Automatic gene:-:-ation of break point s for minimum description length edge classification
Tony Lindeberg, Meng-Xiang Li · 1995
T his article shows how break points for minimum description length (MDL) based edge segmentation and edge classification can be automatically gener ated. A schem e is proposed where junction candidates are fi rst detected in n mult i-scale p re-processing step , which generates junction candidat es with asso ciated regions of interest. These junct ion features are matched to edges based on spatiai coincidence. For each matched pair, a tentative break p oint is introduced at the edge poin t closest to the junction. Finally, these feature combinations serve a.l; input for an MDL approximation method which tests the validity of the break point hyp otheses and classifies the resulting edge segments as either straight or curved. Experiments on real world image data dem onstrate t he viability of t he approach .