The Research Of Natural Scene Image Seg-mentation Based On Graph And FCM

With the rapid development of imaging equipment, the natural scene images are increasingand relevant to people’s lives.Because of the natural scene images have the characteristic of lightuncertain and complex background, the segmentation of natural scene image is difficult andbecome a hot spot in area of image segmentation.This dissertation takes natural image segmentation as the object of the research, researchesinto the popular image segmentation based on graph theory and Fuzzy c-means clusteringalgorithm.The main work is as follows:(1) The purpose of natural scene image segmentation are introduced, and the problem ofimage segmentation which need to be solved are analyzed. Introduces the basic theory of imagesegmentation and the evaluation criteria widely used in image segmentation. Finally, theexperimental results show that this algorithm is effective and practical.(2) The image segmentation algorithm based on graph theory are researched, focuses on thestudy of Graph-Based image segmentation algorithm. For the Graph-Based algorithm which existsunder combined phenomenon, a weighted function based on HSI color space and spatialinformation is proposed, and the improved union-find sets is used to merge the regions.(3) Fuzzy c-means clustering algorithm is analyzed, and an algorithm of the natural sceneimage segmentation combine graph-based with fuzzy C-means clustering (GB-FCM) is proposed.For fuzzy C-means clustering algorithm lack of spatial information and large amount ofcalculation,the graph is divided into subgraph by graph-based algorithm. The amount of clustersare reduced by calculate the cluster centers of subgraph which can accelerate the fuzzy C-meansclustering.For the set of clustering classification number is not optimal necessarily,the oversegmentation may be exist, then a weighted distance based on the color of clustering center andspace information is propose as the merge rules of union-find sets.The improved union-find sets isused to merge the regions of clustering results.Experimental results show that the algorithm basedon graph and FCM is feasibility.Finally,the full text is summarized and prospected．...