【作者】 瞿中；

【导师】 杨丹；

【作者基本信息】 重庆大学 ， 计算机应用技术， 2009， 博士

【摘要】 图像分割和图像三维表面重建是图像处理中的两个重要研究领域,现有的图像分割算法在求最佳阈值时存在局限性;图像分割存在重叠、粘连等,对分割质量的评价也非常困难;利用多目标优化求解方法对图像三维表面重建研究也往往找不到最优解。针对这些问题,本文在模糊集理论的基础上,提出用并行搜索求最优解来对图像进行分割,用颗粒参数测量方法对分割图像的质量进行评价,用循环极大极小求解方法来进行图像三维表面重建,具体包括以下内容:①在图像分割中,研究基于阈值的图像分割算法和基于边缘检测的图像分割算法,分析这两类算法的基本思想,针对不同的图像进行了仿真实验,对实验结果进行了分析和比较。本文采用等22.5°角的16个方向的Laplace算子的检测估算模板,设定适当的权向量,提高边缘检测的精度,检测出来的边缘不但更清晰,而且也检测出8个方向的Laplace算子所没有检测出的一些边缘,同时合理地设置参数,避免一些伪边缘的提取。②基于聚类的图像分割算法中,对HCM和FCM算法进行比较,运用FCM聚类算法对图像进行分割实验。由于FCM算法需要初始化,并且目标函数存在许多局部极小点,如果初始化落在目标函数的局部极小点附近,就会造成算法收敛到局部极小点。为了解决此问题,本文提出GOS算法,利用GOS算法的并行搜索能力,对FCM算法加以改进,形成具有并行能力的GOS算法。同时,分析在不同初始条件下,对许多样本的聚类分析时,GOS算法比传统的FCM聚类算法更加有效,对算法性能进行理论分析,并通过仿真实验进行验证。③图像分割质量的评价是最难解决的问题之一,本文提出用图像颗粒参数测量方法来计算颗粒形状和粒度分布等,以此评价图像分割的精确度,并和已有的七种分割评价方法进行比较,通过仿真实验进行定性和定量的分析,证明此方法能更好地评价图像分割算法的优劣。④在图像三维重建中,提出模糊多目标优化图像三维重建的数学模型,将优化图像三维重建中普通目标函数以隶属函数的形式表示,构成新型数学模型来表述问题,采用极大极小求解方法。在大大简化求解过程的同时,有效地保证自动搜寻出原问题的有效解。⑤在医学图像处理中,从受损颅骨图像的背景物质检测、边缘检测,到重建出受损的颅骨图像,对比不同算法,并通过仿真实验,验证算法的可行性;给兔子的血管注射造影剂,重建图像,观察兔子的各种组织,通过仿真实验验证算法的可行性及结果的合理性。⑥在上述理论研究的基础上,本文开发一个图像分割、图像三维表面重建和颗粒参数测量原型系统,能对图像进行处理和分析,特别是将本文提出GOS算法、重建算法和图像颗粒参数测量方法加入该系统,并对该系统的应用前景进行展望。最后,对研究内容进行总结,并指出后续的研究方向,为进一步的研究开拓思路。更多还原

【Abstract】 Image segmentation and image three-dimensional(3-D) surface reconstruction are two important research fields in image processing. There are limitations about existing image segmentation when applying for the best threshold value. There are problems of overlap, adhesion in image segmentation, and also there are some difficulties in evaluation of the segmentation quality. It is often unable to find the optimal solution when using multi-objective optimization method to solve 3-D surface reconstruction of the images. To solve these problems, based on fuzzy set theory, this thesis discuss the optimal solution to image segmentation by parallel searching methods, evaluating the quality of images segmentation method by means of particle parameters, reconstructing the 3-D surface using the circle maximum-minimum solution for image. The whole thesis includes follow parts:In image segmentation, the segmentation algorithms based on threshold and edge detection were discussed in detail. The ideas of the two algorithms were analyzed. The results of the simulation by different images were analyzed, studied, compared. It can improve the accuracy of edge detection by use the estimate pattern of 22.5°angle in 16 direction and set a reasonable parameters, avoid pseudo-edge by improve the operator of Laplace. The improved operator of Laplace not only detected on edge clearer, but also detected more edges which have not been detected before, avoiding the extraction of some pseudo-edge.Based on the clustering image segmentation, algorithms of hard c-means clustering(HCM) and fuzzy c-means clustering(FCM) were compared by means of the FCM clustering algorithm for image segmentation experiments carried out. However, the initialization was needed in FCM algorithm and there were lots of local minimum in the objective function, if the initialization abstained the local minimum vicinity point, it would cause a convergence to local minimum. In order to solve this problem, global optimization search(GOS) algorithm is proposed in this thesis, using the parallel search algorithm GOS ability of the FCM algorithm to improve the capacity of the formation with a parallel algorithm for the GOS. At the same time, the fact that GOS algorithm are more effective than the traditional FCM clustering algorithm is analyzed , the algorithm performance has been analyzed theoretically, and has been verified through simulation experiment. The evaluation of the quality of image segmentation is one of the most difficult problems to be solved. It is proposed in this thesis that it is possible to calculate the parameters of particle shape and particle size distribution and so on by means of particle image measuring method. So that we can evaluate the accuracy of image segmentation, and compare the seven types of segmentation evaluation methods. It has been proved that this method can better evaluate the advantages and disadvantages of image segmentation algorithms through simulation of the qualitative and quantitative analysis.In images 3-D reconstruction, the multi-objective optimization of fuzzy images mathematical model of 3-D reconstruction to optimize the image has been proposed, and the ordinary objective function of 3-D reconstruction is represented in the form of membership functions, and has been described by new type of mathematical model, and so the circle maximum-minimum solution were proposed. The proposed algorithm can not only predigests the seeking process, but also finds the available solution of original problem automatically.The sequence image of damaged skull and rabbit’s thigh were reconstructed in the image 3-D reconstruction. The background material detection, edge detection and reconstruction a damaged skull were studied in this thesis. The simulation experiments verified the feasibility of the algorithm by comparing different algorithms. In order to observe a variety of rabbit’s thigh, the contrast media was injected into the rabbit’s blood vessel. The feasibility of reconstruction algorithm and the reasonableness of results were discussed through the simulation.A system of experimental platform has been developed based on the above research, which is based on image segmentation and image 3-D reconstruction. This system can provide image processing and analysis, especially the GOS algorithm, image reconstruction algorithms and measurement methods of particle’s parameters and so on, at the same time the application of the system of experimental platform is forecasted in this thesis.The thesis ends with the conclusions of the achieved and future research, which extend our thoughts in the future.更多还原