Research On Image Segmentation Based On Level Set

As an important part of digital image processing,image segmentation has been concerned by many domestic and foreign scholars.On the one hand,specific types of images often have abundant information.Pictures obtained through cameras and other devices may be affected by pollution and uneven light.It is very difficult to manually divide them.This is one of the problems in image segmentation;On the other hand,because the existing image methods are targeted and there is no unified method,the segmentation results have many problems,such as slow segmentation time,inaccurate segmentation,and so on,which is also one of the factors for the slow development of image segmentation.The level set method is widely used in image segmentation due to its easy formation of topological changes,good stability,and rigorous mathematical theory as a support.Using the advantages and disadvantages of the level set,combined with other theories,the purpose of this paper is to improve the stability of the segmentation and reduce the time complexity.Different algorithms are proposed based on the level set method in different contexts.The segmentation method based on the level set is studied.The main contents and innovations are as follows:1.For the problem of difficult segmentation of noise-containing images,this paper proposes a level-noise image segmentation algorithm based on region information.The model firstly improves the kernel function based on the LCK model(Local Correntropy based K-Means),constructs the velocity function based on the local information according to the gradient information of the image,and filters out some noise points contained in the image when performing the evolution curve.,and then combined with the local fitting LIF(Local Image Fitting)model to get the final model.Finally,the segmentation of noise-containing images is compared with LCK model and LIF model.The accuracy of segmentation is much higher than that of LCK model and LIF model.2.For the problem of over-segmentation with weak boundary images,this paper proposes a fast horizontal-set image segmentation algorithm based on bias field.The model first simulates the original image with the bias field theory model,then improves the velocity function in the Regularized Level Set Evolution(DRLSE)model,reduces the evolution time,and then combines the local statistical information of the image with the global information CV.(Chan-Vese)method.Finally,a new energy function is established by combining the image CV model and the image DRLSE model,and the new energy function is embedded into the levelset framework to obtain the final improved segmentation model--Proposed Bias Chan Vese model.The improved velocity function reduces the evolution of time,and the bias field theory corrects the image's deviation information and further simulates the original image.The PBCV model can not only speed up the convergence of the contour curve,but also can deal with those grayscale uneven or blurred images.Finally,the composite image and the real image are segmented.The PBCV model not only has better segmentation effect than other models,but also has higher segmentation accuracy and segmentation speed than other models.3.For the problem of image segmentation interference with complex background,this paper proposes a fast image segmentation algorithm based on saliency region detection and level set.This model proposes a new energy function.First,the optimal salience region of the image is selected according to the saliency model of the cellular automata,which helps to generate the initial boundary curve of the image.Then use the DRLSE model and locally fitted information to construct a new energy function to guide the evolution of the curve.Finally,image segmentation is accomplished by an improved level set model.Experiment with complex background images.Experimental results show that the average time required for image segmentation by the improved model is greatly reduced.Compared with the DRLSE model,the proposed algorithm not only has shorter convergence time,but also has higher segmentation accuracy.