Computer Science Colloquia
Thursday, April 14, 2011
Advisor: Jason Lawrence
Attending Faculty: Worthy Martin, Kevin Skadron, Bill Walker and Todd Zickler
Olsson 228E, 14:30:00
A Ph.D. Proposal
Methods for the Synchronous Acquisition of 2D Shape and Material Appearance
Measuring the 3-dimensional shape and material appearance of physical objects is a challenging task, but has a wide range of applications in areas including cultural heritage preservation, architecture, law enforcement, industry, and entertainment. This thesis extends existing research on acquisition pipelines to enable synchronous measurement of shape and appearance. Previous work that measures each independently suffers from errors due to the difficult alignment and merging problem. We overcome the difficulties of synchronous acquisition by exploiting surface reflectance symmetries and a novel optical design.
We demonstrate the first photometric approach for measuring the shape and appearance of objects with anisotropic surface reflectance. For such materials, both the surface normal as well as tangent direction at each surface point must be recovered, unlike in the case of isotropic materials where the tangent direction can be ignored. We present an optimization strategy to locate vectors of maximum reflectance symmetry using a dense set of images taken under variable point lighting, resulting in a normal and tangent vector recovered at each pixel along with a dense 2D slice of the surface reflectance function.
We present an analysis of sinusoidal illumination in the context of shape and appearance measurement, which shows how to simultaneously perform optical descattering, recover 3D geometry, and acquire reflectance measurements from images taken under phase-shifted sinusoidal lighting. In addition, we show that accurately measuring the geometry of translucent objects using sinusoidal illumination is possible for a wide range of materials and frequencies. These two theoretical results provide a basis for the design of practical scanning systems.