一种同步人脸运动跟踪与表情识别算法

针对单视频动态变化背景下的人脸表情识别问题,提出了一种同步人脸运动跟踪和表情识别算法,并在此基础上构建了一个实时系统.该系统达到了如下目标:首先在粒子滤波框架下结合在线外观模型和柱状几何模型进行人脸三维运动跟踪;接着基于生理知识来提取人脸表情的静态信息;然后基于流形学习来提取人脸表情的动态信息;最后在人脸运动跟踪过程中,结合人脸表情静态信息和动态信息来进行表情识别.实验结果表明,该系统在大姿态和丰富表情下具有较好的综合优势.

A Simultaneous Facial Motion Tracking and Expression Recognition Algorithm

In view of facial expression recognition from monocular video with dynamic background,a real-time system was proposed based on the algorithm in which facial motion is tracked and facial expression is recognized simultaneously.Firstly,online appearance model and cylinder head model were combined to track 3D facial motion from video in framework of particle filtering;secondly,the static knowledge of facial expression was extracted through facial expression anatomy;thirdly,the dynamic knowledge of facial expression was extracted through manifold learning;fourthly,facial expression was retrieved by fusing the static knowledge and dynamic knowledge during facial motion tracking process.The experiments results confirmed the advantage on facial expression recognition even in the presence of significant head pose and facial expression variations of this system.