Finding parallel structures

Abstract

The world is full of structured objects which appear in images as groups of features that have spatial regularity. By looking for one type of spatial regularity it is possible to design a set of processes that will extract all the parallel structures of any curvature (light or dark ribbons) in images from a number of different domains, such as X-rays of arteries and manufactured objects. The processes start with edge detection and proceed to look for non-accidental relationships between close edges that indicate the presence of a ribbon. Evidence from the early processes is used to infer ribbon extension and also to infer the presence of junctions between ribbons.