Research On Image Segmentation Model Based On The Prior Information And The Level Set Method

Accurate image segmentation is the premise and basis for medical image analysis and disease diagnosis.However,intensity inhomogeneity,multi-tissues interference and blurred boundaries are common problems faced by medical images,which poses a huge challenge to the accurate segmentation of medical images.In order to overcome the above problems,aiming at different characteristics of medical images,this paper establishes different energy functionals,and presents the following three segmentation models to effectively improve the accuracy of medical image segmentation.The intensity of each channel in a color image has unique properties and is often disturbed by the inhomogeneity,which increases the complexity and challenge of color image segmentation.In order to achieve accurate segmentation,this paper integrates the bias field,local region intensity with the level set method to propose an accurate segmen-tation and correction model for color images.Taking the intensity inhomogeneity as the bias filed,the proposed model defines the data fitting term in the energy functional based on the image intensity in all channels.With the edge detection function as the weight,the weighted length term is also defined to minimize the length of the segmentation curve.Benefiting from the weighted length term,the proposed model can get a smoother seg-mentation contour,and makes it easier to detect the target boundary.Combining the data fitting term and the weighted length term,the energy functional is given with the two-phase level set formulation.Then,the split Bregman method is applied to efficiently minimize the energy functional,which significantly improves the computational speed and stability of the proposed model.Applying the proposed model to color medical and natural images,experimental results demonstrate that the proposed model not only can accurately segment color images,but also can eliminate the intensity inhomogeneity in the original images to give the more homogeneous correction images.In addition,quali-tative and quantitative comparison results with traditional segmentation models verify the superiority of the proposed model in terms of the segmentation accuracy,the correction effect and the segmentation speed.What is more,the proposed model is robust to the initial contours and noises.In the tooth image,the boundary of the tooth is very weak,and some tissues around the teeth have the similar intensities to the teeth,such as the alveolar bone and gums,which makes it extremely difficult to accurately extract the tooth boundary.In order to overcome these problems,in this paper,a tooth segmentation model based on the shape prior information is proposed,whose energy functional consists of the image data term,the shape prior constraint term,the length term and the regularization term.Here,the def-inition of the shape prior constraint term is the most crucial step in the proposed model.In detail,the proposed model uses a series of elliptic curves as the shape prior informa-tion to describe the roughly position and contour of the teeth.Then the error between the shape prior information and the segmentation function is defined as the shape prior con-straint term.During the segmentation process,the shape prior constraint term enforces the segmentation curve to move only around the given shape prior information.Not on-ly this,the shape prior constraint term also makes the segmentation contour as close as possible to the shape prior information such that the accuracy of the tooth segmentation can be guaranteed.After that,the steepest descent method is applied to solve the mini-mization problem about the energy functional.Applying the proposed model to the tooth images,experimental results show that the proposed model can accurately segment the tooth boundary without interference from the surrounding tissues,and its segmentation accuracy is significantly higher than other models.In addition,the proposed model is not sensitive to the shape and location of the initial contours.The left and right ventricles,amygdala and hippocampus are important structures in brain magnetic resonance images.Not only do they have similar intensities,but they are adjacent to each other and in contact with each other in the brain,which makes it difficult to accurately segment them by only relying on the intensity similarity.As a consequence,this paper proposes the double level set segmentation model based on the mutual exclu-sion of adjacent region,which can accurately and independently segment two adjacent tissues.In order to avoid the generation of common segmentation regions,the proposed model introduces two level set function,and defines the mutual exclusion term according to the area of the regions jointly segmented by two level set functions to ensure the in-dependence of the adjacent tissue segmentation.Besides,we manually extract the rough contour of the target tissue as a prior information,and define the prior constraint term to ensure only the tissues we want are segmented.Combining the local region data term,the mutual exclusion term,the prior constraint term,the length term and the regularization term,the energy functional is given with the level set formulation,which is minimized by the steepest descent method.Applying the proposed model to brain magnetic resonance images,experimental results show that the proposed model can accurately and indepen-dently segment two adjacent tissues in the brain.Comparison results with other models demonstrate that the proposed model h as higher accuracy than other models in the appli-cation of segmenting adjacent tissu es.Besides,the test experiments with the synthetic images prove that the proposed model can accurately segment inhomogeneous images and is robut to noises.