Bayesian Face Recognition Approach Based on Feature Fusion

Abstract

Feature extraction and matching recognition are two critical stages in face recognition process. While traditional Bayesian classifier exists the small sample problem in the matching recognition stage, a novel Bayesian face recognition approach based on feature fusion is proposed. In the feature extraction stage, the global non-linear feature is extracted by kernel principal component analysis (KPCA), and the local manifold structure information is extracted by the orthogonal locality sensitive discriminant analysis (OLSDA), achieving the purpose of extracting the low-dimension essential facial feature with high-discrimination, and the constraints of the fusion features make the obtained matrix be closer to the desired solution. In the matching recognition stage, a maximum entropy covariance selection (MECS) method is utilized to solve the small sample problem. Extensive experimental results on several datasets show that these two stages can significantly improve the accuracy of face recognition.