Compressive sensing with a block-strategy for fast image acquisition

Abstract

Compressive sensing is a recent technique that was developed for the reconstruction of large signals from a small number of measurements. It relies on the assumption that the signal to recover is sparse, and the performance of the reconstruction is depending on the level of sparsity. However, in practical case the sparsity of the image to recover is unknown and it is then difficult to estimate the number of measurements necessary to reconstruct the image with a satisfying quality. In this study, we examined a strategy where the image is reconstructed by CS in two steps. A first step with a small number of measurements to estimate the number of points needed, and a second step for the final reconstruction. In addition, we investigated the benefits to create a partition of the image of interest to estimate locally the number of measurements needed for the reconstruction. We demonstrated that our strategy could be used to reconstruct images presenting a PSNR similar to the one obtained with the conventional method, but with fewer measurements.