Local Feature-based Face Recognition Research

Local features are inspired from biology.Biological visual system is naturally based on local features,and the visual receptive field is a processor of natural local feature.The well-known Marr's graph theory is to the bottom of the image coming down to dealing with the extraction of local features.Non-negative matrix factorization (NMF)is a new matrix decomposition method based on the part of the study,which has a reflection of human thinking'partial constitute the overall'concept.Non-negative matrix factorization(NMF)based on statistical learning applied to image recognition can be more effective.For single-sample case,with the recent emergence of scale-invariant,rotation-invariant and affine features of the same characteristics of the local descriptors-SIFT.SIFT algorithm to be applied to human face image recognition can be more stable and partial characterization and will be helpful to improve recognition rate.In this paper,based on the non-negative matrix factorization(NMF),wavelet transform(WT)and support vector machine(SVM)for face recognition,as well as key areas of the SIFT-based feature matching method for face recognition are proposed. Based on NMF,WT and SVM face recognition method,first of all through the two-dimensional discrete wavelet processing to remove high-frequency part of the image thereby reducing the computational complexity and filter out the emotional face change and glasses,hair and other unnecessary details of the impact of identification. Then,the sparse non-negative matrix factorization method has been used to extract features after the pre-processing of these images.Combination of support vector machine(SVM)is applied to build image feature classifier,and thus the formation of a complete face recognition system has a success.Based on SIFT feature matching of the key areas of the face recognition methods,to locate and extract the human eye and mouth,and then extracted from these key areas for SIFT feature matching feature points, and then complete the task of face recognition.In order to verify the validity made by WT+NMF+SVM algorithm performance, the classical PCA-based face recognition methods,as well as the K-neighbors based method for face recognition methods are included to take a comparison.Experiments are acted on the ORL face database,and by identifying those who under normal circumstances face samples,to study these three methods in the local feature extraction or feature extraction on the overall performance.The experimental results not only confirmed the view put forward by this paper also shows that face recognition methods based on the local feature extraction with robustness.Similarly,in the ORL face database of SIFT feature matching methods overall,K-means methods and key areas to match the characteristics of the three matching methods of face recognition analysis and comparison,the ultimate validation key areas based on SIFT feature matching methods of efficiency and stability.