The Application Of Level Set Method In Image Segmentation

With the development of computer science and technology,digital image has been widely used each domain.The way of image segmentation by computer to extract image's useful information and analyze objects.Level set methods in the field of image segmentation and computer vision have a wide range of application and achieved a good performance.In this paper,we have a comprehensive analysis and introduction for snake model,GAC model,M-S model,C-V model and LBF model in basic segmentation principle,model construction,and the advantages and disadvantages of each model,moreover,we have some improvements in light of some problems.In conventional level set formulations,the level set functions need to keep signed distance function,the existing models need to reinitialize level set function as a signed distance function.The re-initialization may cause numerical errors and eventually destroy the stability of the evolution.Besides,part of the models are only rely on the homogeneity of the limage intensities,which often fail to provide accurate segmentation results due to the intensity inhomogeneity.To solve these problems,I propose two new image segmentation models combined with the idea of C-V model and LBF model.1.A new four-region level set model:This model is proposed due to the idea of C-V model and have two level set functions are applied,especially,its regularization term is defined with a double potential function,so that the derived level set evolution has a unique forward-and-backward diffusion effect,which is able to maintain a signed distance function during the evolution of level set and eliminate the need for re-initialization.Finally,we show numerical examples and verify the new model can efficiently segment some images with intensity inhomogeneity,moreover,the proposed model have a less time-consu-ming.We efficiently show feasibility and effectiveness of the new model.2.A new LBF model:This model is proposed due to the idea of C-V model and LBF model,it has combined with local and global information to segment image.The model regularization term is still defined with a potential function to ensure level set keep signed distance function.So we establish energy functional expressions of new model.According to the variational level set method,the energy functionals are minimized by Euler-Lagrange equation,and we got the gradient descent equation.Finally,we show numerical examples obtain the satisfactory segmentation results through the inhomogeneity medical images.With the advantages of double potential function,the level set always keep signed distance function and eliminate the need for re-initialization.The experiments verify the feasibility of the new model.