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.
出处
《计算机科学》
CSCD
北大核心
2013年第5期307-310,共4页
Computer Science
