Adaptive Image Restoration Algorithm Based On Total Generalized Variation

Image data acquired in the medical,aviation and communication fields often suffer from hardware limitations and environmental interference leading to quality degradation.Therefore,it is necessary to develop relevant technical methods to solve the typical ill-posed problem of image restoration,and recover potential clear image from degraded image.This paper focuses on the restoration of blurred image with impulse noise.According to the statistical property of impulse noise,the Total Variation(TV)image restoration algorithm based on the L1-norm data-fidelity term can eliminate the blurring effects and impulse noise effectively.However,this algorithm often suffers from the staircase-like artifacts and loses image details resulting in image quality degradation.In order to enhance the quality of restored image,this paper constructs an adaptive variational method based on the Total Generalized Variation(TGV)regularization term for recovering the blurred image with impulse noise,the specific research work is described as follows:Firstly,this paper proposes an image restoration model TGVL1 that integrates the L1 data fidelity term and the TGV regularization term based on the analysis of image structure characteristics.This model makes full use of the advantage of TGV regularizer that can approximate arbitrary-order polynomial function.Therefore,it can protect the important image textures and details.To further improve the quality of restored image,an edge detection operator is introduced into the TGV regularizer to adaptively distinguish the image edge regions and smooth regions.The adaptive TGVL1 model can reduce the diffusion to preserve the image edge features in the image edge regions,while enhancing the diffusion to remove the impulse noise and overcome the staircase-like artifacts in the image smooth regions.Secondly,aiming at the problem that imaging performance is highly dependent on the selection of regularization parameter,this paper proposes a method for selecting spatially adaptive regularization parameter based on image local variance estimation,and constructs a spatially adaptive image restoration model LDTGVL1 based on TGVL1 model.In particular,this model selects small regularization parameter in homogenous regions to enhance denoising;whereas large regularization parameter will be selected in texture regions to preserve image details and features.Furthermore,the validity and stability of LDTGVL1 model are verified by gray and color blurred image restoration experiments with impulse noise.Then,this paper proposes a corresponding numerical solution algorithm based on the Alternating Direction Method of Multipliers(ADMM)to deal with the problem that non-smooth image restoration model is difficult to be solved.The algorithm can decompose the complex image restoration model into multiple subproblems and each subproblem can be alternately solved in the limited iterations.Consequently,it can achieve a stable image restoration effect while guaranteeing the convergence of solution algorithm.Finally,compared with several state-of-the-art image restoration algorithms,the proposed algorithms can achieve satisfactory performance in terms of both subjective visual effects and objective evaluations.