Based Clustering For Image Segmentation Method

There are two trends in the research of image segmentation, the one is seeking new method or new tools to process the image, the other one is looking for the combination of several methods, which can facilitate each other and have more satisfying effects. The paper does the research and exploration on the way of seeking mixed algorithms. The main method is clustering, which is combined with other approach reasonably. These methods combined together can improve each other and make the new segmentation algorithm is more efficient than normal methods. In this paper, clustering and region growing methods were combined to process color image firstly, and then another clustering algorithm, graph algorithm, was introduced. According to the learning of the normal graph methods, especially SWA cut algorithm, an optimized graph algorithm was proposed. Exact organizations are as follows:This paper has two parts: The first part is for color image segmentation. Some normal color image segmentation algorithms were analyzed in this part. Multi-scale clustering and density-based region growing are most important and a new method was proposed based on them. The new approach is image segmentation based on region growing and multi-scale clustering and it was approved to be adaptive same as clustering algorithm and be able to suppress noise same as region method, and it also had a big improvement in multi-object segmentation effect.The second part is clustering based on graph algorithms. Some normal image segmentation based on graph algorithms were studied and analyzed and the most important part is segmentation by weighted aggregation (SWA). Considering the defect of lower efficiency of SWA, we proposed a new algorithm which based on region growth algorithm as coarse segmentation before SWA. The new approach is drove from the pyramid structure of SWA and can reduce the levels of the pyramid in order to decrease the processing time. The experiment results approve that this combination can promote the segmentation speed obviously.