【作者】 张军； 徐光佑； 陶霖密；

【Author】 Zhang Jun1+ Xu Guangyou1 Tao Linmi11(Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China)

【机构】 普适计算教育部重点实验室清华大学计算机科学与技术系；

【摘要】 实现了基于MPEG-4动画框架,支持包括话语动作、表情动作及自由脸部动作等多种动因人脸动作及其融合的人脸动画编著系统。提出了能够对与话语无关的自由脸部动作的风格进行描述的ISM-FFA模型和模拟合成算法,使人脸动画效果更加生动自然且可以表现出个性化的脸部动作。提出了基于“主次动作及其重要性的动态调整”的动作融合算法,使并发脸部动作能够更协调平滑地融合。更多还原

【Abstract】 Facial animation plays an important role in conversational communicationbetween human beings. A realistic computer-synthesized facial animation can improvethe realism and liveliness of the human-computer interaction. This paper presents aMPEG-4 based facial animation authoring system which supports several kinds of facialactions including phonetic facial actions, emotional expressions and free facial actions.Phonetic actions are driven by phonemes derived from text or voice by integrated TTSand speech recognizer. Emotional expressions are driven by bookmarks in the text.Free facial actions indicate the actions independent of the speech content, which was notemphasized in research before. But without them, the facial animation appears stiff.There are different types of free facial actions such as blinking, shaking head, etc. Oursystem differentiates 13 types, which are represented by 13 groups of FAP (FacialAnimation Parameters). Different persons have different styles of free actions. This paperproposes a model named ISM-FFA (Individual Style Model of Free Facial Animation) torepresent styles of free facial actions. An individual style can be represented by theranges of duration time, internal time, occurrence probability and the range of intensityof every type of free actions. Based on a given ISM-FFA model instance, the systemsimulates the personal free facial actions by determining duration time, internal time andstrength of each occurrence of each type of free actions at run time within the rangesconstrained by the model, concatenates every occurrence of the same type, combines alltypes of them and then constructs the whole personal free facial actions. The facialanimation with free actions by our method looks lively and more comfortable.Different kinds of facial actions normally occur concurrently. This paper proposes analgorithm to blend them smoothly. The result action is a weighted combination of twoactions. But the weights and the choice of the primary and the secondary actions aredynamically adjusted at every facial part at every time point. The combination of facialactions is similar to the combination of real person’s combination of those. And thealgorithm ensures the whole result facial animation is within reasonable action strength,temporal continuous at every part and spacial continuous at any time point. So thesynthesized facial animation by this authoring system is harmonized, smooth and natural.更多还原