IEEE

Talk on Spatiotemporal Gabor Filters for Motion Processing

Dear All,

Department of Electrical Engineering, Indian Institute of Science IEEE Signal Processing Society, Bangalore Chapter and IEEE Bangalore Section invite you to a talk On Spatiotemporal Gabor Filters for Motion Processing

Venue: Multimedia Classroom, Department of Electrical Engineering, IISc

Time and Date: 4 PM, September 7, 2012

Speaker: Dr. Easwar Subramanian
Scientist, TCS Innovation Labs, Hyderabad

Abstract
The talk is on the topic of biologically-motivated visual processing. Specifically, we study operators based on the receptive field (RF) properties of simple and complex cells found in the area V1 of the visual cortex. We consider 2D Gabor filters as models of spatial RF profiles which have applications in edge detection.  Thereafter, we introduce the concept surround suppression and propose models to explain its role in computer vision. We then further our study by considering the temporal aspect of the RF profile to model the direction and speed tuning
properties of simple and complex cells as 3D Gabor filters. These operators are then augmented with surround suppression to model the inhibitory effect of stimuli present outside the classical receptive field. We show that spatiotemporal integration and surround suppression lead to substantial noise reduction. We propose an effective and straightforward motion detection computation that uses the population code of a set of motion energy filters tuned to different velocities. We also show that surround inhibition leads to suppression of texture and thus improves the visibility of object contours and facilitates figure/ground segregation and the detection and recognition of objects.

Biography
Dr. Easwar Subramanian is a scientist in TCS Innovation Labs, Hyderabad. He obtained his PhD from University of Groningen in the topic of dynamical systems and visual processing. Primarily the thesis was focused on two objectives. One objective was to study the dynamics of neuronal networks using coupled oscillators. Specifically, the focus was on understanding the mechanism of unstable attractors and the occurrence of heteroclinic cycles. The other objective was to examine the benefits of spatiotemporal filters and surround suppression in processing image sequences.

He was a postdoctoral-fellow in Ecole Normale Superior-INRIA (France) where he was involved in developing mathematical tools for the estimation of synaptic conductances from voltage recordings of biological neurons. In his current role in TCS, he works on a completely orthogonal area of Mathematical Finance. Apart from his area of work, other professional
interests include visual perception, biological models for vision and mathematical principles associated with neuroscience.

Leave a Reply