基于轮廓波变换的模糊聚类图像分割算法研究

The final result and quality of image analysis and image understanding is decided by image segmentation. Fuzzy clustering algorithm is applied widely because it can retain more of the original image information. But two major drawbacks were found in the fuzzy clustering algorithm, which are (1)It is not only needed to determine the number of classification beforehand but also it is noise-sensitive and is easy in local optimum; (2)Many of the proposed image segmentation algorithms are for the specific study object and versatile segmentation theory has not been proposed by far. It is an urgent problem to improve the quality and efficiency of image segmentation and the versatility of the algorithm. Therefore, this project would focus on the following four scientific problems: (1)The noise immunity performance of traditional clustering algorithm is improved from the angle of transform domain by using contourlet transform; (2)The noise immunity performance of conventional algorithm is improved by introducing context information into it; (3)The problem of bandwidth selecting is solved by use of nearest neighbor algorithm, correlation comparison algorithm and hybrid particle swarm optimization algorithm respectively in mean shift algorithm; (4)The proposed algorithms are used in different types of image and these new algorithms are classified in order to provide the basis for choosing the right image segmentation algorithm and improve the quality and efficiency of image segmentation.

图像分割决定着最终图像分析和图像理解的结果和质量。由于模糊聚类算法能够保留更多的原始图像信息，因此在图像处理中得到了广泛的应用。但是该算法存在以下主要缺陷：(1)需要事先确定分类数，对噪声敏感，易陷入局部极值；(2)现有的图像分割算法大都是针对具体研究对象的，至今还没有通用的分割理论提出。因此提高图像分割的质量和效率以及提高算法的通用性成为亟需解决的问题。为此，本项目重点解决以下问题：(1)从变换域的角度，利用轮廓波变换提高传统算法的抗噪声能力；(2)通过空间上下文信息的引入，解决传统算法仅仅在特征空间中进行聚类，对噪声敏感的问题；(3)利用相关性比较算法、最近邻算法和混合粒子群算法解决经典均值漂移算法的带宽选择问题；(4)将所提算法应用于各种不同类型图像的处理，并对其进行分类，为选择合适的图像分割算法提供依据，从而提高图像分割的质量和效率。

图像分割是图像分析和模式识别中的关键和必要步骤之一，决定着最终的图像分析和图像理解的结果和质量。模糊聚类算法能够保留更多的原始图像信息，因此本项目针对模糊聚类图像分割算法开展研究。将轮廓波变换的方法用于模糊C均值图像分割，从变换域的角度，利用轮廓波变换提高传统算法的抗噪声能力。引入空间上下文信息，从隶属度的改进和目标函数中距离度量的修正两方面入手，提高传统模糊C均值算法的抗噪性能。分别采用最近邻算法、相关性比较算法和混合粒子群算法自适应的选择均值漂移算法的带宽。提出了基于共轭梯度法的快速均值漂移算法，利用共轭梯度算法收敛速度快、算法原理简单的特点，通过交替执行共轭梯度算法和均值漂移算法提高原有均值漂移算法的收敛速度。将所提算法应用于各种不同类型图像的处理，并对其进行分类，为选择合适的图像分割算法提供依据，从而提高图像分割的质量和效率。