Subspace-based Face Recognition Algorithm

Currently, face recognition algorithm based on subspace has attract-ted widespread attention for its high efficiency and strong character-rization advantages. However, because of face images are more vulnera-ble to the interference of various external factors, Many researchers have been tirelessly explored that how to use face image information more effectively and how to obtain a higher recognition rate.Having Analyzed and studied a lot abroad and home academic lite-rature on face recognition in recent years, this paper deeply studied the method of subspace-based feature extraction and the classification meth-od of nearest neighbor classifier-based. The research work is as follows:Firstly, for the heavy computational complexity of the human face image, this paper proposed an improved kernel principal component analysis algorithm (KPCA).The algorithm uses the method of best sam-ple mean estimate vector to instead of the original sample vector, redu-cing the amount of data calculation and then improves the algo-rithm computation rate.Secondly,for the small samples and the edge information interferen-ce problems,this paper proposed an improved linear discriminant analysis algorithm(LDA).The algorithm uses the method of between class scatter matrix feature weight and inner class scatter matrix mean center treat,enhancing the algorithm ability of dealling with eigenvalue error and edge classification. Then the improved KPCA and improved LDA algori-thm is combined and applied to the face recognition feature extraction.Thirdly, for the curse of dimensionality and noise issues in the sam-ple characteristics, this paper proposed a K neighbors (KNN) classificati-on algorithm basing on similarity auxiliary of feature weighted.This al-gorithm reduces the dimension of sample data, effectively suppresses the interference of the noise data.Fourthly, this paper describes the face recognition algorithm comp-rehensively, and analysis the advantages of the improved face recogniti- on algorithm rate by large number of experimental data.This paper studies the improved feature extraction algorithm and classification algorithm by MATLAB software.The simulation results show that:this paper presents an improved performance better in face re-cognition algorithm.