A Research On Kernel-based Face Recognition Applications

Face recognition continues to be a hot topic in pattern recognition field due to its wide range of applications such as commercial and law enforcement applications.A critical issue to a successful approach for face recognition is how to extract discriminant feature from the facial images.Many feature extraction methods have been proposed and among them the linear subspace analysis has received extensive attention owing to its simpleness and efficiency. But there are two problems in linear face recognition : The first one is that the distribution of face images with different pose,illumination and face expression is complex and nonlinear.The second one is the small sample size (S3) problem.This problem occurs when the number of training samples is smaller than the dimensionality of feature vector,which results in a sigular within-class scatter matrix. For the former, kernel technique can be used to extract nonlinear feature,and for the latter,there are two kinds of solutions:algorithm based methods and transform based ones.The former ones exploit new algorithm aiming at S3 and the latter ones apply dimension reduction before discriminant analysis to eliminate singularity of the within-class scatter matrix.The main work of the thesis is summarized as follows:1 Reorganized and summarized the domestic and foreign academic circles about the research results of the kernel methods in statistical learning theory,deeply researched on the basic theory of the kernel and applied it to face recognition.2 Proposed a algorithm based method to solve S3, namly Kernel machine-based One-parameter Multiple Discriminant Analysis (K1PMDA) .This approach avoid calculate inverse within-class scatter matrix by building a new optimization criteria and introducing a disturbing parameter.3 Proposed a transform based method to solve S3,namly Kernel Inverse Orthogonalized Fisher Discriminant (KIOFD).This approach inversed the Fisher optimization criteria,applied a dimension reduction method to extract important components for linear discriminant analysis and combined orthogonalized technique.