Virtual Actor's Facial Expression Modeling Based On Interactive Evolutionary Computation

Three-dimensional computer graphics (3DCG) animated movies have become a modern form of entertainment. Virtual actor's facial expression modeling has become an important and practical topic. But at the same time, the demand for the virtual facial expression is increasing, so it has been an important topic for animated directors how to make the expression realistic, vivid, and fit for the people's requirements. Previous works have demonstrated that increasing levels of knowledge about expressive facial animation can be encoded within the animation system, thereby transferring expert knowledge from human animators to increasingly knowledgeable virtual actors maybe feasible to generate virtual facial expression.So this paper tries to establish a model which people can get the virtual actor's facial expression based on the expert's knowledge. And the training samples can be got from the expert fuzzy rules, with aid of supervised training of the virtual actors by a human director via TSK fuzzy neural network that fine tunes the fuzzy rules, results more satisfactory to a human director can be produced. In order to speed up the convergence of the learning algorithm and avoid the disadvantage of BP, PSO algorithm is incorporated into the TSK fuzzy neural network. Our experimental results indicate the success of our approach here compared to BP.However during the experiments, it is found that not all expert fuzzy rules are fit for all animated director's demand. So Interactive Evolutionary Computation (IEC) is utilized to training the TSK fuzzy neural network, IEC fuses the capability of Evolutionary Computation optimization and the human evaluation. Our experimental results indicate that the convergence of IEC is faster than EC. Different human directors can use such a system to influence the knowledge-based system to generate results suitable to their personal preferences.