Abstract
Detection of digital step edge in this paper has been looked both as a mathematical and statistical problem. The step-edge is fitted to actual data but before that existence of an edge on the null hypothesis that there is no edge is tested by small sample statistics tests. This reduces the search space for an edge and eliminates spurious edges. While the edge model is quite simple and well known, a set of directional edge masks have been derived and the local neighbourhood has been treated as a sample from the image population.
Indexing terms: