基于改进的LDP人脸识别算法  

The Face Recognition Algorithm Based on Improved Local Directional Pattern

作　　者：武伟[1] 江艳霞[1] 张喆[1] 

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《光电工程》2014年第12期72-77,共6页Opto-Electronic Engineering

基　　金：国家自然科学基金青年基金项目(61203143);国家自然科学基金(60874002);沪江基金(C14002)资助项目

摘　　要：LDP算法是将与Kirsch算子运算后得到的一些负值的邻域灰度值作为中心灰度值的编码因子,这将会导致中心灰度值不能很好的反映邻域局部特征信息,从而降低识别率。针对于此,本文提出了一种改进的局部定向模式(CLDP)算法。该算法在LDP的基础上,去掉邻域灰度值为负值的因子,对中心灰度值进行重新编码,由于该编码值是将与Kirsch算子运算后的正值最大值作为图像边缘输出,使中心灰度值能很好反映邻域的局部特征信息,从而提高人脸的识别率。本文还将提出的CLDP算法用在YALE,ORL,JAFFE等人脸数据库中进行人脸识别。从实验的结果表明,该方法识别性能较LDP算法,LDN算法以及ELDP算法等更好的。Local Directional Pattern （LDP） algorithm is a local feature extraction method which is robust to random noise. It has been widely used in face recognition. As LDP algorithm encodes the negative directions convolved by the Kirsch masks, the value of center gradation can’t fully reflect the information in the field of the local features. Thus face recognition performance will be influenced. A new algorithm named CLDP based on LDP algorithm is proposed. The CLDP algorithm encodes the positive directions which are convolved by the Kirsch masks. These directions can fully reflect the information in the field of the local features. And thus face recognition accuracy can be enhanced. To evaluate the performance of the proposed CLDP algorithm, some experiments are conducted in YALE, ORL and JAFFE face databases. Experimental results show that our CLDP algorithm has better recognition performance compared with LDP, LDN and ELDP algorithm.

关 键 词：LDP 局部邻域 人脸识别 KIRSCH算子 鲁棒性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]