Research On Subspace Based Face Recognition Algorithms

Face recognition is an important research field in image processing and pattern recognition. No one can find two human faces that exactly the same through out the world. Therefore, the research topic has a significant theoretical and practical value as the demand of highly efficient and effective Bio-Identification nowadays. After 40 years of research in pattern recognition, computer vision, neural network and physiology etc., researchers still can not be theoretically prove whether it is practical to invent a human-like auto-machine for face recognition, which makes the research topic much more challenging.After an abundant study and research of papers in this field, an in-depth research of subspace based face recognition algorithms are presented in this paper. Experiments show that the improved algorithms presented in this paper are of better performance and certain theoretical and practical value as well. The research work above mainly includes the following several aspects.First, research on the quality of kernel methods based on the study and implement PCA and KPCA for face recognition.Second, an improved LDA method based on regularized parameter is presented due to the SSS problem after an in-depth study the LDA and Direct LDA methods for face recognition. Further more, the kernel form of the improved LDA method has presented based on two kinds of kernel functions.Third, an improved Locality Preserving Projection is presented based on optimal linear embedding, further more, its kernel form and supervised form are presented based on theoretical deduction.Last, feasibility analysis of the algorithms presented in this paper has been addressed when applied to real world face recognition systems.