Detecting junctions using properties of the Laplacian of Gaussian detector

This paper describes a fast junction detection algorithm using the Laplacian of Guassian detector. We propose a general junction model and analyze its behavior in scale space. This study shows two properties: the Laplacian of Gaussian is zero at any junction and it has one or several elliptic extrema that always lie inside the junction sectors. These extrema move in scale space on lines passing by the junction point. Using these properties, we build an accurate and efficient junction detector. Experimental results confirm, in practice, the efficiency and the reliability of this detector.