【作者】 赵于前；

【导师】 桂卫华；

【作者基本信息】 中南大学 ， 计算机应用技术， 2006， 博士

【摘要】 医学图像是反映人体生物组织的复杂图像，图像信息量大，处理困难。医学图像处理是医学图像研究的一个关键环节，对临床医学的应用和发展有着巨大的实用价值。 数学形态学是一种非常实用的图像处理方法，其关键是结构元素的构造。但传统的数学形态学方法结构元素不变，且是对整个图像的一种均衡的处理过程，这就会导致图像的“过处理”或“不及处理”。为此，本文研究了数学形态学医学图像滤波—医学图像边缘检测—医学图像边缘连接—医学图像分割，提出了相关的理论和方法，并通过实验加以验证。本文的主要研究工作和创新性成果包括： (1)提出了基于遗传算法的数学形态学医学图像滤波方法。针对中值滤波的滤波效果将受到噪声强度以及滤波窗口的大小和形状等因素的制约，且滤波窗口形状和大小的设计没有规律可循等局限性；以及形态学开闭滤波器的滤波效果严重地依赖于结构元素的形状和大小，且形态学开闭滤波器具有幂等性等局限性，提出了基于遗传算法的数学形态学医学图像滤波方法。这一方法通过遗传算法对滤波窗口及结构元素进行优化，利用合理构造的适应度函数，求出适应度函数最大值所对应的滤波窗口与结构元素的组合即为所求最优的中值滤波窗口及形态学开闭滤波结构元素。利用这一方法滤波，图像的信噪比得到明显提高。 (2)研究了基于数学形态学的医学图像边缘检测方法，提出了复合型数学形态学医学图像边缘检测算法，以及基于多结构元素与多尺度数学形态学的医学图像检测方法。常用微分算子边缘检测法和传统的数学形态学边缘检测法，不仅提取的边缘模糊，效果差，而且其抗噪声能力很差。论文利用数学形态学膨胀、腐蚀、开闭运算以及形态学梯度等运算的综合特性，提出了复合型数学形态学医学图形边缘检测方法。形态学边缘检测效果的好坏，与结构元素的形状、大小、方向以及尺度有密切的联系，论文接着又提出了多结构元素与多尺度的数学形态学医学图像边缘检测方法。实验结果表明，这些方法提取的医学图像边缘都较微分算子边缘检测方法和传统的数学形态学边缘检测方法有明显的改善。更多还原

【Abstract】 Medical images, which reflect complex and abundant information of human histology, are difficult to process. But medical images processing is a key part to medical images research, and possesses enormous use value to the application and development of clinical medicine.Mathematical morphology, whose most important procedure is to construct structuring element, is one of the very practical image processing methods. But traditional mathematical morphology uses fixed structuring element to process the whole image uniformly, and therefore, results in over-processes or the below-processes the input image.Based on these, in this paper, medical images processing theories and methods based on mathematical morphology are put forward, which mainly include noise filtering, edge detection, edge linking and segmentation of medical image. A lot of experimentations are made to validate the effectivity of the presented methods or algorithms. In this paper, the main reseach work and innovative results are as follows:(1) Mathematical morphological medical images filtering method based on genetic algorithm is put forward.The effect of median filtering is restricted by the size and shape of filtering domain, by the magnitude of noise intensity, and the design of filtering domain has no regular pattern to follow; the effect of morphological Open-Close filtering badly depends on the shape and size of structuring element, and morphological Open-Close filtering have the property of idempotency. To utilize the advantages and overcome the disadvantages of median filtering and morphological Open-Close filtering, mathematical morphological medical images filtering method based on genetic algorithms is put forward. The method optimize filtering domains and structuring elements by genetic algorithm, which constructs reasonable fitness function and computes the maximum value of fitness function to search for the best combination of filtering domain and structuring element. The experimental results show that the method is excellent for更多还原