基于Contourlet变换的任意光照下的人脸识别

Face Recognition Under Varying Illumination Based on the Contourlet Transform

为了识别任意光照下的人脸图像,提出了一种多分辨率多方向性的人脸特征表示方法。利用Contourlet变换,提取人脸图像的低频系数特征和高频方向子带特征,并将高频方向子带特征作为识别依据而舍弃了对光照变换敏感的低频系数特征。在Yale B人脸库上的实验结果表明,提出的方法可以显著提高任意光照下的人脸识别系统的识别率。

This paper presents a directional multiresolution face representation for face recognition under varying lighting.In this work,the features of the low frequency and directional subbands in the contourlet transform are extracted respectively.The effect of illuminations is effectively reduced by discarding the low frequency components which are sensitive to illumination changes,and meanwhile the directional multiresolution face representation is generated.The PCA method is chosen to recognize the face representation.Experimental results on the Yale Face Database B demonstrate significant performance improvement of our method in the face recognition system under large illumination variations conditions.