面向光照可变的人脸识别方法  

Face recognition method for scenario with lighting variation

作　　者：李昕昕[1] 陈丹[2] 许凤娇[2] 

机构地区：[1]四川大学锦城学院,成都611731 [2]重庆邮电大学计算机科学与技术研究所,重庆400065

出　　处：《计算机应用》2013年第2期507-510,514,共5页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61202191);中央高校基本科研业务费专项资金资助项目(SWJTU12CX095)

摘　　要：传统Retinex算法在侧光严重的情况下难以消除阴影,为此提出一个对数形式的传导函数,取得了很好的光照补偿效果。为提高人脸识别率,将该问题看成一个典型的模式分类问题,提出基于局部二值模式(LBP)特征的支持向量机(SVM)人脸识别方法,使用"一对一"的方法将多类问题转化为SVM分类器可以解决的两类问题,实现了高效的人脸识别。在CMU PIE、AR、CAS-PEAL以及自行采集的人脸库上进行了仿真实验,结果表明该方法能够有效地去除光照影响,相对传统方法具有较优的识别性能。With serious sidelight, it is difficult for the traditional algorithm to eliminate shadows. To improve the illumination compensation effect, a logarithmic transformation function was presented. In order to improve the performance of face recognition, by taking this problem as a classic pattern classification problem, a new method combining Local Binary Pattern （LBP） and Support Vector Machine （SVM） was proposed. One-against-one was used to convert multi-class problem to two-class problem, that can be used by SVM. Simulation experiments were conducted on the database of CMU PIE, AR, CAS-PEAL and one face database collected by the authors. The results show that lighting effects can be well eliminated and the proposed method performs better than the traditional ones.

关 键 词：人脸识别 光照 局部二值模式 支持向量机 视网膜皮层 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]