用SIFT词汇树实现的姿态无关的人脸识别

Pose-invariant Face Recognition via SIFT Vocabulary Tree

自动人脸识别是智能视频监控的重要组成部分.为提高人脸姿态不确定时的识别准确率,提出一种姿态无关的识别方法.该方法分为训练与识别2个阶段:在训练阶段,利用样本图像的SIFT特征构造词汇树,基于该词汇树对每幅图像进行定量表示,并利用保局投影进行维度约减;在识别阶段,通过提取待识别图像的SIFT特征,利用已有词汇树表示图像,并通过保局投影得到低维特征,采用K近邻方法进行识别.实验结果表明,该方法在人脸姿态不确定的情况下能够有效地提高识别准确率.

Face recognition is an important part of intelligent video surveillance. To improve the recognition accuracy in case of unknown facial pose, we propose a pose invariant face recognition approach which includes two stages. In training stage, we construct a vocabulary tree using SIFT features of sampled images, and compute the feature descriptor of each face based on the vocabulary tree. We then reduce the feature dimensionality through locality preserving projections （LPP）. In recognition stage, we extract the SIFT features of the test face and compute its feature descriptor through the existed vocabulary tree, then obtain the low dimensional feature by LPP algorithm, K-NN algorithm is used for face recognition. Experimental results show that the proposed approach improves recognition rate under unknown pose to a large extent.