人脸特征点跟踪及夸张人脸动画系统

Facial Feature Point Tracking and Exaggerated Face Animation System

为了研究视频中人脸特征点的跟踪问题,根据人脸特征点的不同特征,提出了不同的跟踪方法.对于眉都等特征点,利用光流跟踪方法,用拉普拉斯金字塔图代替了通常的灰度图作为光流输入图,减少了累积误差,提高了光流跟踪的准确性;对于嘴部特征点,将光流与弹性图匹配相结合,通过光流预先得到大致位置,减小了弹性图匹配时的搜索范围,提高了跟踪速度;对于眼部特征点,采用图像二值化方法进行跟踪.通过MPEG-4机制将跟踪到的运动数据克隆到系统根据真实人脸特点生成的夸张人脸上进行动画,具有很强的娱乐色彩.

In order to track facial feature points, according to the different features of different points, many methods are used to track. For the points on the brow, the authors use optical flow to track, and the authors also use Laplacian pyramid image replacing gray image as the input image to improve the tracking veracity. For the points on the mouth, the authors combine the optical flow with EBGM, using optical flow to get the position roughly and reduce searching area to accelerate the tracking speed. For the points on the eyelids, image binarization method is used. Then based on MPEG-4 mechanism, the authors clone the tracked motion data to an exaggerated character, which is automatically generated by the system according to a real face. Experiments show that the result is recreational.