人脸识别中遮挡区域恢复算法研究  

Research of Face Occlusion Recover Algorithm in Face Recognition

作　　者：杜杏菁[1,2] 郭明雄[2] 

机构地区：[1]北京理工大学光电学院光电成像技术与系统教育部重点实验室,北京100081 [2]华北科技学院计算机系,燕郊065201

出　　处：《计算机科学》2013年第5期307-310,共4页Computer Science

摘　　要：在分析脸部遮挡处理各算法的基础上,提出了自动多值掩模PCA人脸重建模型(MMPCA模型)。该模型首先进行特征提取,计算待测人脸和标准样本的特征脸差,判断遮挡部位,即遮挡类型;接着使用M估计器对遮挡掩模进行估计,为不同像素点估计符合自身特性的幅度参数,生成多值遮挡掩模;再通过3个半二次型函数迭代保证最优合成系数唯一与收敛,获得最优合成系数,重建人脸。实验结果表明,该算法能恢复遮挡部位,减弱遮挡对识别准确率的影响。Based on the analysis of facial occlusion handling various algorithms, a automatical multi-value mask PCA face reconstruction model （MMPCA model） was proposed. First the facial features are extracted, and the eigenface difference between a test face and a standard face is calculated to judge the facial occlusion accurately and determine the facal occlusion type. Then a occlusion mask is estimated with a M estimator, and the pixel＇s amplitude parameters are estimated, and the multi-value occlusion mask is generated. Last three half secondary-type functions are used to ensure optimal synthesis coefficients is only and convergent, and then the optimal synthesis coefficient is obtained to recon struct the face. The experimental results show that the occlusion region is restored and the occlusion influence to face recognition accuracy is weaken through using the automatical multi-value mask PCA face reconstruction model.

关 键 词：人脸识别 遮挡恢复 多值掩模PCA人脸重建(MMPCA) 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]