Research On Two-stage Image Segmentation Based On Nonconvex And Convex Approximation And Thresholding

Image segmentation is of great importance in image processing and computer vision.It is to extract the objects of interest from the image according to the characteristics of the image,so as to serve for the higher level of image processing.At present,image segmentation has been widely used in target detection,medical image,remote sensing satellite,and other fields.Over these years,approaches to image processing based on the calculus of variations and partial differential equations have been extensively studied.This dissertation makes improvements based on the Mumford-Shah model,two-stage image segmentation strategy,and K-means method.This dissertation mainly studies the image segmentation method based on the variational model.The main research contents and innovations of this dissertation are as follows:1.A two-stage image segmentation model based on nonconvex Tp V regularization and thresholding is proposed.In the first stage,the Mumford-Shah model is approximated by the variational model based on the L₂-L_p regularization.In this model,the Tp V regularization is used as a priori information to better approximate the empirical distribution of the original image gradient.The split-Bregman algorithm is used to solve the proposed model quickly.In the second stage,a thresholding strategy is used for segmentation.Numerical experiments show that the proposed method is flexible and can be extended to segmentation for blurred images and color images.2.For the first stage of the two-stage image segmentation strategy,a variational model based on the convex and nonconvex coupling is proposed to approximate the Mumford-Shah model.Based on TGp V regularization,convex and nonconvex coupled regularization terms are designed to protect the edge information of the image.The Alternating Direction Multiplier Method(ADMM)is used to solve the proposed model.In the second stage,a thresholding strategy is used for segmentation.Numerical experiments show that the proposed method achieves good segmentation results.3.Based on the K-means method and lifting method,a variational model based on the improved K-means method is proposed.This model uses a combination of ₁L and ₂L regularization to maintain edge information of objects in images while overcoming the staircase effect.The variational model is used to determine the most suitable color for each pixel,which effectively solves the problem of color misclassification.The Chambolle-Pock algorithm is used to quickly solve the proposed model.Numerical experiments show that the proposed method is superior to the existing traditional image segmentation methods.