Medical Image Segmentation Based On Level Set Method

Image Segmentation as a basic problem of image processing and computer vision is that the image is divided into a series of disjoint homogeneous regions, is to analyze images from a deal, and then completed a critical understanding of the image operation. The mathematical definition as follows: If an image is segmented image I(x, y) is divided into sub-regions I1, I2…IN to meet the following conditions:(1)∪kN=1 Ik (x,y)＝I(x,y),(2)Ik is Connected regions,(3)Ik(x,y)∩Ij(x,y)=Φ(k, j=1,2,…,N),(4)regions Ik meet certain conditions of similarity.In the field of image segmentation,medical image segmentation as one of the important branches of image segmentation,Domestic and interna-tional academic community has been given a high priority.With the clini-cal precision in large numbers and the use of medical imaging equipment and high-speed computer,technical level of medical images in just the past thirty years has made a significant upgrade.CT,Ultrasound,Magnetic res-onance,Digital subtraction,Positron emission tomography and other ad-vanced technologies have been successfully used in clinical,and gradually become a clinical study, diagnosis and treatment of essential and con-ventional means,not only provide an adequate image information data,but also play a role in promoting medical image segmentation for the fur-ther development,at the same time in many biomedical imaging applica-tions has made great success. Medical image segmentation has A wide range of algorithms,for example:Threshold,Clusterig,Based on active con-tour,Combinatorial Optimization,Wavelet,Ridgelet,Curvelet,Neural network,Fuzzy Sets,Morphology,Genetic algorithm and so on.In this ar-ticle.we aimed primarily at the level set based on active contour method.In the second chapter,we detailed discussion curve evolution theory,the principle of the level set method and the theory guarantees,introduced two classical models which are given by the previous,one is based on edge Information model another is based on regional information model,and then we introduced later proposed without re-initialization technique. after analying,summarizing and comparing the work of predecessors,we can find that the two classical models have their own problems:based on edge model have to use gradient information,for some weak edge images,using this model the result of segmentation is not satisfactory. Although based on regional information model can solve the problem of weak edge im-ages,lack of using the important information of gradient,so the first model we proposed in this article is combining the two classical models,get a specific combination of edge and region information without re-initialization the level set method model.Based on regional models only suitable for having two distinct gray and gray smoother internal target and background region images'segment prob-lem,if for the multiple gray scale images of different organs or tissues or within the same region, there are ups and downs of the image gray,it is hard to get the correct result,Literature[3]proposed using local information into level set method model to solve this problem,but this model have to re-initialization the level set function during the iterations,it's obvious influence the calcu-lation of the rate,so the second model we proposed in this article add the constraint item into the model and get the using local information into level set method without re-initialization model which can reduce the computation and improve computing speed.the two models we proposed in this article is as follows: The partial differential equation of combination of edge and region infor-mation without re-initialization the level set method is as follow: The partial differential equation of using local information into level set method without re-initialization is as follow:In chapter three,for the first experimental model we carried out number of experiments.through the experiments,we can proved that the new model algorithm is effective,through a serial of comparative Tests contrasted with the original models,the computing speed had been improved and this model does not depend on the selection of initial level set function,the segment results is also satisfy.For the second experimental model,due to the time limit, we will carry out number of experiments in the future,and I believe that it can achieve the desired results.