The traditional segmentation method which is based on local information search technique gives little regard for the global information of the image and ignores the randomness and uncertainty of image segmentation.In view of this

this paper proposes a new segmentation method which is based on cloud model

graph theory and mutual information.Firstly

we could use the cloud model to reflect the uncertainty and randomness when pixel cluster into regions.Secondly

when the graph theory method is introduced into a quasi-optimal cut sets

we could obtain a globally optimal segmentation.Thirdly

by using the multidimensional characteristics which are showed by regional concept of cloud model

we could use a comprehensive heterogeneity measure to improve border weights

and therefore improve the ability to distinguish regional dissimilarity.From the experimental results

the proposed method can produce meaningful

complete and internal-homogeneity divided region

moreover

the segmentation accuracy can meet the basic human visual requirements.