Algorithms For Fast Image Segmentation With Convex Relaxed Chan-Vese Models

The research of image segmentation technology focuses on the precision and speed of the segmentation target.At present,there have been a variety of edge extraction and region segmentation models and methods,but many methods still have the problems of low computational efficiency,low segmentation accuracy,poor robustness,and so on.Therefore,the precision and speed of segmentation still need to be researched and improved.Chan-Vese model is a classic segmentation variational model,and multiphase image segmentation is a successful extension of Chan-Vese model.In this thesis,two fast iterative methods are proposed by taking Chan-Vese model as an example.The improved algorithm in this thesis applies to the image segmentation of Chan-Vese model,and it is easy to be applied and extended to other complex models.The research methods in this thesis rely on the convex relaxation method to solve the problem based on energy minimization of the total variation.Taking Chan-Vese model as an example,the classical Chan-Vese image segmentation model is transformed into a convex optimization model through the discrete binary labeling function.The alternative optimization method is used,and the projection method is used for constraints.Finally,the threshold processing is carried out.The main innovations and implementation work are as follows:(1)A variational fast iterative alternating direction method of multipliers(FIADMM)is proposed.Combined with a fast iterative shrinkage-thresholding algorithm(FISTA),this method introduces accelerating variables to accelerate the iterative process of variable and reduce the number of iterations in the process of cross-iteration.(2)Acceleration of chambolle-pock dual method(ACPDM)based on the variational theory is proposed.Based on the dual algorithm framework,combined with the chambolle-pock algorithm,the optimization problem is transformed into the calculation of dual variables and original variables by introducing dual variables,to achieve the speed of the whole iterative process.(3)Two fast segmentation algorithms are applied to the Chan-Vese image segmentation model.In the experiment,one binary labeling function is used for the two-phase image,and several binary labeling functions are used for the multiphase image.At last,through a large number of numerical experiments with multiple sets of images and existing models and the comparison to the traditional method.The experimental results show that the two methods in keeping the image region boundary conditions,the final convergence speed can be increased by more than two times,visually verify the validity of the method,shows that the new method the numerical algorithm is simple,fast calculation speed,segmentation accuracy is higher.