Adaptive Sensing with Structured Sparsity

Adaptive sensing strategies have been proven to outperform traditional (non adaptive) compressed sensing, in terms of the signal to noise ratios that can be handled, and/or the number of measurements needed to accurately recover a signal of interest. Most existing adaptive sensing schemes for sparse signals, while work well in practice, do not take into account potential structure present in the sparsity pattern of the signal. In this paper, we focus on the Markov tree structure inherent in the wavelet coefficients of signals, and propose an adaptive sampling technique to recover the same. We adopt a simple “follow the scent” strategy, and show that it outperforms traditional non adaptive techniques in practice.