Abstract: Watershed segmentation is a widely used automatic scheme to generate close outlines.It might give rise to closed contour of single pixel width.Using this method the image needs to be divided into many small areaswhich leads to huge workload.This paper proposes a watershed and multi-scale comprehensive image segmentation method.Firstly,calculate the comprehensive gradient which merges the brightness gradient and the texture gradient by using watershed algorithm,then,merge the area of the smallest homogeneity measure (spectral,shape and texture homogeneity metric),and the last,improve region adjacency graph.The experimental result is compared with watershed and region merging image segmentation algorithm,this method proves that it can not only take the advantageof high resolution remote sensing image features in the spectrum,shape,textureand other characteristics,and reduce the computational time.

Key words: image segmentation, watershed algorithm, multi-scalesegmentation, farmland information extraction