Stable Segmentation of 2D Curves

Abstract

This thesis deals with the specific subproblem of converting sequences of two-dimensional data points into higher level descriptions in terms of curved shape primitives. This is an important member of the wider family of segmentation problems, as rawcurve data are readily available from edge-detection processes, but higher level processes—such as the extraction of geometric invariants or 2D object recognition—require the higher level descriptions. The overriding principle of the work is that if the segmentation process is accurate and repeatable, the later stages of processing are greatly simplified.