Abstract: To solve the problems that available imagery segmentation models cannot exploit the plentiful information of image objects fully,and that the global fixed parameters which ignores the role of local statistical properties in feature space,a new approach is proposed,utilizing multiple features and adaptive model parameters.Firstly,multiple features of objects are extracted,and construct the feature space of objects;secondly,through the analysis on the KNN spatial distribution patterns of feature space,the model parameters are adjusted adaptively;finally,the above optimizations are incorporated into multi-scale image segmentation.The results demonstrate that the proposed method is able to make a better use of features and local scale information of feature space,obtains more reasonable segmentation.