第18卷第4期 2010年4月
1004-924X(2010)04-1011-10
文章编号
光学精密工程
Optics and Precision Engineering
Vol.18No.4
Apr.2010
利用改进的Retinex进行人脸图像光照处理
葛微1,2，李桂菊1，程宇奇1,2，薛陈1,2，朱明1
（1.中国科学院长春光学精密机械与物理研究所，吉林长春130033；
2.中国科学院研究生院，北京100039）
摘要：为了提高光照变化条件下的人脸识别率，针对Retinex算法处理人脸侧光照图像时会误增强阴影的边缘提出了一种基于新的传导函数的自适应平滑Retinex算法。该传导函数用空间梯度和像素的局部不一致性两种方式共同测量灰度的剧烈变化，在平滑图像的同时没有边缘增强效应，且不会损失人脸特征边缘。在平滑估计的送代过程中，选取上一次与此次选代结果中的较大值作为约束条件来保证估计出的亮度图像能满足Retinex理论的约束条件。在YaleB人脸库上的实验结果表明，本文算法能有效克服强侧光照时的阴影现象且没有损失人脸特征边缘；与较经典Retinex算法相比，侧光照时的识别率在最好情况时提高了24.2%，无强侧光照时也可提高4%左右，具有光照鲁棒性，可适用于任何光照条件下的人脸识别。
键词：人脸识别；Retinex算法；自适应平滑；传导函数；侧光照；明影关
中图分类号：TP301.4;TP391.41
文献标识码：A
FaceimageilluminationprocessingbasedonimprovedRetinex
GE Weil.2,LI Gui-ju',CHENG Yu-qi-2,XUE Chen'-,ZHU Ming(1.ChangchunInstituteof Optics,FineMechanicsand Physics
ChineseAcademyofSciences,Changchun130033,China;
2.GraduateUniversity of ChineseAcademyof Sciences,Beijing100039,China)
Abstract: Illumination variation is one of the most significant factors affecting the performance of face recognition systems, As the adaptive smoothing estimation used in traditional Retinex algorithm can enhance the shadow edge falsely during smoothing face images, a novel Retinex algorithm based on a daptive smoothing with new a conduction function is proposed . This conduction function uses both spatial gradient and local inhomogeneity to measure the severity extent of pixel variation, which can smooth face images well without enhancing edge effects and lossing feature edges, During iterative process, the maximum between this iteration and last iteration is chosen to be the constraint,so this a daptive smoothing method with the proposed conduction function can be applied in Retinex theory to estimate illumination. Experimental results based on the Yale B face database show that the proposed algorithm can overcome shadows in side-illumination efficiently without losing feature edges of face images. The recognition rate is improved by 24. 2% in the side-illumination condition at the best rate,
收稿日期：2009-01-22；修订日期：2009-04-01.
基金项目：国家863高技术研究发展计划资助项目（No.2005AA778032)）