DRM-Based Colour Photometric Stereo Using Diffuse-Specular Separation for Non-Lambertian Surfaces.

Area of Science:

• Computer Vision
• Computer Graphics
• Photometry
• Surface Reconstruction

Background:

• Accurate surface orientation estimation is crucial for 3D reconstruction and analysis.
• Non-Lambertian reflectance properties, particularly specularities, complicate traditional photometric stereo (PS) methods.
• Existing PS methods often struggle with surfaces exhibiting complex reflectance behaviors.

Purpose of the Study:

• To develop a robust photometric stereo (PS) method capable of handling non-Lambertian reflectance using color images.
• To improve the accuracy of surface orientation estimation by effectively separating diffuse and specular reflection components.
• To leverage specular highlights for refining surface normal estimates and potentially enabling applications like digital relighting.

Main Methods:

• The proposed method utilizes the dichromatic reflectance model (DRM) with color images.
• It employs a two-step approach: diffuse-specular separation in a specular invariant color subspace and RGB space, followed by surface orientation refinement.
• Specular parameters are initialized via log-linear regression and the DRM is fitted using the Levenburg-Marquardt algorithm.

Main Results:

• The method successfully separates diffuse and specular components, enabling robust surface orientation estimation even with fewer observations.
• The surface normal refinement step, utilizing specular signals, enhances accuracy by approximately 30% on average.
• Validation on synthetic datasets with dielectric materials and comparison against nine other PS methods demonstrate superior performance and validity.

Conclusions:

• The proposed photometric stereo method effectively estimates surface orientations for non-Lambertian surfaces using color images and the DRM.
• The integration of diffuse-specular separation and a novel refinement step significantly improves accuracy and robustness.
• The method's ability to exploit specularities offers potential for advanced applications beyond surface normal estimation.