Corner Detection and Curve Partitioning Using Arc-Chord Distance

Abstract

Several authors have proposed algorithms for curve partitioning using the arc-chord distance formulation, where a chord whose associated arc spans k pixels is moved along the curve and the distance from each border pixel to the chord is computed. The scale of the corners detected by these algorithms depends on the choice of integer k. Without a priori knowledge about the curve, it is difficult to choose a k that yields good results. This paper presents a modified method of this type that can tolerate the effects of an improper choice of k to an acceptable degree. The new algorithm seems to yield generally good results.