如何处理人脸识别中的多态性一直是人脸识别领域的一个难题。传统的图嵌入算法忽视了同类中多态子类间的同属关系，而且也没有恰当地处理异类间的区别信息。该文提出一种鲁棒的图嵌入人脸识别算法，该算法可以恰当地模拟同类中的多态间关系，而且能在局部流形结构与全局区别信息间实现平衡。基于多个公开数据库的人脸识别实验证明了该方法的有效性。%It is always a difficult problem in face recognition on how to process the multimodal information (e.g. variation in lighting or orientation). Traditional graph embedding algorithms neglect congener correlation between different multimodal clusters of the same class (i.e. subject) and do not properly incorporate discriminative information between classes. In this paper, a robust graph embedding face recognition algorithm is proposed. It properly captures multimodal structure within one class and also realizes a balance between local manifold structures and the global discriminative information. Experiments in several public databases demonstrate that the proposed algorithm can achieve better performance than the state-of-arts reported in recent literatures.
展开▼
