关键词: 超像素, 非参数贝叶斯聚类模型, 区域合并, 空间信息

Abstract: According to the common over-segmentation problem of superpixels image segmentation, a new image segmentation algorithm called Mean Shift based Bayesian Region Merging（MS-BRM） based on Mean Shift algorithm and Nonparametric Bayesian Clustering Model（NBCM） is proposed. Firstly, over-segement an image to superpixels using Mean Shift algorithm. Then, a merging criterion based on NBCM and the spatial information of superprixels is proposed, which is used to form the final segment result by merging the superprixels. Experimental results show that MS-BRM algorithm solves the problem of superpixels over-segmentation, and the final segment result of the images reserves many boundary information, which highly matches human vision.

Key words: superprixels, Nonparametric Bayesian Clustering Model（NBCM）, region merging, spatial information