3D Blend-Shape Generation From Scans Using Neural Segmentation

The patent replaces the manual mechanical process of analyzing 3D data shots and creating blend-shapes with an automated neural network system. The NN model processes 3D scan data to automatically generate blend-shapes, substituting human animation designers' manual work with machine learning-based automated generation, thereby reducing time consumption while maintaining accuracy through the NN's learning capabilities.

The neural network model performs self-service by automatically learning from 3D scan data and generating blend-shapes without requiring manual analysis by animation designers. The NN model independently processes the data, extracts facial movement patterns, and creates the necessary blend-shapes, eliminating the labor-intensive manual analysis process while maintaining the ability to reproduce detailed facial expressions.

The patent applies preliminary action by pre-processing the 3D scan data into a standardized format with identified landmarks and mesh structures before feeding it to the neural network. This preliminary organization of data reduces the computational complexity during the actual blend-shape generation phase, allowing the NN model to work more efficiently with structured input data, thereby reducing the overall computational energy required while maintaining high generation speed.

The patent divides the facial mesh into multiple independent segments (such as lips, cheeks, forehead) that can move independently to create realistic facial expressions. Each segment is assigned specific blend-shape weights and transformation parameters that allow independent control. This segmentation enables precise control over facial expressions by manipulating individual segments, while the neural network automatically manages the complexity of coordinating these segments through learned relationships from training data.