Research On Gray Image Object Segmentation Based On Level Set Theory

Due to gray images' low signal-to-noise ratio,few texture feature and unobvious target area,the traditional image segmentation algorithm cannot get accurate segmentation results for this kind of images.As one of the basic steps of image processing,the results of image segmentation have a direct impact on the reliability of target classification,identification,tracking and other following image processing.Level set image segmentation algorithm has good topological property and the algorithm can combine more image information when segmenting.For different image features,the level set algorithm can achieve the purpose of exact segmentation by combining different energy terms.This thesis puts forward some improvement ideas for the characteristics of gray image within the framework of the level set segmentation algorithm,and specific research content can be expressed as follows:Firstly,this thesis introduces the theoretical basis of level set segmentation algorithm.Three kinds of classical level set segmentation algorithms are introduced and simulated,including level set model based on regional information,level set model based on edge information and level set model based on local information.The advantages and disadvantages of each algorithm in gray image segmentation are analyzed.At the same time,two objective evaluation indexes are presented to provide an objective comparison basis for the improved algorithms.Secondly,we propose a new level set segmentation algorithm based on fuzzy C-means(FCM).In this new level set segmentation algorithm,the membership function is first introduced into level set frame as variables,which avoiding the problem that the original level set depends heavily on step function,impact function and getting more accurate details outline.Moreover,a new regularization term is presented to highlight the outline information of image and increase the degree of segmentation precision.Meanwhile,aiming at the phenomenon of pixel intensity's non-uniform distribution in the magnetic resonance images,we introduce the bias field function to correct the regional average deviation caused by pixel changes of image.Experimental results show that the proposed algorithm can effectively segment the magnetic resonance images of inhomogeneity.Finally,a level set image segmentation algorithm based on probability statistics isproposed in this thesis.It aims at the problem that the level set segmentation algorithm has high computational complexity and slowly convergence speed in high pixel gray image after introducing the bias field function.The proposed algorithm introduces the Bayesian formula to describe the regional distribution of pixels and use the total probability formula to integrate this distribution with the level set algorithm.So the evolutionary distance of the level set function is increased after each iteration,thus the number of iterations of the algorithm is reduced.By segmenting the high pixel image,it is verified that this algorithm can effectively reduce the time consumption.Meanwhile,the segmentation of the brain magnetic resonance image is verified,which proves that the algorithm does not reduce the segmentation accuracy of the original algorithm.