Gabor字典及l_0范数快速稀疏表示的人脸识别算法  

Face recognition based on fast sparse representation of Gabor dictionary and l_0 norm

作　　者：曾军英[1] 甘俊英[1] 翟懿奎[1] 

机构地区：[1]五邑大学信息工程学院,广东江门529020

出　　处：《信号处理》2013年第2期256-261,共6页Journal of Signal Processing

基　　金：国家自然科学基金项目(No.61072127);广东省自然科学基金项目(NO.S2011040004211;No.10152902001000002;NO.S2011010001085;No.07010869);广东省高等学校高层次人才项目(粤教师函[2010]79号);广东高校优秀青年创新人才培养计划项目资助(No.2012LYM_0127)

摘　　要：针对光照、遮挡、伪装情况下,识别率比较低,识别时间长的问题,本文提出了基于Gabor字典及l0范数快速稀疏表示的人脸识别算法。Gabor小波提取的特征能够克服遮挡、光照等干扰对人脸识别的影响,平滑l0算法通过平滑连续函数来近似l0范数,只需较少测量值并且较快速度便能重构稀疏信号。本算法通过提取人脸的Gabor特征、主成分分析法(PCA)降低维度,l0范数快速稀疏分类完成识别。在伪装人脸情况下,分块计算Gabor人脸特征,提高Gabor字典的形成速度。基于AR人脸数据库的实验结果表明,本算法可在一定程度上提高识别速度和识别时间,即使在小样本情况下,依然具有较高的识别率。Many classic face recognition algorithms degrade sharply when they are used at identifying an individual under various conditions such as illumination, camouflage. A fast sparse representation face recognition algorithm based on Gabor dictionary and smoothed 10 norm is presented in this paper. Gabor filters, which could effectively extract local directional features of the image at multiple scales, are less sensitive to the variations of illumination and camouflage. Smoothed 10 algorithm requires fewer measurement values by continuously differentiable function approximation 10 norm. The algorithm obtains the local feature by extracting Gabor feature, reduces the dimensions by principal component analysis （PCA） and realizes fast sparse by 10 norm. Under camouflage condition, the algorithm blocks Gabor facial feature and improves the speed of formation of the Gabor dictionary. Experimental results on AR face database show that the proposed algorithm can improve recognition speed and recognition rate and can generalize well to the face recognition, even with a few training image per class.

关 键 词：稀疏表示 Gabor字典 l0 范数 伪装 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]