Roads as important artificial objects are the main body of modern traffic system, which provide many conveniences for human civilization. With the development of remote sensing and hyperspectral imaging technology, how to automatically and accurately extract road network from high-resolution multispectral satellite images has become a hot and challenging research topic of geographic information technology. In this paper, an automatic road extraction method from high-resolution multispectral satellite images is proposed by using multiple saliency features. Firstly, road edge is extracted by detecting local linear edge with Singular value decomposition (SVD). Secondly, road regions are constructed by K-means clustering after extracting the feature of background difference. Then road network is achieved by integrating multiple saliency features with Total variation (TV) based image fusion algorithm. Finally, the non-road parts and noises are removed from road network by optimizing multiple salient features with post-processing and morphological operations. The experimental results show that the proposed method can achieve a superior performance in completeness and correctness.