Deblurring of motion blurred images using histogram of oriented gradients and geometric moments

In image processing, various motion deblurring methods have been proposed that are based on camera shake and object motion. This paper proposes an effective method of estimating the point spread function (PSF) parameters based on the concepts of Histogram of oriented gradients (HoG) and statistical properties of an image, given the blurred image itself. In particular, HoG and statistical properties helps in estimating the angle and velocity parameters of the motion based PSF. Once these parameters are estimated, restoration of the blurred image is performed using recently introduced non-blind method in moment domain. The advantage of using moment domain are fast convergence and robust to the variations in the parameters of the PSF. The experiments on both the synthesized and real images show that our method can effectively restore the original images and get comparable results with existing algorithms. This is validated by using PSNR and a well known quality metrics BRISQUE and SSIM for evaluating the image quality.