Application And Study Of Active Contour Model In Image Segmentation Based On Variational Level Set Theory

The image segmentation method based on level set active contour model is a combination of interpretation and cognition from all kinds of image data,which has attracted extensive attention from researchers because it is closer to the understanding of human visual.Although the level set method made much progress in the theory and applications,it is still in the developmental stage.Therefore,it is necessary to be explored and improved in the future.The classical image segmentation models based on the level set method cannot do well with intensity inhomogeneity and strong noise.Furthermore,this method is easy to fall into local minimum and run in slow speed.In order to slove above problems,the innovative work has been done as following:(1)In order to reduce the effects caused by low contrast and weak boundary on segmentation accuracy of medical image segmentation,a novel geodesic active contour model based on the dynamic combination of local information and global information has been proposed in this paper.A signed pressure function(SPF)has been constructed based on the regional statistics information to combine the local and global information of image,which could replace the edge detection function in GAC model and has improved the ability of evolution curve to recognize the weak boundary in image segmentation.The dynamic adjustment of the weight parameter has been realized so that the effect of global and local terms are more rationally distributed in the different regions in the segmentation process,(the local information plays a leading role when the curve is close to the target region and the global information plays a dominant role when the curve is far to the target boundary).Meanwhile,the proposed model has been extended to multi-phase and it can be successfully applied to the brain MRI image segmentation.We use the Gaussian filter to replace the time-consuming re-initialization process in traditional level set method to regularize the level set function.Experimental results show that the proposed model has strong robustness to the contour initialization and the segmentation accuracy and the running speed have been greatly improved compared with the classical models.(2)The segment results of the traditional active contour model may not be satisfactory when it confronters with the strong noise images and the intensity inhomogeneit images,and the algorithm has low computational efficiency.So some improvements against these defects are made in this paper.The kernel function has been introduced to the active contour model based on the variational level set method and a convex optimal segmentation model based on kernel function and local information has been proposed.Based on assumption of piecewise constant,which is similar to CV model.For each point in a region,a local energy is defined according to the kernel function metric between the intensities of all points within its neighborhood and the intensity average of the region.Then for the whole image domain,a global energy has been defined to integrate the local energy of each point.Due to the introduction of the kernel function metric and the local intensity information,the updating of the region mean values become more accurate and more robust.Thus can overcome the influence of intensity inhomogeneity and noise on the segmentation result,and improve the accuracy of segmentation.In addition,the global optimization technique has been used to obtain the global convex segmentation model and the optimal solution,which overcomes the dependence of the segmentation result on initialization.Finally,in order to improve the computational efficiency,the Split-Bregman method has been used to achieve the fast solution.Experimental results show that compared with the CV model,LBF model,Li model and DRLSE model,the proposed model can overcome the intensity inhomogeneity and has strong robustness to noise and singular values,and can obtain higher segmentation accuracy and faster segmentation efficiency.(3)In order to overcome the problems in the LBF model that was easily trapped in local minimum and could not accurately segment images with complex background,a variational level set image segmentation model has been proposed,which combined the local intensity information of image and the global constraint function.Based on the LBF model,a new global constraint gradient function has been introduced to eliminate the lines or points with large gradient amplitude at the outer region.The combination of the local intensity energy and the global function can lead the active contour curve to reach the target boundary more accurately,thus avoid the energy function trapped in the local minimum and increase the segmentation accuracy and stability.In addition,the level set function regularization term has been introduced in the energy function,which can avoid re-initializing the level set function.The experimental results show that the proposed model can not only effectively deal with low contrast and intensity inhomogeneity,and reduce the effect of noise on image segmentation,but also obtain good segmentation results for medical images and texture images with complex background.