An Iterative Graph-based Image Restoration using Data-Adaptive Objective Function

Abstract

An image is defined as a function of weighted graph encoded with Laplacian matrices and its associated kernel similarities. An Iterative Graph (IG)-based Image Restoration with data-adaptive objective function is used to deblur the images that are degraded due to unconstrained conditions. From
a normalized graph Laplacian, cost function is defined with a new regularization term and new data fidelity term. From the fast symmetry preserving balancing matrix, the normalizing coefficients are derived. This results in determining the spectral properties like symmetric, positive semi-definite and returning zero vector when applied to a constant Image. This algorithm has inner and outer iterations. In the inner conjugate gradient iterations, an updated objective function is minimized and the similarity weights are recalculated with earlier estimate in each outer iterations. The performance of this method is more effective for various restoration problems like sharpening, deblurring and denoising. Experimental results show that IGbased algorithm performs more powerfully in terms of objective criteria and visual quality.