基于泛函极值的图像分割算法研究

【作者】 李慧芬；

【导师】 赵于前；

【作者基本信息】 中南大学 ， 生物医学工程， 2009， 硕士

【摘要】 本文以基于泛函极值的图像分割算法为研究对象,主要研究泛函极值的求解策略。首先,研究经典的Otsu算法,从运算量上阐释穷举策略对算法的影响,在此基础上提出改进算法,即基于模拟退火算法的多阈值图像分割,这一部分是本文的创新点所在。为了提高Otsu算法的运算效率,提出了两个方法,分别是引入模拟退火算法和引入先验的图像分类信息,使运算量随阈值个数的几何级数增长转化为近似的线性增长。然后研究基于变分法求解泛函极值的图像分割算法,即可变轮廓模型,研究了可变轮廓模型的两个分支,即Snake模型和利用水平集方法的几何可变轮廓模型及两类算法之间的联系。对在性能上具有代表性的两个模型做了具体介绍。其次,针对图像照度不均所致的部分区域误分割问题,研究构建基于互信息的能量泛函的分割方法,对以互信息与类间方差构建的能量泛函的性能进行了分析。本文在Matlab701平台上,对Otsu算法及改进算法的实验结果进行对比分析,对基于互信息的图像分割方法和可变轮廓模型算法的处理结果分别进行分析。更多还原

【Abstract】 This paper chooses algorithm of image segmentation based on functional extreme as its research subject.First of all, the classical Otsu algorithm is taken for the research of algorithm based on Exhaustive Attack searching and functional extreme. Throuth the investigation of a deficiency of speed, the affection from Exhaustive Attack searching to the operation quantity of the algorithm is analysed, and this draw forth the second stage of a research of algorithm based on Intelligence Optimum Algorithm and functional extreme which conclude the originality of this paper. In this stage two tactics is proposed, that is an injection of simulated annealing and also an injection of metempirical classification information of images. This makes a chage for increase of operation quantity as the number of threshold from geometric progression to analogous linear progression. Then the research is carried out with algorithm based on variations, named deformable contour models. Two kinds of deformable model algorithms are studied, and more research are done on two typical models which differ in the function i.In this paper in order to conquer the problem of mis-segmentation of partial region, a method of using mutual information to construct a new functional is proposed.Experiment results of Otsu and the processed algorithm has been discussed on matlab 701, and also the result of the method based on mutual information and deformable contour models are analysed.更多还原