摘要
提出了一种对相似度空间进行寻优的新方法,以提高开集人脸识别的准确率。该方法首先将开集识别问题转化为二分类问题,然后引入寻优方法寻找分割相似度空间的最优超平面,该超平面能够将相似度空间分割为接受空间和拒绝空间两部分。在判别过程中,利用相似度向量在空间中的位置判断样本是否为已知类。由于利用了相似度空间中向量分布的信息,训练出的特征具有更强的分类能力。通过不同人脸库的实验表明,相对于传统的方法,本文所提的方法能显著地提高开集识别的准确率。
This paper studies a novel classification algorithm based on similarity space division to enhance the accuracy in open set face recognition. In this algorithm, open set problem is transformed to a 2-class classification problem, and a hyper plane is searched to divide the similarity space and separate vectors of the 2-class. Then the algorithm rejects the unknown identity by the relative position of the hyper plane. Hence, the feature has strong classification ability, in view of the discrimination information abstracted from the similarity space. Experimental results on several face databases demonstrate that similarity space division based method in this paper significantly outperforms the traditional method for open set face recognition.
出处
《微型电脑应用》
2010年第6期31-32,44,共3页
Microcomputer Applications
关键词
人脸识别
开集
相似度空间
Face Recognition
Open Set
Similarity Space