Bayesian Compressive Sensing of Images Based on Effective Blocking

The traditional reconstruction method of Compressive Sensing (CS) was mostly depended on L1-norm linear regression model. And here we propose Bayesian Compressive Sensing (BCS) to reconstruct the signal. It provides posterior distribution of the parameter rather than point estimate, so we can get the uncertainty of the estimation to optimize the data reconstruction process adaptively. In this paper, we employ hierarchical form of Laplace prior, and aiming at improving the efficiency of reconstruction, we segment image into blocks, employ various sample rates to compress different kinds of block and utilize relevance vector machine (RVM) to sparse signal in the reconstruction process. At last, we provide experimental result of image, and compare with the state-of-the-art CS algorithms, it demonstrating the superior performance of the proposed approach.