Research On Methods Of Road Extraction From Remotely Sensed Image

The extraction of road networks from remotely sensed image is an important part in many scene recognition applications. Road network supply strong contextual information which can be used to improve the performance of automatic target recognition systems by directing the search for targets and adjusting target classification confidences. During the past 20 years, a number of semiautomatic and automatic road detection algorithms in remotely sensed images have been developed. Existing methods possess respective merits and drawbacks, without universal adaptability.This thesis considers development of a fully automated road network extraction strategy. We present a new feature fusion strategy, especially implemented to characterize road network from high resolution remotely sensed images. The proposed approach applies"AND"and"OR"rules to the road segments to be compared. At last ,the proposed techniques are applied to a set of two multi-temporal SPOT images of a urban area. Results show that it is possible to fully exploit the potentialities of the two multi-temporal images, by appropriately fusing their information.In addition, several new methods are presented in the road network extraction strategy. The core algorithms implemented include 1) Corner Extractor,2) Line Extractor. The Corner Extractor is developed based on SUSAN(Smallest Univalue Segment Assimilating Nucleus) and MIC(Minimum Intensity Change)algorithm. The Line Extractor enhancement by using corner information which extracted by the new Corner Extractor. Experimental results show these methods are superior to original methods.