Event Location: NSH 1507
Bio: Devi Parikh is a Ph.D. student in the Electrical and Computer
Engineering (ECE) Department at Carnegie Mellon University. She is
advised by Tushan Chen, and is part of the Advanced Multimedia
Processing Laboratory. She received her MS in ECE at CMU in 2007, and
her BS in ECE at Rowan University, NJ in 2005. Her research interests
include computer vision, machine learning and pattern recognition. More
specifically, she is interested in exploring the role of context for
enhanced image understanding, and developing unsupervised methods to
learn contextual relationships from images.

Abstract: Our visual world is highly structured, and hence images have significant
contextual information. Traditionally, exploiting this context has been
limited to boosting the performance of higher level tasks such as object
detection. Moreover, context has often been utilized in high quality
images where the need for context is not clear. In this talk, we will
take a closer look at the role of context for enhanced image
understanding. We raise two specific questions. First, when is context
really helpful? We find that context provides improvements in
performances mainly when the appearance information is weak (such as in
low resolution images), which we also verified with human studies. And
second, what tasks can context be leveraged for? We find that it can be
useful for low level tasks as well, such as identifying salient patches
in an image.

This work was done in collaboration with Larry Zitnick at Microsoft
Research (Redmond), the first part of which was recently presented at
CVPR 2008, and the second part will appear at ECCV 2008.