Research Of Multi-pose Face Recognition Based On SIFT

Automatic face recognition technology is a multi-subject cross subject,which involves machine vision,pattern recognition,image processing,expert system and other fields.With the rapid development of various fields,face recognition technology has been widely used in exam monitoring,criminal investigation,case control,access control,certificate verification,video surveillance,intelligent transportation and so on.At present,the research of face recognition algorithm has extended from ideal state to multi-pose face recognition in complex and non-ideal condition.Age,occlusion,illumination and attitude change are the main factors that affect the recognition performance of the present face recognition system.Therefore,the characteristics of SIFT(Scaled Invariant Feature Transform)with high unique characteristics have entered the vision of face recognition researchers.The SIFT descriptor has the scale,rotation and partial affine invariance.This paper studies how to apply the unique advantage of SIFT descriptor to multi-pose face recognition system.First,the SIFT algorithm is used to extract the feature points of face images,and use the phase consistency information to eliminate a large number of unstable edge points,reducing the number of feature points,and improve the recognition efficiency.Secondly,in view of the multi-pose changes of the face,multi view fusion is used to recognize the multi-pose face.Each class selects face images with a horizontal interval of 45 degrees as a sample set,extracts SIFT features and performs intra class feature fusion,and then using sparse representation to perform intra class feature selection,effectively filtering similar or identical features and reducing data redundancy.Finally,on the premise of testing the best parameters of the key parameters,we simulate the multi-pose face recognition method in this paper.The results illustrate that each improvement is effective in improving the recognition effect and robustness.In the case of uncertain face attitude,this paper obtains better recognition accuracy and does not need face markup and face alignment.