This paper presents a new strategy that exploits Artificial Neural Networks (ANNs) for a direct selection of edge points from an image. First, the Canny spatial filtering is used to obtain a set of candidate edge points which turn out to be the local maxima of the filtered image. A preliminary smooth selection of these points that exploits neighbourhood information is made to produce a set of pseudo-edges. Some features are extracted from this set and are used to teach an ANN to classify whether or not a point belongs to a real edge. Since the selection works at the pixel level (even if on a strongly reduced subset of the whole image), the generation of training data is easy even with less expert users. Concerning performances, the ANN locally improves edge extraction where significant edges are missed by a selection criterion that is fixed all over the image (e.g. the classical hysteresis selection). The proposed method demonstrates how to provide an easy man-machine interface in those visual sensing systems used in autonomous applications that need a powerful and flexible edge extraction.
Artificial Neural Networks, Edge Extraction, Image Processing, Visual Systems
All articles published by Studies in Informatics and Control (SIC) are published in accordance with the terms and conditions of the Creative Commons Attribution License (CC BY-NC-SA 4.0).