Numerical Reflectance Compensation for Non-Lambertian Photometric Stereo

Area of Science:

• Computer Vision
• Computer Graphics
• Computational Imaging

Background:

• Photometric stereo relies on Lambertian surface reflectance assumption for accurate surface normal estimation.
• Deviations from Lambertian reflectance introduce significant errors in traditional photometric stereo methods.
• Physics-based modeling of non-Lambertian reflectance is complex and computationally intensive.

Purpose of the Study:

• To develop a robust and efficient method for surface normal estimation in photometric stereo under non-Lambertian conditions.
• To minimize angular errors in surface normal estimation by addressing non-Lambertian effects.
• To improve the reliability and computational speed of photometric stereo.

Main Methods:

• Proposed a numerical compensation scheme to minimize angular error in surface normal estimation.
• Introduced an alternating strategy simplifying angular error formulation by treating reflectance as a known variable.
• Developed a compensation weight reflecting reflectance estimation reliability to mitigate simplification impacts.
• Utilized cosine difference for efficient computation of the proposed scheme.

Main Results:

• Significantly improved performance over state-of-the-art methods on both synthetic and real-world data.
• Achieved state-of-the-art results with a baseline initialization, showing substantial computational savings.
• Demonstrated approximately eight times faster computation compared to existing state-of-the-art methods.

Conclusions:

• The proposed numerical compensation scheme effectively addresses non-Lambertian effects in photometric stereo.
• The method offers a significant improvement in accuracy and computational efficiency for surface normal estimation.
• This approach provides a practical solution for challenging photometric stereo applications.