Image Segmentation Method Based On Image Structure Information And GVF Snake Model

Image segmentation is a kind of image processing technology,separateing the region of interest and background in the image,and then extracts the target area.As the key and basic techniques,image segmentation has become one of the most important subjects in image processing.Because of the good segmentation performance,active contour model(also called Snake model)is widely used in image segmentation.The gradient vector flow active contour model,or GVF Snake model for short,have been improved the sensitivity of the initial contour in the traditional Snake and promoted the performance of the concave boundary convergence.GVF Snake model as a classical modified model in external force field,has gained tremendous popularity.Firstly,this paper introduces the background and significance of image segmentation,and describes the concept of image segmentation;Then several classical and basic image segmentation algorithms are summarized in this paper.Particularly,the Snake model and GVF Snake model are introduced in detail.GVF Snake model has larger capture range and better segmentation performance than traditional Snake model.However,there are some deficiencies in the GVF Snake model,to solve the inadequate of GVF Snake,two improved models are proposed in the paper.GVF method is inaccurate in capturing the sharp corner of the object during image segmentation.To solve the inadequate of GVF Snake,a new technique of GVF Snake model based on angular point information to extract boundaries of objects having sharp corners is presented.First,we check the real corner points at the edge using curvature-based corner detector,and locally modified GVF force field.Then local angular point force is put forwarded based on corner information.Finally,combining the corner force and the modified external force of the GVF Snake,a new external force of improved GVF Snake is proposed.Experiments indicate that the new algorithm have a better convergence to the sharp corners.Compared with the traditional Snake model,the GVF Snake model has a improvement in the performance of the concave boundary convergence,But for deep concave region segmentation,the GVF Snake model is still difficult to converge to the bottom of the deep concave region,in addition,GVF Snake model is also inadequate in noise robustness and edge protection.To solve these problems,according to the generalized GVF(referred to as GGVF)Snake model,The anisotropic GGVF Snake model based on image structure information(referred to as ISAGGVF)is proposed.Firstly,the image structure tensor is obtained,and then the anisotropic diffusion matrix is constructed according to the image structure tensor;Finally,substituting the anisotropic diffusion for isotropic diffusion terms in GGVF Snake model.Therefore,the diffusion term in the external force field is an adaptive to the image structure;Experiments indicate that the new algorithm have a better convergence to deep concave region and more robustness to noise.