Abstract: The main advantage of image segmentation methods based on MST(minimum spanning tree) is its fast speed,but the segmentation results are often coarse and lack of clear structures.Inspired by Mumford-Shah theory,an improved MST-based segmentation method is proposed in this work.In our method,a good consideration is taken into account about the connectivity of neighboring regions as well as the geometric properties of each region.More specifically,weights between regions are computed and added into the region merging criteria to make the regions with higher connectivity easier to merge.Experiments showed that our method could preserve weak edges between objects and merge regions effectively within the same object,and the contours of resultant segmented regions are simple and smooth.