Semi-automatic Road Extraction Methods For High Resolution Remotely Sensed Imagery

With the development of science and technology, the acquisition of very high resolution (VHR) remotely sensed imagery is much more easier than before, whose application range has become wider. How to extract objects from VHR remotely sensed imagery automatically or semi-automatically is one of the hottest topics in the areas such as computer vision,pattern recognition etc. Road is one of the most obvious geographic elements in VHR remotely sensed imagery. However,the real automatic road extraction system has not come true so far because of the complexity of the problem itself and the limitations of existing technology level.In this article,we use probability statistics,graph theory and the technology of computer software system to design the two road extraction tool. The concrete research content and results of this paper including the following aspects:(1) Road extraction method based on template matching. We use Kullback-Leibler divergence as the similarity measurement in template matching,which is basd on probability. This algorithm can effectively overcome the noises brought by cars,shadow of trees and buildings on both sides of roads.based on this algorithm, A road semi-automatic extraction tool has been made in which Kullback-Leibler divergence is used as the core algorithm,and it has a good man-machine interaction pattern mode.We carried out road extraction tests on different images and verified that under the same conditions, the KL divergence algorithm has obvious advantage compared to other algorithms.in addition we demonstrated that the efficiency of the operation mode of the road extraction tool.(2)Road extraction method based on the shortest path. Considering the traditional edge detection methods which depend entirely on the edge extraction algorithm itself and the process of the methods cannot be controlled easily,the results of such methods cannot be predicted especially under the complicated conditions.In this article, the shortest path algorithm utilizing the feature of edges of road is applied.This method deals with the problem of controlling and improves the efficiency. Experimental results show that this road extraction method can obtain good results.(3) The system design and tool integration. We have designed two road extraction tool named KL road extraction method and Dijkstra road extraction. When extract complex road networks from VHR remotely sensed imageries,not only the KL road extraction method was used,but also the Dijkstra road extraction method.The extraction experiments show that the results are much more better than simple use one of them.