Multiple Level Set Region Based Single Line Road Extraction

Abstract

A new region detection approach for GIS system has been proposed to detect the roads in image. After extraction of road parts from the other parts of image, binarized road features are mined and then used to get a single line road features by thinning. The pruning of the extra unrelated branches and linking of the related branches have been done using Bottom-Top Nega-Max Formulation and Gradient Circum-circle Based Linking. Our proposed method shows high accuracy in detecting roads from satellite images. We have proved by experiment and analysis that proposed method of road extraction is best suited to deal with an average of 95.6% accuracy.