Adaptive Compressed Sensing Image Reconstruction Using Binary Measurement Matrices

摘要

This article introduces an adaptive block-based, low-complexity Compressed Sensing (CS) image recovery algorithm that permits to reconstruct a picture from a reduced number of CS measurements with very high accuracy. Our algorithm first detects the regions of interest based on an initial non-adaptive block-based reconstruction; based on this picture recovery, the subrate is adapted for each block based on its texture content, such that the total acquisition rate does not overpass the target one. Our experimental framework proves that our algorithm outperforms several fixed block acquisition recovery methods, in terms of both image quality, as well as associated complexity.