Natural images segmentation for patterns recognition using edges pyramids and its application to the leather defects

A node linking edges pyramid is described and implemented for segmentation of natural images. Objects, as defects and marks, can show random features for texture and shape. For thin objects, borders seem to be good features for image segmentation. An edge weighting function is defined for linking nodes of different resolutions. Image segmentation is built from roots by following the inter-resolution links. Objects are fairly represented on the ground for a further identification, and the number of active pixels is highly reduced