Shape From Polarization With Distant Lighting Estimation.

Area of Science:

• Computer Vision
• Optical Engineering
• Material Science

Background:

• Surface normal recovery is crucial for 3D reconstruction.
• Existing Shape-from-Polarization (SfP) methods often rely on prior assumptions.
• Polarization imaging offers rich geometric and material cues but is sensitive to lighting.

Purpose of the Study:

• To develop a novel approach for surface normal recovery from polarization images under unknown distant lighting.
• To analytically investigate the role of estimating distant lighting in resolving SfP ambiguities.
• To propose a learning framework that integrates polarization, shading, and normal priors.

Main Methods:

• Utilized a polarimetric Bidirectional Reflectance Distribution Function (pBRDF) based image formation model.
• Developed a two-stage learning framework for estimating reflectance, lighting, and refining surface normals.
• Created a data generation pipeline based on the pBRDF model and a real dataset for evaluation.

Main Results:

• The proposed two-stage framework effectively estimates lighting and reflectance.
• Leveraging normal priors enhances surface normal detail and accuracy.
• The approach demonstrates superior performance over existing methods on real-world data.

Conclusions:

• Estimating unknown distant lighting is beneficial for accurate surface normal recovery from polarization images.
• The proposed learning framework effectively integrates polarization, shading, and geometric priors.
• The method achieves state-of-the-art results in quantitative and qualitative evaluations.