Feature preserving image sharpening using coupled bidirectional flow

Abstract

In the past decade there has been a growing amount of research concerning partial differential equations in image sharpening. Most of these models indicate edges by a binary zero-crossing decision, however, which will produce a false result with piecewise constant regions, whose textures and fine part are lost. In this paper, we propose a feature preserving coupled bidirectional flow process, where an inverse diffusion is performed to sharpen edges along the normal directions to the isophote lines (edges), while a normal diffusion is done to remove noise and artifacts ("jaggies") along the tangent directions on the contrary. To preserve image features, the nonlinear diffusion coefficients are locally adjusted according to the directional derivatives of the image. Experimental results demonstrate that our algorithm substantially improves the subjective quality of the enhanced images.