基于视觉条件评估的自适应人脸验证  

Adaptive face verification based on visual condition estimation

作　　者：黄仁杰[1,2] 李涛[1,3] 窦育民[1] 刘枫[2] 

机构地区：[1]电子科技大学计算机科学与工程学院,成都611731 [2]西南大学计算机与信息科学学院,重庆400715 [3]河南广播电视大学信息工程系,郑州450008

出　　处：《计算机应用研究》2015年第12期3805-3809,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61375038);河南省科技厅科技攻关项目(142102210010)

摘　　要：为了减少非限定环境下干扰对人脸验证的影响,提出一种基于视觉条件评估的自适应人脸验证算法。该算法首先把输入的一对人脸图像分割成多个人脸区域;然后采用支持向量回归机方法评估每个区域上图像对的视觉条件并选出可靠区域;最后联合多特征融合、度量学习和支持向量机等技术在所选区域上执行分类验证该对人脸。实验结果表明,根据人脸图像间的视觉条件自适应地选择人脸区域,可以有效地减少复杂干扰的影响,并能显著提高非限定环境下的人脸验证精度。For reducing the effects of interferences in face verification in unconstrained environments,this paper proposed an algorithm of adaptive face verification based on visual condition estimation. It firstly cropped a pair of input faces into multiple facial regions. Then it adopted support vector regression to estimate the visual conditions of image pair on each region and select such reliable regions. Finally,by combining multiple technologies of fusion of multi-feature, metric learning and support vector machine, it implemented the classifications to verify the pair of faces on the selected facial regions. The experiment results show the effects of the complicated interferenees can be effectively reduced by adaptively selecting facial regions according to the visual conditions between face images. It significantly improves the accuracy of face verification in unconstrained environments.

关 键 词：视觉条件评估 支持向量回归 人脸验证 自适应选择 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]