variable parameters such as a center between both eyes, a size scaling parameter of the entire graph, an upper size scaling parameter of the both eyes, or a lower size scaling parameter of the both eyes; and

wherein the personal particle is a first data point from the single standard face graph, and the particle within the swarm is a second data point from the swarm, and the swarm is a collection of data points;

determining, using the processor, whether an input facial image is disguised using the standard face graph and the optimal classification plane, when the facial image to be recognized is input.

10. The method of claim 9, wherein the generating of the standard face graph detects the rectangular facial area with a predetermined size from each of the input facial image samples using an Adaboost algorithm, normalizes each of the detected rectangular facial areas to a predetermined size, extracts a facial feature point from each of the normalized rectangular facial areas, and generates the standard face graph using the extracted facial feature points.

11. The method of claim 10, wherein the size of the normalized rectangular facial area is set to be greater than or equal to a size of the detected rectangular facial area.

12. The method of claim 9, wherein the generating of the optimal classification plane detects a rectangular facial area from each of the input facial image samples and the disguised facial image samples using an Adaboost algorithm, normalizes each of the detected rectangular facial areas to a predetermined size, generates an optimal face graph using the normalized rectangular facial areas, extracts a Gabor feature value from the generated optimal face graph, and generates the optimal classification plane for discriminating the disguised face using the extracted Gabor feature value of the optimal face graph and a position value.

13. The method of claim 9, wherein the determining of whether the input facial image is disguised detects a rectangular facial area from the input facial image using an Adaboost algorithm, normalizes the detected rectangular facial area to a predetermined size, generates an optimal face graph using the normalized rectangular facial area, and determines whether the input facial image is disguised using a Gabor feature value and a position value that are extracted from the generated optimal face graph, and the generated optimal classification plane.

14. The method of claim 13, wherein the determining of whether the input facial image is disguised generates the optimal face graph using the normalized rectangular facial area, extracts the Gabor feature value and the position value from the generated optimal face graph, and provides the extracted Gabor feature value and position value of each of nodes of the optimal face graph as an input value of an SVM classifier and as a result, obtains an output value of the SVM classifier, and determines whether the input facial image is disguised using the output value and the generated optimal classification plane.

15. A method of recognizing a disguised face, the method comprising:

detecting a rectangular facial area from an input facial image using an Adaboost algorithm;

normalizing the detected rectangular facial area to a predetermined size;

generating an optimal face graph using the normalized rectangular facial area, and extracting a Gabor feature value and a position value from the generated optimal face graph; and

determining whether the input facial image is disguised using the extracted Gabor feature value and the position value, and a pre-generated optimal classification plane for recognizing the disguised face, wherein generating includes generating an initial face graph by adjusting a size of a standard face graph based on position points of both eyes within the rectangular facial area, and generates an the optimal face graph by comparing a similarity between the Gabor feature value at each node of the initial face graph and a standard Gabor feature value at the each node of the standard face graph and repeatedly modifying the initial face graph using a particle swarm optimization (PSO) algorithm based on a result of the comparison,

wherein the PSO algorithm is an evolutionary calculation used to obtain an optimal solution from a complex function by exchanging information with a personal particle and a particle within a swarm, using variable parameters such as a center between both eyes, a size scaling parameter of the entire graph, an upper size scaling parameter of the both eyes, or a lower size scaling parameter of the both eyes and

wherein the personal particle is a first data point from the single standard face graph, and the particle within the swarm is a second data point from the swarm, and the swarm is a collection of data points.

16. The method of claim 15, wherein the determining of whether the input facial image is disguised provides the extracted Gabor feature value and position value of each of nodes of the optimal face graph as an input value of an SVM classifier and as a result, obtains an output value of the SVM classifier, and determines whether the input facial image is disguised using the output value and the optimal classification plane.