Compressive sensing image reconstruction using universal Hidden Markov Tree model

For images, the conventional compressive sensing reconstruction algorithms assume no particular structure aside from the sparsity of them. Although some model-based recovery algorithms take the structure into account, they cost a large amount of computation when computing the model parameters. In this paper, we propose a new recovery algorithm which enables fast and accurate reconstruction of real-world images. We analyze the feasibility of using the wavelet universal Hidden Markov Tree model which fixes the parameters directly to characterize images, and combine iterative reweighted l1 norm minimization with this model. Moreover, we give a novel weighting scheme for reweighted l1 minimization to obtain better reconstruction quality. Experimental results demonstrate that our proposed method outperforms the previously proposed model-based image recovery algorithms as well as conventional recovery algorithms.