Research On Image Segmentation Algorithm Based On Level Set

Image segmentation is very important in the fields of image understanding.It is the basis part of image understanding and analysis,and has been a hot problem in the field of image processing.Level set method for image segmentation has advantages over highly flexible and topological changes in a natural way and can be implemented by more information which received extensive attention.At present,level set method is still at the stage of development.There have been a number of algorithms,but none of them for any type of images.It's difficult to deal with multi image with intensity inhomogeneous.Therefore,it's necessary to further improve and study in terms of theory and application.In this thesis,level set methods have been deeply investigated,the research focuses on the theoretical model and the application model.Theoretical model mainly focus on interactive selection segmentation.Application model research model focus on local intensity clustering model based on priori shape constraints,and apply it to the disc segmentation.The main works can be summarized as follows:To solve the problem caused by intensity inhomogeneous in multi target images.We proposed interactive selection segmentation model.The GAC model uses image gradient information and has an interactive segmentation property.The CV model uses image region information and has global segmentation property.The GAC-CV model can combine the advantages of the former two models under the correct setting of level set.By introducing an improved symbolic constraint function which specifies a unified level set function,the propose model can efficiently segment the image with intensity inhomogeneous.There are two kinds of segmentation modes:All targets in an image can be segmented in a global way,the intended target can be segmented in a local way.Comparing with the GAC model,the CV model and the GAC-CV model,the proposed model can improve the computational cost and validity of segmentation.It is difficult to segment the optic disk due to its intensity inhomogeneous.In this thesis,a local gray level clustering segmentation model based on priori shape constraints is proposed.The LIC model is a region based segmentation model,which uses the K means principle to estimate the local intensity clustering centers.Local weighted mean intensity of the image is used to combine global information.Therefore LIC model can be used to segment images with intensity inhomogeneous.However,there are numbers of blood vessels surrounding the optic disc.The shape of the disc is easily deformed due to light,lesion,etc.To solve the problem,we introduce the affine transformation of ellipse to LIC model which can guide the evolution of active contour.Compared with the LGDF model and LIC model,the proposed model can greatly improve the accuracy rate on optic disc segmentation.