基于轮廓波变换和核Fisher判别分析的人脸识别

摘要

Contourlet变换是一种新的多尺度几何分析方法,它不仅具有小波变换的多分辨率特性和时频局域特性,还具有很强的方向性和各向异性.提出基于Contourlet变换和核Fisher判别分析的人脸识别方法,研究了Contourlet变换的低频系数、各层高频系数与核Fisher判别分析相结合进行人脸识别的识别率和识别时间.实验表明,Contourlet变换的低频系数与核Fisher判别分析相结合,有优异的识别率,也减少了识别时间;高频成分有一定的识别性能,但识别率较低.将低频成分与高频方向子带相结合能获得最优的识别率.%Contourlet transform is a new multiscale geometrical analysis scheme possessing main features of the wavelets (namely,multiresolution and time-frequency localisation) as well as high directionality and anisotropy. In this paper, a method based on contourlet transform and kernel Fisher discriminant analysis (KFDA) for face recognition is proposed. The recognition rate and recognition time of face recognition are studied in regard to the combination of low frequency coefficient and high-frequency coefficients at each level of contourlet transform with KFDA. Experimental result show that to combine low frequency coefficient of contourlet transform with KFDA achieves higher recognition rate with decreased recognition time; while the high-frequency components do help the recognition to a certain extent but its recognition rate is low. The combination of low frequency component with high frequency directional subband can optimise the recognition rate.