Multi-feature fusion for image segmentation based on granular theory

Image segmentation in the big data context is a hot topic in the field of image understanding. Contrary to traditional computing paradigm with precise description of problems, Granular Computing (GrC) is studied by utilizing the toleration of imprecise, incomplete, uncertain and mass information to make systems manageable, robust, low-cost and harmonious. Thus it is an efficient measure to simplify calculation. In this paper, a multi-feature fusion approach based on quadtree and Grc was presented in accordance with the mechanism of human vision. In this technique, firstly original images are reduced into gray images, binary images and quadtree-segmented images, then features are extracted with different granularities from the reduced images respectively, and finally original images are partitioned precisely by the fusion of features according to quotient space theory (QST). Based on the technique of granularity hierarchical and synthesis, this paper gives the example and validation of color image segmentation. Experimental results demonstrate that the algorithm is valid for image segmentation with both speed and accuracy obviously approved compared with common segmentation methods.