Multi-points fast marching: A novel method for road extraction

This paper introduced fast marching and minimum path into road extraction from remotely sensed images and proposed a new road extraction method. The work builds on a novel variant of fast marching approach, named as multi-points fast marching. The underlying idea is to calculate minimum action map with multi-points propagating simultaneously. Efficacious saddle points are selected with judgment criterion to remove potential erroneous shortcuts. The extraction roads are continuous line features between couples of starting points that dispense with thinning and broken lines connection. The programming implementation of the method is given and remote sensing imageries of ETM+, IKONOS and airborne images have been taken as case studies. The test images have different spatial resolution, size and road network conditions. The experiment shows that multi-points fast marching could extract roads from 163×161 ETM+ image in 0.10s, 1653×1630 IKONOS image in 21.20s, 1579×1698 airborne image in 20.80s. The results demonstrated that the proposed method can extract multiple roads with high efficiency and little post-processing.