基于非约束图像参考集匹配的视频人脸识别

针对视频人脸识别中系统不能很好地学习人脸图像有效特征的问题,提出了非约束图像参考集匹配方法,通过在两个图像集之间预先定义参考集构造多个离线的局部模型,并分别与参考集进行匹配,无需考虑所有的成对情况,从而计算出它们的相似度,有效地将视频人脸识别问题转化成二次规划问题。所提方法的有效性在Honda、MoBo及YouTube三大视频人脸数据库上进行了验证,实验结果表明,与现有的视频人脸识别方法相比,所提方法取得了更好的识别效果。

Video Face Recognition based on Non-Binding Image Referring Set Matching

For the case that system cannot extracts effective feature of the face images in video face recognition (VFR), Non-Binding Image Referring Set Matching (NB-LRSM) is proposed by which in advance between the two image-sets constructing multiple offline local models. Match with the reference set rather than considering all pairs of cases which used for calculating similarity. This challenge is transformed into a quadratic programming issue, which contains the local neighboring relations and consistency of the local geometry. The effectiveness of the proposed method is validated in Honda, MoBo and YouTube video face databases. Experiments results indicate that proposed method has better recognition accuracy comparing with several existing video face recognition approaches.