SSPO:一种部分遮挡单样本人脸识别算法

SSPO: A Partial Occlusion Single Sample Face Recognition Algorithm

单样本人脸识别问题是一个比较具有挑战性的问题。针对部分遮挡单样本人脸识别困难的问题,本文提出了一种SSPO算法。首先对所有样本进行遮挡,通过MSEP算法进行分块,遮挡识别和剔除。为了对人脸姿势、光照等变化进行识别,我们抽取样本库中邻接块构造一个样本字典,通过类内变化字典预测可能的人脸变化。最小化待测人脸与样本库、变化字典的残差预测样本分类。最后,在AR人脸数据库和Multi-PIE人脸数据库上验证了本算法的有效性。

The single sample face recognition is a challenging problem. Aiming at the issue of partial occlusion single sample face recognition, an algorithm called SSPO is proposed. Firstly, all samples are occluded, and the MSEP algorithm is used for block identification and removal. In order to recognize changes in face pose, illumination, etc., adjacent blocks are extracted from the sample library to construct a sample dictionary, and possible face changes are predicted through the intraclass change dictionary. The researchers minimize the residual of the face to be tested, the sample library, and the change dictionary to predict its classification. Finally, the algorithm is verified effective in AR face database and MultiPIE face database.