A Method Based On Geodesic Distance For Image Segmentation And Denoising

Image is an important information source. We can easily and intuitively catch a lot ofinformation which can’t be expressed by words from an image. So the research for animage is getting more and more attention, and has developed many fields. Imagesegmentation and image denoising are two important fields. Image segmentation is animportant step from image processing to image analysis, and also is the importantcomponent of image understanding. However, during the process of gathering andtransmission, an image is likely to be polluted by noise, which affects the quality of theimage.This paper introduces the present situation of the research of image segmentation, andexplains the defect of using Euclidean distance in image processing. The geodesic distanceis proposed to solve space limitations coursed by Euclidean distance, and the paper gives asimple method which based on image gradient’s magnitude and direction to calculategeodesic distance. In addition, a model of image segmentation, Edge-Weighted CentroidalVoronoi Tessellation (EWCVT) is introduced in this paper. CVT model is special Voronoitessellation whose generators are also the centers of mass (centroids) of the Voronoiregions. The improved EWCVT model in this paper is a kind of image segmentation whichis based on geodesic distance and EWCVT model.In this paper, the CVT model in image segmentation and the definition and theadvantages of geodesic distance have been introduced in detail. Then the paper introducesEWCVT model, and how to apply the geodesic distance to this model to imagesegmentation. The result of the method is compared with the result of other method at lastto show the advantages of this method. And based on the result, the more work plan aboutthis method is given.The focus of the algorithm is combining the geodesic distance with EWCVT model. Theconcept of image gradient simplifies the computing method of geodesic distance. And the combination of geodesic distance and EWCVT model resolves the problem of imagedenoising satisfactorily.