With applications in both computer graphics and computer vision, surface reflectance functions characterize the appearance of many materials such as brushed aluminum, plastic, velvet, cloth and polished wood.  In addition to a material's texture, these functions also encode its directionally- varying behavior that is responsible for our perception of surfaces being either glossy, dusty, wet, matte etc.

I will describe how matrix factorization provides a flexible framework for representing these high- dimensional functions from a dense set of input measurements.  With this framework in hand, I will first introduce a new data-driven appearance model that allows sampling in the context of Monte Carlo-based global illumination rendering algorithms.  Second, I will describe a tree-structured representation of measured appearance that addresses the challenge of interactively editing a data-driven model. This representation relies on a new suite of factorization algorithms based on linear constrained optimization.  We expect these algorithms to be generally applicable to data dimensionality reduction applications, beyond the task of material representation considered here.

Joint work with Szymon Rusinkiewicz, Ravi Ramamoorthi, Aner Ben-Artzi and Chris DeCoro.

Bio:
Jason Lawrence is currently a PhD candidate at Princeton University working with Professor Szymon Rusinkiewicz.  He received his MS in Computer Science from Princeton in 2003 and a BS in Electrical and Computer Engineering from Carnegie Mellon University in 2001.  His PhD thesis addresses several research problems in computer graphics related to the acquisition and representation of complex appearance data for editing and rendering.
More information can be found at http://www.cs.princeton.edu/~jlawrenc.

Tuesday , March 28, 2006

337 Towne Bldg

3:00pm - 4:15 pm