Few existing image segmentation methods simultaneously take into account both the distinguishability of different features and the relationship between neighboring pixels.In this paper

a novel image segmentation algorithm is proposed by combining the adaptively weighted features with the Gibbs random field.First

the distinguishability of each component of image features for each land-cover type is defined as a weight parameter

which is determined by the corresponding component of the training samples belonging to the same land-cover type.Second

the initial segmentation is obtained by using the minimum distance classifier

and the spatial correlations of neighboring pixels are modeled by the Gibbs random field.Finally

the label field

which is modeled as the label prior of Gibbs random field

and feature field

which is represented as the normalized weighted distance of weighted features

are combined together to generate the segmentation result.Experimental results demonstrate that the Gibbs random field can effectively describe the spatial relationship

and the adjusted weight can strengthen the distinguishability of the feature component

which can distinguish different land-cover objects accurately.