A generative sketch model for human hair analysis and synthesis

Area of Science:

• Computer Vision
• Computer Graphics
• Image Analysis

Background:

• Traditional 3D hair models are physically-based.
• Existing methods lack efficient analysis and synthesis capabilities.

Purpose of the Study:

• To present a novel generative sketch model for human hair analysis and synthesis.
• To develop a view-based representation for hair images.

Main Methods:

• Treating hair images as 2D piecewise smooth vector fields.
• A three-level generative model: high-frequency band, vector field, and attribute sketch graph.
• Utilizing a Swendsen-Wang cut algorithm for inference.

Main Results:

• Successful analysis, synthesis, and rendering of diverse hair styles.
• Demonstrated realistic hair image synthesis from sketch graphs.
• Enabled stylistic drawings (rendering) from sketch graphs.

Conclusions:

• The generative sketch model offers a straightforward method for hair image synthesis and rendering.
• The view-based representation is effective for hair analysis.
• The model handles diverse hair styles effectively.