基于统计学习优化SIFT的面部遮挡人脸识别

Face Recognition with Facial Occlusion Based on SIFT Optimized by Statistical Learning

针对传统的人脸识别算法受面部遮挡的影响导致很难兼顾鲁棒性和保持原始图像核心信息的问题,本文提出了一种基于统计学习优化尺度不变特征变换的面部遮挡人脸识别算法。首先,利用SIFT将所有给定训练图像用一组局部特征描述符表示出来;然后,通过执行统计学习获得正常脸部图像SIFT特征的概率分布函数,利用获得的概率分布函数在新观察到的测试图像中检测异常SIFT特征;最后,计算测试图像与训练图像之间的相似度,并利用K近邻分类器完成人脸识别。在AR人脸数据库上的实验验证了本文算法的有效性及可靠性,实验结果表明,相比其它几种较为先进的人脸识别算法,本文算法取得了更强的识别鲁棒性。

The traditional face recognition algorithms do not keep core information of original images with robustness, for which the algorithm of scale-invariant feature transform optimized by statistical learning is proposed.Firstly, given training images are denoted by a group of local features descriptors by using SIFT. Then, probability distribution function（PDF） of facial images SIFT features is got by performing statistical learning, and PDF is used to detect abnormal SIFT features of testing images. Finally, similarities between testing and training images are calculated and K near neighbor classifier is used to finish face recognition. The effectiveness and robustness of proposed algorithm has been verified by experiments on AR database. Experimental results show that proposed algorithm has stronger robustness than several advanced face recognition algorithms.