Image Restoration by Projection onto Convex Sets with Particle Swarm Parameter Optimization

Image restoration is the operation of obtaining a high-quality image from a corrupt/noisy image and is widely used in many applications such as Magnetic Resonance Imaging (MRI) and fingerprint identification. This paper proposes an image restoration model based on projection onto convex sets (POCS) and particle swarm optimization (PSO). For this task, a number of convex sets are used as constraints and images are projected to these sets iteratively to reach restored image. Since relaxation parameter in POCS has a significant effect on restoration results, PSO is developed to find the best value for this parameter to be used in restoration process. The proposed scheme for image restoration is evaluated on three popular images with 4 configurations of noise, compared with 5 competitive restoration models. Results demonstrate that the proposed method outperforms other models in 32 out of 48 cases in images with different noise configurations with respect to relative error, ISNR, MAE and MSE measures.