3D Animation Automatic Generation System Design Based on Deep Learning.

Area of Science:

• Computer Graphics
• Artificial Intelligence
• Animation

Background:

• Generating realistic facial expressions in 3D animation is challenging due to insufficient image detail.
• Existing methods often lack the realism required for believable animated characters.

Purpose of the Study:

• To propose a novel deep learning-based method for generating realistic facial expressions for 3D animated characters.
• To address the limitations of current expression generation techniques by enhancing image detail and feature extraction.

Main Methods:

• Utilized improved deep learning with cascade classifiers trained on real facial expression images.
• Implemented techniques to extract, soften, and enhance facial expression features from real images.
• Developed a unified content and style approach with a loss function incorporating content and style constraints for feature fusion.

Main Results:

• The proposed method demonstrated accurate feature point localization in expression generation.
• Achieved a high Pearson correlation coefficient and low root mean square error between input and generated images.
• Significantly enhanced the realism of generated animated character facial expressions.

Conclusions:

• The improved deep learning method effectively generates realistic facial expressions for 3D animation.
• The approach successfully enhances feature details and fuses information under specific constraints.
• This technique offers a promising solution for improving the quality of animated character expressions.