21. UNIFIED SHAPE AND BRDF ACQUISITION BY PHOTOMETRIC STEREO

Department: Computer Science & Engineering
Research Institute Affiliation: Agile Center for Visual Computing
Faculty Advisor(s): David Kriegman | Ravi Ramamoorthi

Primary Student
Name: Zachary Paul Murez
Email: zmurez@ucsd.edu
Phone: 310-913-1490
Grad Year: 2017

Student Collaborators
Matteo Mannino, mtmannin@eng.ucsd.edu

Abstract
Photometric stereo is a common method for 3D reconstruction, but is often limited to Lambertian surfaces. Recently, a few methods have been developed to estimate shape and reflectance, but they require a large number of images. On the other hand, recent work in the graphics community has demonstrated the ability to recover an accurate non-parametric BRDF from only a small number of images if the shape is known (flat or spherical). In this work we unify these two directions by providing a method that simultaneously recovers shape and non-parametric BRDF from a very small number of images. For homogenous BRDFs our method performs well on as few as 4 images, similar to Lambertian photometric stereo, and only requires a couple more to handle general spatially varying BRDFs. Unlike previous work, we simultaneously optimize over the entire normal field using a continuous nonlinear objective. By doing so we are able to include global priors that allow us to share information between pixels which is crucial for constraining the problem when only a few images are available. We present extensive results on synthetic and real data.

« Back to Posters or Search Results


Contact:   researchexpo@soe.ucsd.edu   (858) 534-6068