Polygonal approximation of contour shapes using corner detectors

Abstract

A great amount of corner detectors that appear in literature are based on using the Freeman chain code of eight directions, which is used to represent contour shapes. We propose a new method for corner detection based on a three-symbol chain code representation, which requires lower storage memory and an easy way to obtain shape corners. We compare it with five existing methods, which are well known in the literature, giving our method a better performance. Furthermore, in order to reconstruct the original shapes through polygonal approximations, we propose an error parameter to quantify the efficiency. This can be accomplished by considering the redundancy of points produced when looking for corners and when computing the difference between the original region and the approximated polygon.