Restoration Method for Spatially Variant Blurred Images
Saïma Ben Hadj, Laure Blanc‐Féraud · 2011
Most of existing image restoration techniques suppose that the blur is spatially-invariant. However, various physical phenomena related to the optical instrument properties make that degradations may change in the image domain. Taking into account space-variance of optical aberrations in the restoration process is an important task that should enhance the accuracy of the estimated object. This latter issue has received little attention by researchers in these last years. In this work, we derive a restoration method for spatially-variant blurred images. In our approach, we consider a blur modeled by a space-varying linear combination of spatially invariant blurs. We develop the example of a piecewise-constant PSF model with regular transitions between areas in order to alleviate blur alteration effect. Furthermore, we develop for this model, an appropriate deconvolution method based on minimization of a criterion with total variation regularization. For this purpose, we fit a domain decomposition-based minimization approach that was recently developed by Fornassier et al., 2009 to the deconvolution problem with a spatially varying PSF model. We thus obtain a fast restoration algorithm where the true image estimation is performed in a parallel way on different sub-regions of the image. We also study the convergence of the proposed method especially for the considered spatially varying PSF model.