MPEG-4中脸部动画参数和序列重绘的肌肉模型

摘要

MPEG-4中定义了“人脸对象”这样一种特殊的视频对象,并通过脸部动画参数FAP和脸部定义参数FDP来对这类对象进行编码，以实现极低码率的视频编码.通过对MPEG-4中“人脸对象”这类视频码流的句法结构和参数编码方法的详细分析.以及通过对MEPG-4解码器图象重绘(rendering)过程的研究，在Waters的以肌肉收缩强度为参数的肌肉模型基础上，提出了更适应于MPEG-4参数的位移控制肌肉模型(displacement-controlling muscle model)，从而实现了通过利用MEPG-4码流中的FAP和FDP参数来重建自然表情的人脸视频序列.%Face object, is a special visual object defined in MPEG-4. Facial definition parameter(FDP) and facial animation parameter(FAP) are the sets of parameters to calibrate and animate the face object. The bitstream syntax for FDPs and FAPs in MPEG-4 are analyzed in this paper. The meaning of high level FAPs, viseme and expression, are explained with examples. All FAPs can be compressed with mask-based DPCM and/or DCT encoding. A rendering technology——displacement-controlling muscle model is proposed to reconstruct video sequences with natural facial expression from decoded FAPs and FDPs in MPEG-4 bitstream. There are two muscle models for different muscle stryles, parallel muscle and orbicular muscle. Different from traditional muscle models, which are controlled by intensity of muscle contraction, the proposed models are controlled by displacement of key-points affected by contraction of muscle. The reason behind this technology is that the visible displacement of key points is easier to be extracted from orighnal video sequence then the invisible muscle contraction. As an experiment result, reconstructed “Mona Lisa” with different expression is illustrated.