A method of road extraction for high spatial resolution multi-spectral images is presented based on the advantages and disadvantages of Snake model theory.The quaternion expression of multi-spectral image is used to solve the limitation of Snake model for multi-spectral images.The road initial profile is gotten by marked watershed segmentation in quaternion space with multi-spectral images which solve the question of how to set the initial border in the model.At the same time

how to get the approximate road hollow position is solved to realize the road contour extraction by gradient vector flow model with iterative solution.After experiments on QuickBird multi-spectral image with above method

the results indicate that the presented method is feasible for road extraction from high spatial resolution remote sensing image.Comparing with road extraction result by Snake model directly

the method presented in the paper has a better performance in extracting the main road contour information from high spatial resolution multi-spectral images.