Research On Image Denoising And Segmentation Based On Gray Level Set

Image segmentation is a truly significant step in the early stage of image processing.Accurate segmentation results can simplify image information,reduce irrelevant data in images and improve the effectiveness of subsequent image processing tasks.In recent years,image segmentation methods based on variational and partial differential equations have developed rapidly,and the introduction of level set has injected new vitality into them.Noise contamination increases the difficulty of image segmentation.The existing segmentation models mainly focus on additive noise images,while the segmentation methods of multiplicative noise images have not been studied in depth.A two-stage fast segmentation model based on gray level set is proposed for images damaged by multiplicative noise in this paper.In the first stage,a new multiplicative denoising model based on variation is proposed to denoise the original noisy image as an auxiliary work of the second stage segmentation.In the second stage,a new segmentation method based on discrete gray level set is proposed in the framework of simplified CV model with only region fitting terms,which can segment the smoothed image quickly.In theory,the gradient descent flow of the new model is deduced by using the variational method,and the existence of the global minimum of the functional is proved by theoretical analysis.The two-stage model does not need to re-initialize the level set and shortens the segmentation time on the basis of guaranteeing the segmentation effect.This paper proposes an improved global segmentation model inspired by the idea of LIC model for multiplicative noise image with intensity inhomogeneity.A convex energy function is obtained by using the simplified gradient descent flow equation and the global convex segmentation method,and then an improved global model is obtained by introducing the edge detection function.It not only retains the characteristics of the local clustering criterion function in the original model,but also does not depend on the selection of the initial contour under the global properties.In numerical algorithm,Split Bregman algorithm is used to minimize the energy function of the improved model and improve the speed of segmentation.Finite difference method is used in numerical realization.Two models are applied to a series of images and compared with other models.The experimental results show that the segmentation results of the two stage model are accurate an d the segmentation efficiency is greatly improved.The improved global model overcomes the limitation of the initial contour,and proves the effectiveness of the segmentation model.