Animation Character Mouth Matching Model Considering Reinforcement Learning and Feature Extraction

Abstract

With the development of the times, animation production has become increasingly sophisticated, and mouth matching is one of the key points to ensure the vividness and realism of animated characters. Therefore, a study proposes an animated character mouth matching model that takes into account reinforcement learning and feature extraction. Using the Actor-Critic method in reinforcement learning, the extracted audio and facial features are used as input states to predict the next facial feature in the next time step. The experimental results show that the proposed model has an average accuracy of 95.61% and an F1 value of 97.13% on three databases. Meanwhile, the peak signal-to-noise ratio and structural similarity index are 41.77 and 0.93, respectively, which are better than their comparison methods. In addition, the study tested the error of mouth shape under different emotions, and the results showed an average mean square error of only 6.639. Finally, the user survey results showed that the animated characters generated by the proposed model received more recognition in mouth shape matching and realism, with a highest selection rate of 98.64%. The successful application of the proposed model provides new ideas and methods for research in related fields, laying the foundation for further promotion and innovation of animation production technology.