Research On Face Feature Extraction Based On Subspace Analysis

In the rapid development of today's society,face recognition using computer for face analysis has attracted more and more attention.As the focus of research,face recognition is the key to face recognition.Among them,the feature extraction method based on subspace analysis has attracted extensive attention because of its simple algorithm and efficient recognition.The main contents of this paper include the following:(1)By reading the relevant literatures at home and abroad the status quo,in-depth analysis of principal component subspace analysis method,and the principle of linear discriminant analysis method,and expounds the kernel method and locality preserving projection method.Secondly,for the possible problems in one-dimensional methods,such as computational complexity and small sample problem,this paper discusses the analysis method based on 2-D,and introduces an improved algorithm based on 2DPCA,such as 2DPCA,left to right to 2DPCA compression compression and biaxial compression RL2DPCA.(2)Firstly,the kernel method is summarized,and then the characteristics of kernel principal component analysis and some kernel functions used in face recognition are expounded.For the nonlinear information in the high-dimensional feature of human face,this paper proposes a new method,fusion and principal component analysis,and minimum discriminant projection for facial feature extraction.Firstly,KPCA will use the original face data extractor geometry,obtain the projection matrix after drop dimension is reached after dimensionality reduction using MDP method in subspace,which makes the final projection matrix contains not only nonlinear information,but also contains the neighbor relationship between samples.Finally,experiments on face database verify the effectiveness of the proposed method.(3)A bidirectional compression based face recognition method combining two-dimensional principal component analysis(2DPCA)and two-dimensional locality preserving projection(2DLPP)is proposed.Because we focused on one dimension method of one-dimensional method may exist because of the small sample problem,we decided to use two-dimensional principal component analysis method,I adopt the method of bidirectional compression,the face images using 2DLPP to compress in the column direction,using 2DPCA method to compress the line direction.Finally,ORL face database is used to verify the advantages and effectiveness of the proposed algorithm in recognition rate.