Research On And Application Of Image Segmentation Algorithms Based On Level Set

With the development of computer science and technology,images have become an important way for people to transmit and obtain information.People are usually only interested in some parts of the image,so the image needs to be segmented to extract the regions of interest.Image segmentation is the basis of image analysis and image understanding,and has been widely used in practical applications,such as medical imaging,remote sensing engineering,meteorological forecasting,intelligent transportation and defense military,etc.Due to the image quality and diversity of image content,image segmentation still faces many difficulties and challenges,such as segmentation of images with intensity inhomogeneity,noise and complex background,etc.As a classic active contour model(ACM),the level set method can obtain smooth and closed segmentation contour with sub-pixel precision,and can handle complex topological changes.Therefore,it has been widely used in the field of image segmentation.However,the level set method still has some aspects that need to be improved,such as sensitivity to the initial contour,easy to fall into the local minimum,and slow evolution of the level set.Based on the level set theory,this dissertation focuses on the segmentation algorithm for images with intensity inhomogeneity,noise and complex backgrounds.In addition,the level set segmentation method is combined with correlation filters and applied to visual tracking.The main contributions of the dissertation are as follows: 1.In order to address the problem of low segmentation efficiency for the image with intensity inhomogeneity,a fast level set image segmentation algorithm with adaptive scale is proposed.Firstly,according to the inhomogeneous image model,a pressure force function based on region information is proposed and utilized to construct a new energy functional.Then,by minimizing the energy functional,the proposed algorithm can segment the inhomogeneous image quickly with the fast numerical implementation strategy and estimate the bias field simultaneously.Meanwhile,a new bias field initialization method is introduced to improve the robustness to the initial contour.In addition,the local variance is utilized to design an adaptive scale operator for the clustering kernel function to estimate the bias field accurately.The proposed algorithm first appears as a two-phase level set segmentation form and then extends to multi-phase segmentation.Experimental results show that the proposed algorithm can segment the inhomogeneous image accurately,robustly and efficiently.2.In order to segment the image with severe intensity inhomogeneity,an adaptive multilayer level set image segmentation algorithm is proposed.Firstly,the local variance is utilized to design an improved global adaptive scale operator and a local adaptive scale operator for the clustering kernel function.With the local adaptive scale operator,an improved local intensity clustering level set segmentation algorithm is proposed,which can correctly segment images with severe intensity inhomogeneity,but is easy to fall into the local minimum.Then,the algorithm is extended to a multilayer level set form.The two designed adaptive scale operators are utilized to adaptively determine the number of layers and the scale parameters of each layer to construct an adaptive multilevel level set structure.By the dual minimization method,image segmentation and bias correction can be achieved simultaneously,which can also avoid falling into local minimal solution.In addition,a hybrid bias field initialization procedure is proposed to enhance the robustness.Experimental results show that the proposed algorithm can accurately segment images with severe intensity inhomogeneity.3.In order to segment the image with both intensity inhomogeneity and noise,a hybrid level set image segmentation algorithm based on kernel metric is proposed.Firstly,an improved multi-scale average filter is proposed to estimate the bias field of the image.With bias field correction,the intensity inhomogeneity of the image can be greatly reduced.Then,the kernel metric method is utilized to construct the energy terms based on the local and global information,respectively.In addition,the local similarity measure method is introduced into the energy terms to suppress the effects of noise.A new weight function is employed to adjust the weight coefficients of the two energy terms to construct a hybrid energy functional.Finally,the level set is regularized by counting gradient regularization to further reduce the effects of noise.Experimental results show that the proposed algorithm can accurately segment images with both intensity inhomogeneity and noise,and has strong robustness to various types of noise.4.In order to address boundary effects and object scale change in the correlation filter tracking method,a scale adaptive visual tracking algorithm based on correlation filters and level sets is proposed.Firstly,a fast level set segmentation method based on region and gradient information is proposed to estimate the object region.Then,the estimated object region is utilized to construct an adaptive weight spatial reliability map to spatially constrain the correlation filter,which can effectively alleviate the boundary effects.Meanwhile,a series of candidate object scales with variable aspect ratios are constructed by using the estimated object region.Finally,the correlation filter response,histogram response and scale variation constraint term are combined to estimate the position and scale of the object.Experimental results show that the proposed algorithm has good tracking performance.