Orientation response profile based curvature detection algorithm
Jaeyoung Rhi · 1993
Directional operators can be used to detect the presence of an image edge, to determine the spatial location of the edge, and to measure the edge orientation. We present a structural analysis of the orientation response profile (ORP) of a set of directional operators which shows that the response profile at various locations can be used to measure the curvature of image contours as well as edge location and orientation. The analysis assumes that directional operators of a single spatial scale, but of differing orientations, are applied at each image location. We consider two different types of edges: (1) curved step edges and (2) contour edges of single pixel width. We have developed both a theoretical continuous ORP based algorithm and a discrete ORP based algorithm for curvature detection. The response profile has a particular structure dependent upon the distance from the edge and the type of edge. It is shown to contain edge information such as amplitude, location, orientation, and curvature information which can be used by new local algorithms for detecting orientation and curvature. Even the edge localization of conventional edge detection techniques can be improved by utilizing the ORP data. The curvature detection algorithms can be shown to provide more precise curvature information than previous curvature detection algorithms. In addition the ORP based algorithms utilize the same simple directional operators as commonly used in edge detection, and only one scale of operator is required for measurement within a wide range of curvatures.