A soft image segmentation algorithm based on the generalized fuzzy set is presented in this paper.We incorporate Gibbs random field into the generalized fuzzy set to compensate for the spatial information

and a generalized fuzzy Gibbs random field model is proposed

and the generalized fuzzy Gibbs segmentation algorithm(GFGS) is developed.Each class is considered as a generalized fuzzy class

and the segmented image is regarded as a generalized fuzzy set on the label set in the proposed algorithm.With the proposed algorithm

the outliers in the image data are described by the negative part in the generalized fuzzy membership function

and can be dealt with effectively.Maximum a posteriori(MAP) is used as the statistical segmentation criteria

in which the generalized fuzzy Gibbs random field is used to obtain priori knowledge.Every class center is updated by the centroid of the generalized fuzzy class.Experimental results on both MR real data and the stimulated brain data show that the proposed algorithm is robust

which can filter the noise and partial volume effect significantly.