Segmentation Models Based On Partial Differential Equation For Images With Intensity Inhomogeneity

Image segmentation is an important research topic in image processing,whose purpose is to extract the objects of interest from images.In recent years,segmentation models based on partial differential equation have been widely used in the field of image segmentation,which are widely concerned by scholars at home and abroad because of many advantages that traditional methods of image segmentation don't have.So far,a large number of segmentation models based on partial differential equation have been proposed.This dissertation discusses the shortcomings of several classical segmentation models based on partial differential equation and the cause of the shortcomings,such as Local Image Fitting(LIF)model,Region-Scalable Fitting(RSF)model,and proposes two kinds of segmentation models based on partial differential equation for images with intensity inhomogeneity.The main work is as follows:(1)Due to the fact that LIF model is sensitive to the size,shape and position of initial contour,we propose a local image intensity fitting model combining global information.Firstly,we construct a global term by means of simplifying LPSM model.Secondly,we linearly combine the global term with the local term of LIF model.Finally,we obtain a novel segmentation model based on partial differential equation.The proposed model is not sensitive to the size,shape and position of initial contour,can effectively segment images with intensity inhomogeneity,and needs less CPU time.In addition,the proposed model achieves fast segmentation of partial real images and synthetic images without initial contour.(2)Due to the fact that RSF model is not robust to initial contour and must select an appropriate intitial contour,we propose a segmentation model for images with intensity inhomogeneity without initial contour.Firstly,we linearly combine RSF model with LPSM model.Finally,we obtain a segmentation model based on partial differential equation containing both local term and global term.The proposed model is robust to initial contour and can effectively segment images with intensity inhomogeneity without initial contour.