Abstract: In order to overcome the drawbacks of using either spectral or morphological features for traditional image segmentation methods, a multi-scale image segmentation method using both the spectral and morphological information is proposed. First of all, Differential Morphological Profiles are combined with spectral features to form spectral-morphological characteristics. Then, Hausdorff distance is implemented to calculate the weight of edges based on graph theory and minimum spanning tree algorithm Kruskal is applied to complete the initial segmentation of color images. Finally, the obtained segmentation result is refined by a region merging procedure with the regional heterogeneous criteria proposed in fractal network evolution. Furthermore, object-based Gray Level Co-occurrence Matrix and object-based Pixel Shape Index are proposed on the basis of segmentation results. Experimental results show that the proposed segmentation method is more effective and more efficient than eCognition software 8.0 and Meanshift algorithm. In addition, object-based Gray Level Co-occurrence Matrix and object-based Pixel Shape Index are apparently better than traditional pixel-based methods.