A Novel Nonlinear Feature Extraction and Recognition Approach Based on Improved 2D Fisherface Plus Kernel Discriminant Analysis

Abstract

A novel nonlinear feature extraction and recognition approach which is based on improved 2D Fisherface plus Kernel discriminant analysis is proposed. We provide an improved 2D Fisherface method that designs a new strategy to select appropriate 2D principal components and discriminant vectors, then we use 2D features to perform the Kernel discriminant analysis. The nearest neighbor classifier with cosine distance measure is adopted in classifying the nonlinear discriminant features. The experiments show that the proposed approach achieves better recognition results than several representative discriminant methods.