Subspace KDA Algorithm for Non-linear Feature Extraction in Face Identification

Abstract

Kernel discriminant analysis (KDA) method is a promising approach for non-linear feature extraction in face identification tasks. However, as a linear algorithm to address nonlinear problem, Fisher discriminant analysis (FDA) approach will not give a satisfactory performance. Moreover, FDA usually suffers from small sample size (S3) problem. To overcome these two shortcomings in FDA method, Shannon wavelet kernel based subspace FDA (SKDA) algorithm is developed in this paper. Two public databases such as FERET and CMU PIE databases are selected for evaluation. Comparing with the existing kernel based FDA-based methods, the proposed method gives superior results.