The goal of oversegmentation algorithms is to reduce image dimensionality with minimal loss of information and offer an easy way to use interdependencies of adjacent pixels by partitioning images into groups of superpixels. Unlike natural pictures, it is rather difficult to produce superpixels from very high resolution satellite images due to the complex spatial structures in the images. Distance dependent Chinese restaurant process (ddCRP) can naturally obtain spatial structures of the superpixels through a distance measure and a decay function. Also, ddCRP doesn't need us assign the number of superpixels in advance, because the model could infer the parameter from the observed image data. In this paper, we proposed a new superpixels algorithm based on the distance dependent Chinese restaurant process by combining the spatial distance and the spectral distance to generate superpixels from very high resolution satellite images. The landscape fragmentation and three traditional metrics, i.e., undersegmentation error, boundary recall and achievable segmentation accuracy, are adopted to evaluate the performance of both state of the art oversegmentation methods and the proposed method. The experiments shown that the proposed algorithm outperforms other superpixel methods in nearly every aspects.