A Coarse to Fine Corner-Finding Method

An iterative algorithm for the detection of corner points on a digital curve is presented. T h e algorithm does not require any input parameters and yields intuitively appealing results for pictures varying in size and orientation. We first calculate the changes in curvature over a supporting arc whose lengtli is either a fraction of the perimeter or given by tlie subarc bordered by two previously found corner points. We further scan the curvature changes, select all the pixels placed where the curve bends are locally significant, and finally mark maxima among the selected pixels as corner points. As the analysis is performed from a coarse to a fine resolution, features of diKerent perceptual importance can be found and accordingly labelled. Some examples are shown in order t o illust ra te the performance of the algorithm.