A SCALE-VECTOR APPROACH FOR EDGE-DETECTION

This paper presents a method for selecting proper scale parameters for edge detection. To this aim the image is preliminarily segmented into regions with a homogeneous intensity change model, by using a cluster ing algorithm based on a dissimilarity measure applied on feature vect ors associated to the regions. The used algorithm is a standard cluste ring method, with a reasonable criterion to establish the number of cl usters. Once the image has been properly segmented, in each region the mast proper scale parameter is determined and applied for edge detect ion.