Assessing the behaviour of polygonal approximation algorithms

摘要

In a recent paper, we described a method for assessing the accuracy of polygonal approximation algorithms (IEEE Trans. PAMI 19(6) (1997) 659). Here, we develop several measures to assess the stability of such approximation algorithms under two classes of variations, namely perturbations of the data and changes in the algorithms’ scale parameters. In order to quantify the former, a break-point stability index is defined and tested. For the latter, two indices are introduced; (1) a monotonicity index is applied to analyse the change in the approximation error or the number of line segments against increasing scale, and (2) a consistency index quantifies the variation in results produced at the same scale by an algorithm (but with different input parameter values). Finally, the previously developed accuracy figure of merit is calculated and averaged over 21 test curves for different parameter values to obtain more reliable scores.