A novel image fusion algorithm based on an improved compressed sensing

Abstract

This paper presents a novel image fusion algorithm in an improved compressed sensing(CS). We first develop a multiscale directional sparse image representation by combining the 2D separate wavelet filters with discrete shear filters and assess it to the image reconstruction for CS. Experiments show that the CS system that utilize the new sparse representation is more efficient. Based on the improved CS system, a novel image fusion algorithm is performed on the random projections before reconstruction. Specifically, fuse multiple input measurements which transformed from each of random projections in the nonsubsampled Pyramid transform domain and exploit multiresolution fusion approach to generate an efficient rule called fusion operator. Then combining this fusion operator with scrambling operator from measurement basis to yield composite random projections, and apply them on the improved CS system. The simulations demonstrate that the proposed improved CS-based image fusion algorithm achieves better performance compared with other previous fusion algorithms in CS.