Urban Roads Auto-Extraction Algorithm For High Resolution Remote Sensing Image

Road is the backbone of cityand is one of the most important artificial features. As the basis data in geographic information system(GIS), the update of road information is of great significance for cartography, path analysis and emergency. Remote sensing images become one of most important information in geographic information with the rapid development of space technology, for its wide coverage and high precision. Recently, auto-extraction of interest target from massive remote sensing image is still lagged behind.Road extraction technology is comparative low in the automation level, which mainly rely on manual work. Existing technology of remote sensing processing is confronted with great challenge. At the same time, with the increase of resolution of satellite image, the road characteristics are more complexity. Redundancy texture and detail information produce much interference in road information, which bring more difficulty in road extraction. In the past few decades, although a lot of solution is put forward to extract road from remote sensing imagery automatically, there is still no robust method for road extraction. Therefore, the study of urban road network extraction has important theoretical and practical significance.This paper conduct a research for auto-extraction of road network based on spectral features, shape features and context features of road and proposes a method for auto-extraction of road. The more details are as follows:Firstly, the paper summary the characteristic of urban roads in high resolution remote sensing image based on previous study. The characteristic of high remote sensing image and object-oriented method is introduced, which are the theoretical basis of road extraction. This paper presents an optimized method for watershed algorithm including local homogeneity threshold and region merging, etc, based on the principle and disadvantages of watershed algorithm. Experiments show the proposed method can provide better result in image segmentation. Secondly, several typical geometry factors are used in object-oriented road extraction based on the spectral features, shape features and context features of road. Multi-scale analysis is also used in road network extraction. Meanwhile, road intersection is also one of the most important parts of road network. An approach for auto-extraction of road intersection is proposed in this paper using gray morphological transformation and angular texture signature, in order to obtain the topology information of road network. Binary morphology is applied in post-processing, in order to remove and correct the incomplete results. Topological connection is used in road connection for path-link and road intersection-link. The experimentation shows that the method presented in this paper can extract urban road information efficiently and has fairly accuracy for complex urban context.