VASC Seminar Announcement
       =========================

Date:  Monday, April 15
Time: 3:30-4:40
Place: NSH 1507

Title:  Robust Parameterized Subspace Learning for Computer Vision

Speaker: Fernando De la Torre, Brown University
	
Abstract:

Principal Component Analysis (PCA) and Singular Value Decomposition (SVD)
are two of the most commonly used subspace-related techniques for
dimensionality reduction, filtering, data modeling, etc. In this talk, we
extend PCA/SVD in three ways:

1) PCA/SVD are based on least-squares estimation procedures, therefore
they fail to account for "outliers" which are common in realistic training
sets. We introduce robust subspace learning (RSL) as a new method to deal
with intra-sample outliers.

2) PCA/SVD do not learn a subspace invariantly to geometric
transformations of  the data (problem which typically appears when
gathering visual data). We introduce parameterized component analysis as a
method to learn a subspace invariant to affine or higher order parametric
transformations.

3) PCA/SVD learns the subspace of maximum variation within one training
set. We report preliminary experiments on Coupled Component Analysis (CCA)
a method to learn relations between multiple high dimensional data sets in
presence of limited training data.

Finally, previous subspace visual learning techniques have been applied to
the face tracking and modeling problem. The final part of the talk explores
the use of person-specific facial appearance models for video-conferencing,
 actor animation, visual text to speech, etc.

(joint work with M. J. Black)

Some related papers and results can be seen at
http://www.salleURL.edu/~ftorre

For appointments, please contact Jianbo Shi (jshi@cs).  Fernando will be visiting through 
Wednesday morning.