Research On Image Segmentation Based On Watershed Algorithm And Graph Theory

This paper research the image segmentation technology based on watershed algorithm and graph theory chiefly.Segment the image with watershed algorithm directly will result in over segmentation,since lack pretreatment and post processing,the effort of image segmentation is not perfect and has a serious problem of over segmentation.The improved algorithm in this article based on mathematical morphology and graph theory,has three steps:the pretreatment of the image,segment the pretreated image with the classical Vincent-Soille watershed algorithm,and post process the image.In the pretreatment, this article propose a kind of generalized morphological filters,which contains two groups of four-structure structuring elements in order to adapt the need of different shape,the size of the groups is 3×3 and 5×5 respectively,the orientation of structuring elements is horizontal, vertical,45 degree and 135 degree,can get rid of the region detail and noise which are prone to cause over segmentation effectively.In the post processing of image,to treat the small regions produces by watershed algorithm as node in a graph,and build a concentrated weighted graph. Use the Ratio Cut rule to direct the incorporation of small regions.Using a image segmentation system designed by VC++6.0 to validate the proposed algorithm.The improved algorithm can eliminate the problem of over segmentation efficiently and gain the significant regions.