Radon transform based uniform and non-uniform motion blur parameter estimation

Restoration of a single degraded image from uniform velocity motion blurring is recently being studied in number of machine vision based applications including astronomy, medical imaging, consumer level photography, microscopy etc. Identification of motion blur parameters from single blurred image is truly an ill-posed problem, as there results multiple solutions in terms of estimated blur kernel and blurred image combinations, amongst which actual combination is hardly any, that will result into a faithful quality of restored image. The quality of restoration is also highly dependent on the accuracy of point spread function (PSF) kernel estimation. In this paper, we present a Radon transform based motion blur parameter estimation method under both spatial-invariant and variant blur consideration. The experiments performed on simulated motion blurred images show that taking Radon transform of the spectral gradients of blurred images improve estimation accuracy even in presence of noise. Compared with already existing Radon transform based PSF estimation schemes, our method successfully performs PSF estimation even for typical non-uniform motion blurred imagery.