Blind Image Restoration for Moving Objects Based on Normal Density and Variation Method

Abstract

The blind restoration of a scene is investigated, when multiple degraded (blurred and noisy) acquisitions are available. Distorted frames are modeled as deconvolution process; Finite normal density mixture modeling is performed for each noisy frame. Variation regularizer favors images of bounded variation without penalizing possible discontinuities. Variation approach is further used to piecewise smooth the frames. Variation based image deconvolution/deblurring, which is adaptive in the sense that it does not require the user to specify the value of the regularization parameter. Majaorization Minimization algorithm is used to update the current iteration of the image which satisfies upper bound condition; under mild condition stationary parameters converge monotonically. A quadratic bound function is derived further. The motivation is twofold: first minimizing quadratic functions is equivalent to solving linear systems; second, we do not need to solve exactly each linear system, but rather only to decrease the associated quadratic function, which can be achieved by running a few steps of conjugate gradient (CG).