An Automatic Muscle Construction Method Based on SVM Training for Human Facial Animation

摘要

Artificial muscle construction for human facial animation is an interesting and complex work. In this paper, a Support Vector Machine and Clustering Analysis (SVM-CA) method is proposed to implement the laborious process. In this method, a human head model is considered as a training set whose samples are the vertices on this geometry's profile. The vector in SVM is consist of vertices's properties like spatial location, curvature, normal variations and so on. According to SVM-CA method, 13 feature points are generated which help to build the facial muscles automatically. The experimental results demonstrate an inspiring matching rate comparing with the landmark vertices by character-design artists. The muscles automatically constructed by SVM-CA can fit different human head geometries very well, and using this method, a group of characteristic facial expressions and reasonable mouth shapes are synthesized in real time.