Comparison of canny and V1 neural network based edge detectors applied to road extraction

Abstract

The Anti-parallel edge Centerline Extractor (ACE) algorithm is designed to extract road networks from high resolution satellite images. The primary mechanism used by the algorithm to detect the presence of roads is a filter that detects parallel edges with a specified distance between them. The success of the ACE algorithm thus depends critically on the quality of the edges that are extracted early on in the algorithm, typically using Canny’s edge detector. This paper investigates the viability of an ACE variant that uses a different edge detector, modelled on the primary visual cortex (V1). Considering the experimental evidence, it seems unlikely that the V1-based algorithm is able to produce better results than the original Canny-based algorithm