Evolutionary Reconstruction: Image Restoration for Space Variant Degradation
Mohsin Bilal · The Smart Computing Review · 2013
Pseudo-deconvolution is an effective approach to restore space variant degradation (SVD), where each sub-problem is treated as an optimization problem that simultaneously performs image de-blurring and de-noising. De-blurring is an inverse problem primary to visual processing systems. It becomes ill-posed if noise taints the blurry image. Thus the problem is very sensitive to small perturbation in the data. Additive noise has made obsolete classical approaches of inverse filtering and linear algebraic restorations. The restoration formulation for the ill-posed inverse degradations is constrained least square error (CLSE) minimization. Generally, regularization of solutions by smoothness constraint is an addition in the classical approaches to cater to the sensitivity of solutions for small perturbations. In this paper, two well-known evolutionary computation (EC) algorithms: 1) the genetic algorithm (GA) with binary and real encoding schemes, and 2) the particle swarm algorithm (PSO), are proposed to evolve the estimated image in order to obtain an optimal solution for restoration with adaptive regularization. Thus the restoration framework presented in this paper is new and novel, such that it reconstructs the image guided by evolutionary computations. Furthermore, modifications in the initial evolutionary framework are proposed to make it a novel hybrid meta-heuristic approach for real-world restoration applications. Quantitative and visual results of the proposed framework are presented in the paper, with comparative analysis within the EC domain and state-of-the-art methods.