Sunday, Jan 22, 2006, 11:15-12:15
Room 309
--------------------------------------------------------------------------------
Lihi Zelnik-Manor
Caltech
Title:
Analysis of Dynamic Visual Information
Abstract:
Dynamic scenes captured by a moving camera generate rich and
complex visual
data. To understand the content of
such videos, one needs to be able to
extract the relevant information
from the raw data, separate between its
independent parts and recognize
pieces of similar content. I will start by
presenting a subspace-based
approach to modeling rigid and non-rigid motion
in a video sequence. I will then
show how the suggested representation can
be used for a variety of tasks, including
motion based spatial segmentation,
expression recognition, sequence-to-sequence
synchronization and more. These
applications often require
clustering tools that operate on affinity
relations which are pairwise or of higher order. I will propose new
clustering algorithms and
appropriate data representations accompanied by
empirical results on real video sequences.