Study On Image Segmentation Model Based On Double-functional Level Set Method

Image has become the main way for human to obtain information in modern society,we often use computer to deal with the image.The basic image processing methods including:image transformation,image restoration and reconstruction,im-age compression,image segmentation,image recognition,image registration etc.,image segmentation plays an important role in image recognition and visual analysis.Image segmentation refers that according to the image features such as texture,gray scale,color,shape the image into disjoint parts,so that it shows similarity in the same region and difference in different region.Level set method as a popular one in all the image segmentation method,as a result of the level set evolution can use implicit curve natu-rally said the topology of the image contour,has significant advantages,compared with the traditional image segmentation image of relatively complex has good segmentation effect.Secondly,we can convert the evolution of the level set curve into partial differ-ential equations,with a strong mathematical theory as the support,provide a strong guarantee for image segmentation.we put forward a new model for image segmentation based on the local area model for image segmentation.This paper first:introduces the level set theory knowledge and t,he variational method theory,and then introduces several typical level set image segmentation model,finally,based on image local area information,a new image segmentation model is put forward,the model has a good segmentat,ion effect for the inhomogenous image with weak boundary.In this paper,the main innovation points are as follows:(1)Put forward a double energy function image segmentation model.In order to solve the image gray level uneven phenomenon,in this paper,based on the imaging principle of the image,we improve the image segmentation model based on MI CO model,the first model is called bias field estimation model,through to the member function was improved,and join the two rules,respectively relating to gradient of bias area and the gradient of the members function.This method is compared with MICO method has higher accuracy.The second model is called image segmentation model,by combining image edges and regional information and using the bias area corrected image to segment the image,thus the bias area estimation model use the level set function information through the image segmentation model and the image segmentation model use the bias area corrected image to segment the image,two models depend on each other,restrict each other,improve the accuracy and speed of evolution.(2)In view of the image has weak boundary phenomenon,based on the previous pro-posed speed function,and the result of the bias area estimation model,we proposed a adaptive speed function,so that the evolution has a slowly speed in the weak boundary to avoid over-segmentation phenomenon.In the aspect of numerical solu-tion,in order to improve the accuracy,we use the third order runge kutta method.The experimental results show that the adaptive velocity function of weak bound-ary image has good segmentation results,the iterative steps significantly less than the LBF model.(3)In order to improve the evolution speed,we proposed a adaptive narrow band level set method,because of the evolution of the level set in different positions have different speed around the zero level set evolution,the same width of narrow band reduces the evolution speed,according to gradient information of the zero level set,we improve the width of the narrow band around the zero level set function so that narrow band width varies with the of evolution speed changes at each point on the zero level set.