Abstract: Road extraction from high-resolution remote sensing image is considered as an important and difficulty task. Remote sensing image including complex information, the methods that extract roads by spectral, texture and Linear features have limit. Also, many methods need human-intervention to get the road seeds (semi-automatic extraction). So, they have the great human-dependence and low efficiency. The automatic road-extraction method, which uses the image segmentation based on principle of local gray consistency and integration shape features, is proposed in this paper. Firstly, the image is classified by segmentation and using several shape features, then obtain linear and curve road, so rectified some method’s defect that just only extract linear roads. Secondly, the second step of road extraction is carried out on the region growth, so the road seeds are automatic selected and the road network is automatic extracted. Finally, the extracted roads are regulated by combining the edge information. In experiments, the good symmetrical road surface image and the bad symmetrical road surface were chosen for experiments, and the result proved that the method of this paper is promising.