Abstract: Introduces a newly improved method into scenery image segmentation, based on the CASVFMM (class-adaptive spatially variant finite mixture model) algorithm. By modifying the original potential function to strengthen the dependence of potential function on image pixel features, the segmentation continuity can be kept on with enhanced convergence stability. Thus, the segmented results obviously further conform to the corresponding image regions when the new algorithm is applied to the segmentation of images of different classes. Moreover, the structure expression of potential function is modified to a certain degree so as to accelerate the convergence rate of the algorithm and enhance the reasonableness of the segmented results when the algorithm comes into convergence. The simulation tests for scenery image analysis of the MIT standard image sets reveal comparatively that the newly improved algorithm is more efficient than the original CASVFMM and it is also available to other image analyses.