不可控光照下的LNMCP人脸识别方法  

LNMCP face recognition method under uncontrollable lighting conditions

作　　者：周丽芳[1,2] 李伟生[3] 房斌[2] 王立逗[3] 

机构地区：[1]重庆邮电大学软件学院,重庆400065 [2]重庆大学计算机学院,重庆400044 [3]重庆邮电大学计算机科学与技术研究所,重庆400065

出　　处：《计算机工程与应用》2013年第10期173-176,187,共5页Computer Engineering and Applications

基　　金：国家自然科学基金项目(No.61100114;No.61272195;No.61173129;No.61171060;No.61201383);重庆市自然科学金(No.cstc2012jjA1699);重庆市教委科学技术研究项目(No.KJ100519);重庆邮电大学自然科学基金项目(No.A2009-34);中央高校基本科研业务费资助(No.CDJXS11182240;No.CDJXS10182216);重庆邮电大学博士启动基金(No.A2012-20)

摘　　要：局部二值模式(LBP)在人脸识别领域取得了显著成效,但由于没有考虑像素之间的对比度,导致部分重要纹理特征被丢弃。采用粗粒度分区算法求取各人脸分块对比度值域区间的动态阈值点,并借鉴LMCP算法思想利用上述阈值点分别对各人脸分块编码,得到结合光照变化信息的局部非线性多层对比度特征LNMCP作为人脸特征。实验结果表明,该方法识别率高,特别是在户外人脸库上提高了2.91%,方法在不可控光照变化环境下具有极强的自适应性。Local Binary Pattern（LBP）has been used with considerable success in face recognition field, while it hasn’t considered the contrast between pixels so that some important texture features have been abandoned. This paper proposes to calculate dynamic thresholds of the contrast range by coarse grain algorithm, and extracts the face feature with the thresholds, meanwhile the code method is similar to Local Multi-layer Contrast Pattern（LMCP）. Moreover, Local Nonlinear Multi-layer Contrast Pattern（LNMCP） can get the face features, which contained the information of illumination various. The experiments show that the proposed method has a high recognition rate, the results on outdoor database can be enhanced by 2.91% especially. It proves that the method will be adaptive under uncontrollable varying illumination.

关 键 词：局部非线性多层对比格局 局部二值模式 粗粒度算法 动态阈值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]