【作者】 张宁；

【导师】 王春立；

【作者基本信息】 大连海事大学 ， 计算机科学与技术， 2010， 硕士

【摘要】 人脸动画合成是计算机视觉及计算机图形学领域中的一个热门研究课题,它有着非常广泛的应用领域,如：影视制作、医学研究、虚拟会议、虚拟主持人等等。由于不同类型的人脸在脸的大小、五官的位置、大小、比例等方面存在着差异,如果将运动参数不作调整直接作用于人脸模型,可能会得到与真实人脸不一致的表情,甚至怪异的表情。因此必须针对不同类型的人脸对运动参数做出适当的调整,从而得到比较自然的表情。具体而言,本文的研究工作如下：1.对视频中人脸的特征点进行检测,得到人面部五官的运动参数。2.本文对人脸模型按照一定的分区原则进行功能分区,然后手动获取人脸模型各个控制点的运动因子,利用已经获得的运动参数和运动因子得到每个功能分区的动画单元AU,为了消除人头部晃动带来的误差,对特征点的坐标进行修正,最终实现了视频驱动的人脸动画。3.为了使不同人脸视频驱动模型获得较好的动画,本文采用了全局变换、局部变换、FAPU变换。实验结果表明,采用上述3种变换后,在每个实验者的视频驱动下,都基本得到了令人满意的动画效果。更多还原

【Abstract】 Facial Animation Synthesis is a hot research topic in the fields of Computer Vision and Computer Graphics. It has a very wide range of applications, such as video production, medical research, virtual meetings, virtual host and so on.Because the location, size and proportion of facial features vary greatly from person to person, if motion parameters are employed directly to drive the face model without adjustments, we may get expressions which are inconsistent with the real face, or even the strange ones. Therefore, it is necessary to make appropriate adjustments in the motion parameters of different types of face to obtain more natural expressions. The main research contents are as follows:1. The feature points on the face in the video are detected at first. And then the facial motion parameters are obtained.2. In this paper, the face model is divided into some function partitions in accordance with certain principles. Then the movement factors of the various control points on the face model are obtained manually. The animation units AU of every divided function are obtained by using the motion parameters and movement factors of the various control points. The coordinates of the features are modified in order to get rid of errors which are caused by shaking of head. Ultimately, the video-driven facial animation is achieved.3. In order to obtain a better facial animation when different types of facial videos drive the face model, the global transformation, local transformation, and FAPU transformation are employed. Experimental results show that the satisfactory animation nearly has been obtained when all the experimenters’videos drive the standard face model after using the three kinds of transformation.更多还原