Translucency and Photometric Stereo

The photometric stereo method for acquiring surface geometry is only valid for a specific class of materials captured under directional lighting conditions. The ideal material for this method is Lambertian a material that completely and diffusely reflects incident light. Translucent materials, while apparently diffuse, break these restrictions. Due to subsurface light transport (specifically, multiple scattering events of light beneath the surface) the effect of translucency on photometric stereo is an apparent blurring of the estimated surface normals. In this work, we present an empirical study demonstrating this phenomenon and supported by physical validation. These results indicate the relationship between highly subsurface scattering material and poor surface geometry estimation by photometric stereo. We then propose a wavelength-based approach to refinement and present simulation results for a joint optimization that iteratively improves initial estimations of surface geometry and scattering parameters of the material. Ultimately, we hope to extend this optimization strategy for physical samples and generate less noise by prioritizing single scattering information over multiple scattering.