Single-View 3D Hair Modeling With Clumping Optimization.

Area of Science:

• Computer Vision
• Computer Graphics
• Deep Learning

Background:

• Deep learning enables hair geometry generation from single images, but lacks realism for applications like rendering and simulation.
• Missing hair clumping effects, influenced by scalp oils, are crucial for realistic hair appearance and simulation.

Purpose of the Study:

• To integrate realistic hair clumping effects into single-view hair reconstruction.
• To improve the quality of generated hair geometry for advanced applications.

Main Methods:

• Introduced a hierarchical hair representation with a clumping modifier for guide hairs and skinning-based expressions.
• Developed a fully differentiable framework using neural clumping measurement and line-based rasterization.
• Iteratively solved guide strand positions and clumping parameters for enhanced realism.

Main Results:

• The proposed method successfully integrates detailed and realistic hair clumping effects.
• Qualitative and quantitative evaluations show superior performance compared to state-of-the-art techniques.
• Generated hair geometry demonstrates improved visual plausibility and realism.

Conclusions:

• The novel hierarchical hair representation and differentiable framework effectively address the limitations of previous single-view hair reconstruction methods.
• The integration of clumping effects significantly enhances the realism of generated hair, paving the way for better rendering and simulation.