Abstract
We present a task-specific information (TSI) based framework for designing compressive imaging (CI) systems. The task of target detection is chosen to demonstrate the performance of the optimized CI system designs relative to a conventional imager. In our optimization framework, we first select a projection basis and then find the associated optimal photon-allocation vector in the presence of a total photon-count constraint. Several projection bases, including principal components (PC), independent components, generalized matched-filter, and generalized Fisher discriminant (GFD) are considered for candidate CI systems, and their respective performance is analyzed for the target-detection task. We find that the TSI-optimized CI system design based on a GFD projection basis outperforms all other candidate CI system designs as well as the conventional imager. The GFD-based compressive imager yields a TSI of (out of a maximum possible for the detection task), which is nearly ten times the achieved by the conventional imager at a signal-to-noise ratio of 5.0. We also discuss the relation between the information-theoretic TSI metric and a conventional statistical metric like probability of error in the context of the target-detection problem. It is shown that the TSI can be used to derive an upper bound on the probability of error that can be attained by any detection algorithm.
© 2008 Optical Society of America
Full Article | PDF ArticleMore Like This
Mark A. Neifeld, Amit Ashok, and Pawan K. Baheti
J. Opt. Soc. Am. A 24(12) B25-B41 (2007)
Pawan K. Baheti and Mark A. Neifeld
J. Opt. Soc. Am. A 26(4) 1055-1070 (2009)
Liang-Chih Huang, Mark A. Neifeld, and Amit Ashok
Appl. Opt. 55(34) 9744-9755 (2016)