Thursday, April 05, 2007
10:00 AM
CSB 601
Chris Pal
U. Massachusetts
Machine Vision, Learning and the Future of Interactive Visual Media
We have traditionally thought of video as a simple stack of rectangular images that are played back linearly in time. Cameras have traditionally been devices for capturing these stacks. Can we create expanded representations of scenes with dynamic, high resolution and 3D content? How does the nature of a camera change when it can use powerful machine vision and machine learning algorithms?

I present some of my answers to these questions, highlighting how machine vision and learning help create future interactive visual media. I show how probability models of imaging functions can expand the dynamic range of cameras and generalize the theory of smoothness regularization. I outline the critical computations required for registering high-definition quality video into huge, high quality panoramic representations and present an efficient solution for registration using adaptive feature matching and sparse matrix methods for optimization. When a camera is translated, it is well known that it possible to extract 3D structure from a scene. I show how random field based models can be used to infer depth in an image and present methods for efficiently optimizing model parameters that are also applicable in many other settings. I then outline how these tasks enable the construction of rich, animated and interactive representations. Specifically, I'll highlight some interactive systems for: real-time panoramic photography, video summarization and interacting with spatially organized information.

Bio: Chris Pal earned his PhD from the University of Waterloo in Canada. Prior to his graduate studies he was with Interval Research in Palo Alto, CA. As a result of his research at Interval he was awarded a patent on audio signal processing and machine learning. During his PhD studies he was also a research assistant at the University of Toronto in the Department of Electrical and Computer Engineering and a part of the Machine Learning group. At Toronto he collaborated closely with the Department of Medical Research on machine vision and learning methods for the analysis of large scale genomics and computational molecular biology experiments. Chris then spent two years at Microsoft Research in Redmond, WA with their Interactive Visual Media Group. His research contributions led to the filing of three patents on computer vision, machine learning and interactive multimedia. He also worked closely with the Windows Media product group. He is presently a Research Scientist in the Department of Computer Science at the University of Massachusetts, Amherst.