An edge-adaptive iterative method for image restoration

Abstract

The paper proposes an edge-adaptive iterative method for restoring a blurred image. An original image is represented as a multiple image model composed of four kinds of edge regions and a flat region. The image is restored with the maximum a posteriori (MAP) criterion, which is realized by the steepest descent technique. The parameters of the original image and edge orientation are estimated from the image temporally restored by a deconvolution method. Computer simulations show that this method suppresses ringing artifacts near sharp edges and avoids smoothing edges in the cases of a motion blur and an out-of-focus blur.<>