Discriminative analysis based nonlinear subspace approach for expression recognition

Abstract

Most of the earlier expression recognition system based on nonlinear subspace methods not able to solve the discriminative problems of feature extraction, locality structure preservation and dimensional reduction by increasing the Fishers ratio of discriminant analysis. In this work, adaptive combination approach is framed by combining geometrical and holistic features. Both Gabor magnitude feature vector (GMFY) and enhanced Gabor phase feature vector (GPFV) are separately isolated and feature level fusion is carried out by combining with geometrical distance feature vector (GDFY). Fused phase part was aligned with discrete wavelet moment (DWT) features. High dimensional space was projected into low dimensional subspace by kernel locality preserving Fisher discriminant analysis method. Projected subspace is normalized and final scores of projected space were fused using maximum fusion rule. Expressions are classified using Euclidean distance matching and support vector machine radial basis function kernel classifier. The whole proposed approach is abbreviated as ACEGKLPFDA. An experimental result reveals that the proposed approach is effective for dimension reduction, efficient recognition and classification. Performance of proposed approach is measured in comparison with related subspace approaches. The best average recognition rate achieves 97.61% for JAFFE and 95.62% FD database respectively.