Improved Block Compressed Sensing of Images by Data Sorting

In Block Compressed Sensing (BCS), the image is divided into small blocks and sampled with the same operator. At the decoder side, each block is treated as an independent sub-CS reconstruction task. This often results in generating some blocking artifacts and losing global structure of the image. In this paper, we propose data sorting into the BCS framework to overcome the BCS problems and improve the reconstruction result. We refer this new block-based CS technique as sort-based BCS (SBCS). In this method, the original image is sorted in such a way that a smooth image is produced. Then, block-based sampling and reconstruction are applied on the smoothed image. We use iterative projected Landweber (PL) and iterative soft thresholding (IST) algorithms for image reconstruction. Simulation results show that the proposed SBCS image reconstruction provides significant improvement over the existing block-based CS image reconstruction methods, in terms of both subjective and objective evaluations.