采用二代曲波变换和反向传播神经网络的人脸识别方法  

Method for Face Recognition Using Second-Generation Curvelet Transform and Back Propagation Neural Network

作　　者：许学斌[1] 张德运[1] 张新曼[1] 潘煜[1] 

机构地区：[1]西安交通大学电子与信息工程学院,西安710049

出　　处：《西安交通大学学报》2008年第10期1213-1216,1284,共5页Journal of Xi'an Jiaotong University

基　　金：国家高技术研究发展计划资助项目(2005AA121130);国家自然科学基金资助项目(60602025)

摘　　要：针对小波变换在人脸识别中存在识别正确率较低的问题,提出了一种基于二代曲波变换的人脸识别方法.首先将所有样本图像和测试图像通过基于"打包"的快速离散曲波变换进行分解,获得不同尺度、不同角度的曲波变换系数,再利用曲波变换分解系数中包含了人脸重要特征信息的低频系数,作为特征参数送入反向传播(BP)神经网络中进行学习训练,最后将训练好的BP神经网络用于人脸识别.经剑桥大学ORL人脸库的图像识别实验表明,所提方法的识别正确率达到95%,比Daub(2)小波基的小波变换方法的识别正确率提高了2.5%.To improve the recognition rate of the wavelet-based methods for face recognition, a multiscale face recognition method based on second-generation eurvelet transform is proposed. All face images are decomposed by using digital curvelet transform via wrapping. Curvelet coefficients of low frequency and high frequency in different scales and of various angles are obtained. Most significant information of faces is contained in the low frequency coefficients which are important for face recognition. Then, the low frequency coefficients are applied as study samples to the BP neural network. Finally, low frequency coefficients of some test face images are used to simulate the neural network to get the face recognition results. The experiments that are performed on the Cambridge university ORL database show that the proposed method has better performance than wavelet-based method, and that the recognition rate is improved to 95% （with 2.5 % improvement）.

关 键 词：曲波变换 人脸识别 神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]