|Stochastic Systems Group|
SSG Doctoral Student
How do you distinguish between metal, plastic, and paper from a photograph? If you know the amount of light incident on the surface from all directions, you can invert the computer graphics rendering process to determine reflectance. If you don't know the illumination, on the other hand, the problem is underconstrained. Different combinations of illumination and reflectance can produce the same image. For example, a chrome sphere reflects the world around it, so if the illumination were just right, it could appear to be a ping-pong ball. Yet, in the real world, humans effortlessly recognize surfaces of different reflectance.
This talk will focus on a computer vision system to recognize surface reflectance properties from a single image under unknown illumination. Our reflectance estimation algorithm succeeds by learning relationships between surface reflectance and certain statistics computed from an observed image, which depend on statistical regularities in the spatial structure of real-world illumination. I will also describe a study comparing the statistics of natural illumination to those of typical photographs ("natural images"). Finally, I will summarize the results of experiments investigating the human approach to reflectance estimation.
This talk covers joint work with Alan Willsky, Ted Adelson, Roland Fleming, and Thomas Leung.
Problems with this site should be emailed to email@example.com