基于发育网络的复杂背景下人脸识别  

Face Recognition with Complex Background Based on Developmental Network

作　　者：王东署 郑光璞 

机构地区：[1]郑州大学电气工程学院,河南郑州450001

出　　处：《郑州大学学报（工学版）》2016年第1期19-23,共5页Journal of Zhengzhou University（Engineering Science）

基　　金：国家自然科学基金资助项目(61174085)

摘　　要：针对人脸识别中光线、背景及面部表情的变化,利用密歇根州立大学人脸库构建复杂背景人脸图像库,并将发育网络应用于二维人脸识别.引入突触维护机制来处理复杂背景,可以自动决定突触的完全去除、保留和部分去除,弱化背景并强化人脸,从而降低复杂背景对人脸识别的不利影响.通过和传统方法测试结果对比,采用带突触维护机制的发育网络对复杂背景下人脸的识别可以取得更好的效果,其识别率可达96%以上.Aiming at the change of light,background and facial expressions in face recognition,this paper constructs a face database with complex background based on the face database of Michigan State University and applies the developmental network（ DN） to the two-dimensional face recognition. The synapse maintenance mechanism is introduced to deal with the complex background,which can automatically determine the synapses＇ complete removal,partial removal or retention. Moreover,it can weaken the background,strengthen the face and consequently decrease the negative effects of the complex background. Compared with the test results of the traditional methods,DN with the synapse maintenance mechanism can recognize faces with complex background effectively and the recognition rate can reach above 96%.

关 键 词：发育网络 人脸识别 复杂背景 突触维护 识别率 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]