Continuous material reflectance map for deep photometric stereo.

Area of Science:

• Computer Vision
• Computer Graphics
• Material Science

Background:

• Photometric stereo aims to recover surface normals from images under varying lighting.
• Sparse lighting conditions pose challenges for traditional photometric stereo methods.
• Neural networks offer potential for handling complex material appearances in photometric stereo.

Purpose of the Study:

• To propose a novel BRDF representation for photometric stereo using sparse lights.
• To optimize the computation of BRDF-based photometric stereo maps.
• To evaluate the proposed method's performance against state-of-the-art algorithms.

Main Methods:

• A BRDF representation based on reflectance maps for sparse lights was developed.
• Optimal computation strategies for BRDF-based photometric stereo maps were investigated.
• The method was trained and tested using numerical rendering, DiliGenT, and custom acquisition systems.

Main Results:

• The proposed BRDF representation effectively handles various surface appearances, including specular and diffuse areas.
• The method demonstrated superior performance in normal map estimation compared to observation maps.
• Experimental results confirmed the effectiveness of the BRDF-based approach.

Conclusions:

• The novel BRDF representation is a significant advancement for sparse-light photometric stereo.
• The proposed method offers improved accuracy and robustness for normal map estimation.
• This work has implications for real-world applications requiring accurate 3D surface reconstruction.