Surface Normals and Light Directions From Shading and Polarization

Area of Science:

• Computer Vision
• Optics
• Material Science

Background:

• Polarization imaging offers surface orientation cues independent of light direction but suffers from ambiguities and poor performance at small zenith angles.
• Photometric stereo excels at disambiguating surface normals with multiple light sources but struggles with large zenith angles and unknown light directions.
• Integrating these methods leverages complementary strengths for robust shape recovery.

Purpose of the Study:

• To develop a unified method for recovering the shape of smooth dielectric objects.
• To overcome limitations of individual polarization and photometric stereo techniques.
• To accurately estimate surface normals, light directions, and refractive indexes.

Main Methods:

• Utilizing diffuse polarization images captured with varying directional light sources.
• Implementing a single optimization scheme incorporating both shading and polarization constraints.
• Combining the strengths of photometric stereo and polarization-based imaging.

Main Results:

• Successfully recovered surface normals across both small and large zenith angles.
• Accurately determined object light directions and refractive indexes.
• Demonstrated robust performance in both simulated and real-world experiments.

Conclusions:

• The integrated approach effectively addresses ambiguities and limitations of existing methods.
• This technique provides a comprehensive solution for non-contact 3D shape recovery of dielectric objects.
• The method offers improved accuracy and robustness in challenging imaging conditions.