         
|
GRASP Lab Seminar 2003-2004
June 24 (Thursday), 1:45 PM, Levine Hall 307, hosted by Kostas Daniilidis.
Gianfranco Doretto
UCLA Computer Science Department
Video-based Analysis of Dynamic Visual Processes
Abstract: Dynamic visual processes are video sequences of dynamic scenes. Building modeling frameworks to analyze the spatio-temporal information carried by them is one of the most active research areas in computer vision. In this work I consider two classes of dynamic visual processes. The first one contains the video sequences that exhibit some form of temporal regularity, while the second one contains the video sequences that exhibit both spatial and temporal regularity, intended in a statistical sense. Many video sequences of natural scenes belong to one of these two classes, for example sequences of flowing water, smoke, steam, flames, whirlwinds, etc.
For the two classes of dynamic visual processes I will propose models that will not capture the geometry, radiometry and dynamics of the physical scene that generated the images, instead, they will capture the spatio-temporal statistics of the images. For each modeling framework I will also describe a procedure to learn model parameters from training data, and demonstrate the ability of the model to extrapolate new video sequences. Once we have a generative model that can explain the measured video signal, this can be applied to many different problems. For dynamic visual processes that exhibit temporal regularity I will show how the modeling framework can be augmented to perform recognition and segmentation of video sequences. Finally, I will demonstrate how these models can be used for image-based rendering applications, and show how model parameters can be edited to synthesize new video sequences with desired characteristics. This is joint work with Stefano Soatto (UCLA).Biography: Gianfranco Doretto received the D.Eng. degree in Electrical Engineering (with highest honors) from the University of Padua, Italy, in 1998, and the M.S. degree in Computer Science from the University of California, Los Angeles (UCLA), in 2002. From 1996 to 1999 he was with the Department of Electrical and Computer Engineering of the University of Padua, first as a research assistant and then as a research staff. In 2000, he was a visiting researcher with the Electronic Systems and Signals Research Laboratory at Washington University, St. Louis, Missouri, and in September 2000 he joined the UCLA Vision Laboratory where he is currently a Ph.D. candidate in Computer Science. His research interests span several areas of computer vision, with a focus on statistical modeling of video signals. He is also interested in image processing and computer graphics, in particular image-based rendering.
