【作者】 周鲜成；

【导师】 申群太；

【作者基本信息】 中南大学 ， 控制理论与控制工程， 2008， 博士

【摘要】 随着数字图像处理技术在军事、医学、遥感、工业生产等领域越来越广泛的应用,图像信息呈现出复杂性和多样性特征,普遍存在着图像信息处理的不完整性、不确定性以及建模困难等问题。因此,智能优化算法在图像处理领域得到了广泛应用,并在某些方面取得了比传统方法更好的效果。近年来,将微粒群算法应用于图像处理领域的研究已取得一定成绩,但在图像分割、图像增强和图像复原等方面仍存在值得进一步深入研究探讨的问题。本文在研究微粒群算法基本理论的基础上,提出了微粒群算法的改进形式,并将微粒群算法和模糊理论、模拟退火算法结合应用于图像处理领域,研究基于微粒群算法的图像模糊阈值分割、图像聚类分割、图像增强和图像复原方法。主要研究工作包括以下几个方面。1.研究了基于微粒群算法的多峰函数寻优问题,提出一种基于峰谷函数的小生镜微粒群算法。算法通过峰谷函数判断小生境子微粒群的生成和合并,产生新的小生境微粒群。该算法克服了初始化参数选取依赖于求解问题先验知识、算法收敛速度慢等缺陷,提高了小生境微粒群算法的多峰函数寻优能力,避免了计算资源的浪费,使算法的寻优效率和收敛速度均有明显改善。2.提出了基于微粒群算法的最大模糊熵阈值分割算法。该算法利用微粒群算法的全局优化能力,依据最大模糊熵原理,搜索模糊参数的最优组合,自适应地确定分割阈值,能应用于单目标、多目标以及信噪比较低图像的分割,具有较强的适应性和较好的图像分割效果,并能大大降低计算的复杂度。3.提出了基于微粒群算法的图像模糊聚类分割算法。根据不同的应用对象,对传统FCM算法的目标函数进行修改,设计了不同的适应度函数,利用捕食者-食饵微粒群算法寻找最优聚类中心,能应用于普通图像、噪声污染图像和彩色图像的分割。提出的算法能克服模糊C均值聚类算法对初始聚类中心敏感易陷入局部最优的不足,提高FCM算法的计算速度。特别是当应用于噪声图像分割时,提出的算法由于既考虑了图像所具有的模糊性,又利用了图像的空间信息,对噪声不敏感,具有抗噪性能好、鲁棒性强等特点。4.提出了基于微粒群算法的图像增强算法。该算法利用Tubbs提出的规则化Beta函数拟合对比度变换曲线,自动寻找Beta函数的最优参数,实现灰度图像对比度的自适应变换;针对彩色图像滤波,通过自适应地获得滤波器窗口的最优权值,体现滤波器窗口内像素之间的空间距离对滤波效果的影响,实现彩色图像脉冲噪声的自适应滤波,其性能明显优于现有的彩色图像滤波方法。5.提出了基于微粒群和模拟退火算法的图像复原算法。该算法利用微粒群算法的快速搜索能力和模拟退火算法良好的全局收敛性能寻找最佳复原图像,克服了传统的图像复原方法存在较多约束条件、依赖先验知识、计算求解复杂和对噪声十分敏感等不足,能应用于不同类型退化图像的复原,并能有效地解决经典维纳滤波算法噪信功率比难以确定的问题。最后,对论文进行了总结,并提出了一些有待于今后进一步研究的问题。更多还原

【Abstract】 Digital image processing technology has been widely used in the fields of military, medicine, remote sensing and industry and so on. Because of the complexity and diversity of image information, there are problems of imperfection, uncertainty and modeling difficulties in image processing field. To solve problems above, intelligent optimization algorithms have been widely adopted, which can achieve a better performance compared with other traditional methods in several respects. In recent years, the research of particle swarm optimization (PSO) has made some progress in the image processing field, but there are still many issues about image segmentation, image enhancement and image restoration worth further studying.On the basis of the research on fundamental theory of the PSO algorithms, an improved particle swarm optimization algorithm is presented in this dissertation. The theory of image fuzzy threshold segmentation, the clustering image segmentation, image enhancement and image restoration are researched based on the PSO algorithms combined with the fuzzy theory and simulated annealing algorithm. Main contributions of the dissertation are shown as following:1. A modified niching PSO based on the hill valley function is proposed. A new niching is formed by utilizing the algorithm which uses the hill valley function to judge whether the niching subswarms are produced and merged . The algorithm has solved the problems of choosing the initial parameters depending on prior knowledge and slow convergence speed . which improves searching ability of multiple solutions, avoids wasting computing resource, and makes significant improvement in searching efficiency and convergence speed.2. An image segmentation algorithm based on particle swarm optimization algorithm and maximum fuzzy entropy is put forward. The algorithm utilizes the global searching ability of the PSO and maximum fuzzy entropy theory, and searches the optimal combinations of the fuzzy parameters. It can adaptively obtain the segmentation thresholds, and can be used to the image segmentation for single, multiple objective and low signal-to-noise image. With the strong adaptability and good result, the algorithm can greatly reduce complexity of the calculations.3. A fuzzy clustering algorithm in image segmentation is proposed based on the PSO algorithm. The algorithm is applied to normal image、noise pollution image and color image segmentation by modifying the objective function of traditional FCM algorithm, establishing the fitness function to different application object and utilizing the predator-prey PSO algorithm to search the optimum clustering center . The algorithm has overcome the defects that the fuzzy C-means clustering algorithm is sensitive to the initial clustering center and involuntary to get into local optimum, so that the calculation speed of FCM algorithm is improved. Especially for noise image segmentation, this algorithm not only takes the fuzziness into consideration but also utilizes the spatial information and accordingly gets the characteristics of insensitive to noise, high anti-noise property and strong robustness.4. Image enhancing algorithm based on the PSO algorithm is proposed. The algorithm can automatically search for the optimal parameter of Beta function to achieve gray image adaptive enhancement by fitting contrast transform curve using regularized Beta function proposed by Tubbs. For color image, the algorithm can get the best weight value of filtering window adaptively, and reflects that spatial distance between two pixels in a filtering window has some influence on the filtering effect. The algorithm can also realize the adaptive filter of the pulse noise and have better performance compared to existing methods.5. A method of image restoration based on PSO and simulated annealing algorithm is proposed. The algorithm has overcome some problems of traditional algorithm such as more constraint conditions, depending prior knowledge, calculation complexity and noise susceptibility, and can be applied into the different restored image, and effectively overcomes the difficulties to determine noise-signal power ratio of Wiener filtering.At last, the research results are summarized, and some issues are raised for the further research.更多还原