Multi-View Photometric Stereo: A Robust Solution and Benchmark Dataset for Spatially Varying Isotropic Materials

Area of Science:

• Computer Vision
• Computer Graphics
• Material Science

Background:

• Accurate 3D shape and spatially varying reflectance capture are crucial for realistic digital content creation and analysis.
• Existing multi-view photometric stereo (MVPS) methods often have limitations in handling general materials or require complex setups.

Purpose of the Study:

• To develop a robust MVPS technique capable of reconstructing both 3D shape and spatially varying bidirectional reflectance distribution functions (BRDFs) for general isotropic materials.
• To enable precise 3D shape and reflectance capture using a simple, accessible data acquisition setup.

Main Methods:

• A novel MVPS algorithm integrating photometric stereo from a single viewpoint with structure-from-motion across multiple viewpoints.
• Simultaneous inference of basis BRDFs and their mixing weights at each surface point to capture spatially varying reflectance.
• Utilized a simple capture setup comprising a digital camera, LED lights, and an optional turntable.

Main Results:

• Achieved high-precision 3D shape reconstruction with accuracy up to 0.5 millimeters in a studio setup.
• Captured spatially varying isotropic BRDFs with a relative root-mean-square error (RMSE) of 9% under studio conditions.
• Demonstrated the algorithm's effectiveness and usability with both high-precision studio and accessible desktop setups.

Conclusions:

• The proposed MVPS method effectively captures both 3D shape and spatially varying reflectance for general isotropic materials.
• The technique offers a practical solution for high-quality 3D reconstruction and material capture with a simple setup.
• A new benchmark dataset is released to facilitate future research in MVPS.