Compression and Adaptation in Imaging

Natural scenes are inherently redundant/compressible due to multi-scale spatio-spectral correlation, which has been very successfully exploited in image compression (e.g. JPEG and JPEG2000). Traditional imaging system design however, ignores this prior information and as a result employs significant resources to acquire redundant information. Compressive sensing has been used in imaging to improve the measurement efficiency typically by employing some version of random projections. In this talk, I will discuss our recent work on an information-theoretic framework for compressive measurement design that exploits additional prior knowledge (e.g. structured sparsity) about natural scenes. Furthermore it incorporates various physical/engineering constraints such as fixed exposure time, measurement noise and quantization, rate-allocation that yields significant performance improvements relative to random projections. I will also discuss how adapting the measurement design to exploit information embedded in past measurements can lead to significant performance improvements. Lastly, I will talk about a scalable compressive imager prototype we have developed to test and validate novel measurement designs and the associate calibration challenge that can limit the performance gains achievable with compressive imaging.

Amit Ashok an Assistant Professor in the College of Optical Sciences and the Department of Electrical and Computer Engineering (ECE) at the University of Arizona and leads the Intelligent Imaging and Sensing Lab (I2SL). He received his PhD and M.S. degrees in ECE from University of Arizona and University of Cape Town in 2008 and 2001 respectively. His research experience spans both industry and academia and his research interests include computational imaging and sensing, physical optics, Bayesian inference, statistical learning theory and information theory. He has made key contributions in task-based joint-design framework for computational imaging and information-theoretic system performance measures such as the task-specific information. With his multi-disciplinary contributions he has been invited to speak at various OSA, IEEE, SIAM and SPIE conferences.