Patch-Based Uncalibrated Photometric Stereo Under Natural Illumination.

Area of Science:

• Computer Vision
• 3D Reconstruction
• Photometric Stereo

Background:

• Photometric stereo typically requires controlled lighting and calibration objects.
• Reconstructing 3D shapes under unknown natural illumination presents significant challenges.

Purpose of the Study:

• To develop a robust photometric stereo method for 3D shape recovery under unknown natural illumination.
• To overcome limitations of existing methods that rely on controlled environments and calibration.

Main Methods:

• Utilizes an equivalent directional lighting model for surface patches with slowly varying normals.
• Employs Markov Random Field optimization and rotation averaging to resolve local and global ambiguities.
• Applies an integrability constraint to refine the reconstructed surface normals.

Main Results:

• The proposed method successfully reconstructs 3D shapes without prior calibration or shape assumptions.
• Achieves results comparable to traditional calibrated photometric stereo methods on synthetic and real-world data.
• Effectively handles unknown natural illumination conditions.

Conclusions:

• The developed photometric stereo approach offers a viable solution for 3D shape recovery in unconstrained environments.
• Demonstrates the effectiveness of patch-based processing and ambiguity resolution techniques.
• Paves the way for more accessible 3D scanning applications using natural lighting.