A Fast Image Denoising Algorithm Based on TLS model and Sparse Representation

Abstract

The sparse representation algorithms (SRs) plays important role to remove the noise from noisy images. The Quantum Particle Swarm Optimization based on the matching pursuit algorithm (QPSO-MP) denoising algorithm that takes advantage of the SRs and the meta-heuristic algorithms has shown hopeful results with respect to noise reduction. Despite effective combination between the meta-heuristic and SRs, the learning dictionary and the fixed populations make these algorithms computational exacting, which largely locates its applicability in many applications. To solve this problem, the fast version of the QPSO-MP denoising algorithm, called FQPSO-MP denoising algorithm, based on a pre learned dictionary and Dynamic-Multi-Swarm (DMS), were proposed in this paper. The pre-learned dictionary produces the best average performance of the denoised images. So, the learned dictionary is no longer required to provide the best average performance of the denoising images. While the denoising approach is applied, the learned dictionary is obtained. So, the execution time of the pre-learned dictionary is shorter than the learned dictionary. The proposed algorithm is compared to the original QPSO-MP and the state of the art algorithms after modifications. These comparisons proof that the PSNR results and the execution time of the proposed algorithm are better than other algorithms.