ECE Seminar: Exploiting Saliency in Compressive and Adaptive Sensing
This event has passed.
Friday, December 7, 2012 - 11:45am to 1:00pm
Jarvis Haupt, Ph.D., Assistant Professor, Electrical and Computer Engineering, University of Minnesota
Jarvis Haupt, Ph.D., Assistant Professor, Electrical and Computer Engineering, University of Minnesota. The notion of visual saliency plays a central role in many modern imaging applications. In medical imaging, for example, salient or anomalous regions could indicate areas of pathological significance such as tumors or lesions. The identification of salient regions in an image or video also comprises an essential step in automated surveillance systems, where the goal may be to quickly identify potential anomalies or threats. In an abstract sense, saliency can be interpreted as a generalization of sparsity. Motivated by recent developments in compressive and adaptive sensing, which have established that tremendous improvements in sensing resource efficiency can be achieved by exploiting sparsity in high-dimensional inference tasks, it is natural to ask whether similar techniques can be successfully employed to identify salient regions in an image. In this talk I will discuss results of our recent work on this front. I will describe a novel and computationally efficient saliency-based compressive sensing procedure, and provide a theoretical justification showing explicitly how the performance of the approach depends on the saliency level (the number of ¿interesting¿ regions). Finally, I will demonstrate the performance of our procedure in the context of a two-stage active compressive imaging application.