For spectral clustering is applied to image segmentation is difficult to calculate the spectral weight matrix of the actual problem, we have defined the pixel distance between the point and the class is given a sampling theorem, the design of a hierarchical image segmentation algorithm in the use of this algorithm for image segmentation. By adjusting the scaling factor to merge or split a large class of smaller classes, so the image segmentation both randomness but also has multi-scale feature, called spectral clustering based on multi-scale stochastic tree image segmentation (SCMSTIS). The ex...