Classified road detection from satellite images based on perceptual organization

Extracting roads from satellite images is an important task in both research and practice. This work presents an improved model for road detection based on the principles of perceptual organization and classification fusion in human vision system (HVS). The model consists of four levels: pixels, primitives, structures and objects, and two additional sub-processes: automatic classification of road scenes and global integration of multiform roads. Based on the model, a novel algorithm for detecting roads from satellite images is also proposed, in which two types of road primitives, namely blob-like primitive and line-like primitive are defined, measured, extracted and linked using different methods for dissimilar road scenes. A hierarchical search strategy driven by saliency measurement is adopted in both linking processes. The blob primitives are linked using heuristic grouping and the line primitives are connected through genetic algorithm (GA) evolution. Finally, all of the linked road segments are normalized with centre-main lines and integrated into global smooth road curves through tensor voting. Experimental results show that the algorithm is capable of detecting multiform roads from real satellite images with high adaptability and reliability.