Research On Active Contour Model Based On Target Sign Function For Image Segmentation

Image segmentation is the most basic and important issue in the field of digital image processing.It's the basic prerequisite for visual analysis and pattern recognition.The level set method is a hot topic in image segmentation in recent years.Compared with the traditional image segmentation methods,it has great advantages:(1)The topological structure can be naturally changed by implicit expression of the evolution curve;(2)the evolution of the curve can be translated into a problem of solving partial differential equation,so we needn't to track the evolution of the closed curve;(3)it's easily to extend to highdimensional situations.The evolution model based on the level set method is diverse.This paper improved the distance regularized level set evolution(DRLSE)method for that the DRLSE model is sensitive to the position of initial contour,has slow speed,is easy to fall into the false boundaries,and also easily leaks from the weak edges.The improvements mainly reflected in the following three aspects:(1)Firstly,an adaptive target symbol function is proposed to solve the problem that the traditional DRLSE model is sensitive to the position of the initial contour.Since the evolution curve can only be inwardly contracted or outwardly expanded when using the DRLSE model,the initial contour must completely contain the target or completely within the target.When the initial contour intersects the target boundary,the correct result can never be obtained.The proposed adaptive target sign function can make the part of the evolution curves in the target contour and the part outside the target contour have opposite signs,so that each point on the evolution curve can choose the direction adaptively and evolve toward the target boundaries.(2)Aiming at solving the problem that the DRLSE model is easy to fall into false borders and leaks from weak edges,an edge indicator function is defined.When images have large noise,the evolution curves are easy to stop at the noise when it has not yet reached the target boundary.Furthermore,when the targets have weak edges,it is easy to occur boundary leakage phenomenon.The proposed adaptive edge indicator function can automatically adjust the function curve for different image features,so that the evolution curve can be evolved at an appropriate speed.This improvement is helpful to obtain an ideal result and enhance the robustness of the model.(3)At last,this paper proposed a new double-well function to form a distance regularization term.The new function reduces the evolution speed near the zero potential well,and this is helpful to enhance the ability of detecting weak edges.Furthermore,this function increases the slope near the one potential well,and the model can more efficiently maintain the level set function as the symbol distance function.Moreover,the new potential function solves the problem that the denominator of original function will be zero in some situations,which improves the computational efficiency and stability.The above three improvements are integrated into one model and they form a new adaptive active contour model.The experimental results show that the proposed model can overcome the noise better than other common models when used in natural images.At the same time,it enhances the weak edge detection capability,is insensitive to the initial contour position,has faster speed and higher precision.So,the propose model has stronger robustness.