Extraction of dominant points by estimation of the contour fluctuations

摘要

A new method of dominant points (DPs) extraction is presented. For each boundary point, the method assigns a support region whose length is individually limited. Two steps are used to detect DPs. Firstly, points having higher curvature value than the estimated fluctuations over the support region are extracted as candidates. Secondly, adjacent candidates are merged and a final elimination step is performed on each region. Additionally a new error measurement scheme is introduced, which takes into account the error due to the polygonalisation and the deviation brought by each DP. This error is used to compare our algorithm to other methods.