基于局部特征在线学习的视频人脸识别

摘要

针对视频人脸识别实时性问题，提出一种基于局部特征在线学习的视频人脸识别方法。首先，从人脸任意关键点提取局部特征，采用投票算法挑选出每个簇的代表性特征；然后，进行学习过程，直到簇中人脸数目达到要求的最小值且人脸图像距其簇平均的最远距离低于一个阈值；最后，将检测到的视频帧按顺序与图库中所有个体的簇进行匹配，利用复合时序相似度度量完成人脸的识别。在一个有50个注册对象和20个未知者的数据库上进行在线识别实验，获得了97．8％的识别率。实验结果表明，相比其他几种视频人脸识别算法，该算法取得了更好的识别效果。%For the issue of real-time video face recognition,we propose a video face recognition method which is based on online learning with local feature.Firstly,it extracts local features from any critical point of face and picks out using voting algorithm the representative fea-tures of each cluster.Then,the learning process is conducted until the face numbers in cluster reaches the minimum value and the farthest distance between face image and its cluster average is less than a threshold value.Finally,the method matches the video frame detected with all individual clusters in the gallery according to sequence,and uses composite sequence similarity metric to complete the face recognition. Online recognition experiments are done on a database with 50 registered and 20 unknown objects,the recognition rate reaches 97.8%.Ex-perimental results show that the proposed algorithm has better recognition effect than several other video face recognition algorithms.