The quest for optimal sensing matrices is crucial in the design of efficient Compressed Sensing architectures. In this paper we propose a maximum entropy criterion for the design of optimal Hadamard sensing matrices (and similar deterministic ensembles) when the signal being acquired is sparse and non-white. Since the resulting design strategy entails a combinatorial step, we devise a fast evolutionary algorithm to find sensing matrices that yield high-entropy measurements. Experimental results exploiting this strategy show quality gains when performing the recovery of optimally sensed small images and electrocardiographic signals.The implementation can be found here at:
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.