Multiphase Image Segmentation Algorithm Based On Superpixel

Traditional image segmentation is to separate the foreground information and background information,but multiphase image segmentation is that the main part of each image are separated.Multiphase image segmentation algorithms at present are based on artificial interactive mode,and can't adaptively determine the number of classification of regional image segmentation.This paper proposes an adaptive method of multiphase image segmentation that solves the problem of the interactive multiphase image segmentation.Firstly,the new algorithm uses SLIC method to segment image into superpixels with similar size and shape,and then algorithm use the five features of superpixel including three color vectors and two coordinate vectors as the basic processing unit of the latter clustering.The five features of superpixel are clustered by the DBSCAN algorithm with adaptive parameters,by comparing global distance matrix and density function and fitting inverse Gauss density function.The algorithm use maximum likelihood estimation to compute two parameters of the inverse Gauss density function.Finally,the algorithm calculates adaptive threshold of each superpixel,meanwhile the segmentation results are obtained by segmenting the image with an adaptive threshold of each superpixel.The algorithm does not require user interaction and can determine the number of partitions adaptively.In order to verify the effectiveness of the algorithm,the algorithm compares to artificial annotation of the split image in the standard data set of the University of Berkeley(BSDS500).The segmentation speed is faster than other classical algorithms,and the evaluation index is better than other clustering algorithms.