SymPS: BRDF Symmetry Guided Photometric Stereo for Shape and Light Source Estimation

Area of Science:

• Computer Vision
• Computer Graphics
• Computational Imaging

Background:

• Photometric stereo typically requires known lighting conditions and surface reflectance properties.
• Uncalibrated photometric stereo methods face challenges with unknown isotropic reflectance and non-uniform lighting.

Purpose of the Study:

• To develop uncalibrated photometric stereo methods for accurate shape and light source estimation.
• To address the challenge of unknown isotropic reflectance in 3D surface reconstruction.
• To enable robust surface normal estimation under varying lighting conditions.

Main Methods:

• Introduced the concept of "constrained half-vector symmetry" for general isotropic Bidirectional Reflectance Distribution Functions (BRDFs).
• Developed two surface normal estimation methods based on 1D and 2D symmetry representations.
• Proposed a robust light source estimation technique for uncalibrated scenarios.

Main Results:

• Demonstrated the presence of constrained half-vector symmetry in real-world materials.
• Achieved accurate elevation angle recovery for surface normals with limited light sources.
• Showcased comprehensive surface normal optimization with non-uniformly distributed light sources.
• Obtained state-of-the-art accuracy on synthetic and real-world datasets, including MERL BRDF database.

Conclusions:

• Uncalibrated photometric stereo is feasible and accurate using the proposed symmetry-based methods.
• The methods effectively handle unknown isotropic reflectance and complex lighting.
• The techniques offer a significant advancement in 3D shape recovery from images.