Font Size: a A A

Research Of Face Recognition Based On The Fusion Of Wavelet Transform And Improved Subspace Algorithm

Posted on:2018-10-30Degree:MasterType:Thesis
Country:ChinaCandidate:J P ChuFull Text:PDF
GTID:2348330536969328Subject:Computational Mathematics
Abstract/Summary:PDF Full Text Request
In recent decades,with the rapid development of technology,more and more situations need to be identification fast and effectively.Face recognition technology becomes a hot spot in today's pattern recognition quickly by good performances,and has been applied to all walks of life successfully.The principal component analysis and linear discriminant analysis proposed by Matthew Turk et al and Belhumeur,Peter N et al respectively are great revolutions in the history of face recognition technology,they are data processing methods that turn the face image matrices into vectors,despite the effect of recognition is significant,the curse of dimensionality and small sample problems occur frequently.Based on the internal structure of the face,Two-dimensional principal component analysis was proposed by Yang J and achieved remarkable effect.While the algorithm treat each sample without considering the differences between them and the dimension of the feature is still high.Wavelet transform can extract features from time domain and frequency domain simultaneously with multi-resolution analysis,good compression ratio and adaptive characteristics.More important,it is good at data dimensionality reduction.The majority of scholars have devoted to researching it,and applied it to face recognition technology successfully.Most of people abandoned high-frequency and only extracted features from low-frequency.However,as we know,there are some important details of face features information in high-frequency.Therefore,the two algorithms which are proposed in this paper extract the features of high-frequency and low-frequency components.In order to solve the problems above,this paper describes the wavelet transform,PCA and LDA et al relational theories firstly,then weights wavelet and fuses improved2DPCAand LDA respectively.The main contents are as follows:(1)Weighted fusion of low-frequency and high-frequency of wavelet transform.Meanwhile,considering the feature dimension of2 DPCAand the difference between samples,2DPCAis improved according to the attributes of the sample itself,and we extracted features by it twice.Thus the features of the face are not only fully extracted and the dimension of feature becomes lower than before.The Experiments show that the performance of the new algorithm is better than 2DPCA,WT+2DPCA,2DDPCA.(2)The features of low-frequency are extracted by the improved LDA,the high-frequency are fused by the iterative method,then the features of high-frequency are extracted by the improved LDA,two features are fused at last.The optimize method is used to fuse the high-frequency,not subject to human factors.And the improved LDA solves the edge problems and that the Fisher criterion is useless when the within-class divergence matrix is singular.The experiments show that t the performance of the new algorithm is better than PCA,LDA,WT+PCA,W+LD,TWSBT LDA F+A.
Keywords/Search Tags:Face recognition, Wavelet transform, Recognition rate, Two-dimensional principal component analysis, linear discriminant analysis
PDF Full Text Request
Related items