Gradient detection in discrete log-polar images

In a continuous sense, images and gradient operators can be mapped conformally between Cartesian and log-polar (LP) space and the process of gradient detection can be shown to be equivalent. However, gains in the efficiency of processing LP and other retino-cortical image mappings rely on variable sampling with a high resolution fovea and much lower resolution towards the periphery. In this paper, we consider the implementation and evaluation of three strategies for gradient detection in a space-variant, sampled LP geometry. © 2003 Published by Elsevier B.V.