【作者】 王大鹏；

【导师】 罗斌；

【作者基本信息】 安徽大学 ， 计算机应用技术， 2011， 硕士

【摘要】 图像分割通常作为图像处理的基础性操作,图像分割结果直接关系到后续更高层的图像处理和计算机视觉工作。水平集方法的出现,给活动轮廓模型带来了强大的生命力,并由此出现了大量经典的轮廓模型。可以通过水平集方法,隐性的描述曲线的演化过程,克服了参数化Snake模型的很多固有缺陷,能够自动处理演化过程中曲线的拓扑结构的变化,这使得活动轮廓模型在应用中得到了很大的拓展。本文将主要研究水平集曲线演化理论在图像分割中的应用,并且研究如何将构造出性能更加优越的新活动轮廓模型。本文可以分为以下3个部分：首先,研究了水平集方法在曲线演化理论中的应用及其算法,并且研究了基于水平集方法的活动轮廓模型的发展及其经典模型。利用水平集方法来隐性的描述曲线的演化,可以自动处理演化中曲线的拓扑结构的变化,由此产生了一些性能优越的几何活动轮廓模型。本文介绍了水平集的数值计算方法,和一些性能较好的水平集快速算法和其初始化算法。详细介绍了一些经典的基于水平集方法的活动轮廓模型,并对其进行了实验,分析了相关模型固有的一些优缺点。其次,研究了基于偏微分方程的图像去噪方法,并将图像去噪方法与基于水平方法的图像分割方法向融合,提出了一个新的图像分割模型。噪声对图像的污染会使得在处理图像分割的过程中产生错误,因此,在本文中研究了在各方面性能都比较优越的基于偏微分方法的图像去噪方法,提出了一个新的基于高阶微分方程的图像去噪方法。并将该去噪方法与水平集图像分割方法相融合,目的是使得在处理图像分割和图像去噪过程中能够相互利用对方处理的结果,提出了一个基于图像去噪和水平集方法相融合的图像分割方法。最后,研究了基于模糊理论的图像分割方法,并将模糊理论应用于基于水平集方法的图像分割模型中,提出了一个带有模糊权值的活动轮廓模型。目的是使得基于区域信息的图像分割方法在处理图像分割时,能够达到全局最优,且能应用于那些类别区分模糊的图像分割。更多还原

【Abstract】 Image segmentation is often used as a basic image processing operation, the segmentation result is directly related to the follow higher level of image processing and computer vision work. The appearance of level set brings strong vitality to active contour model, a lot of classic contour mode also arise from this. The level set method can implicitly describe the curve evolution, it can overcome many inherent defects of parametric Snake Model and automatically deal with the changes of curve topology in evolution, this makes the application of active contour model greatly expanded. This article will study the application of level set curve evolution theory in the image segmentation, and how to construct a new active contour model with more superior performance. This article could be divided into three parts:First, we have studied the level set method’s application and algorithm in the area of the curve evolution theory, and studied active contour models’development and classical model based on the level set model. Using level set method to describe the curve evolution implicitly can automatically handle the changes in its topology, this produces some excellent geometric active contour models. This article describes numerical calculation methods and some good performance fast algorithm and initialization algorithms of the level set method’s. This paper also introduces some classic active contour model based on level set methods in detail, experimentalize on these models, and analyzes some inherent advantages and disadvantages of the related models.Second, we studied the image denoising methods based on partial differential equations, with integrating the image denoising methods with the image segmentation methods based on the level set methods, we proposed a new model for image segmentation. Noise pollution in the image will bring an error in the process of image segmentation, therefore, this paper researched some more superior denoising methods based on partial differential approach, then proposed a new image denoising method based on higher order. This denoising methods and the level set based image segmentation methods was integrated in the paper, the purpose of the integration is to use each other’s results mutually in the process of image segmentation and image denoising, then proposed a segmentation method basing on the methods of image denoising and the level set.Finally, this article researched the image segmentation methods based on fuzzy theory, applied the fuzzy theory to image segmentation model basing on the level set methods, proposed a active contour model with fuzzy weights. The purpose is to make the image segmentation method which based on information of the regional can reach the global optimum and can be applied to segmentation of the image which has some pixels of unclear category.更多还原