【作者】 王保平；

【导师】 谢维信；

【作者基本信息】 西安电子科技大学 ， 信号与信息处理， 2004， 博士

【摘要】 图像处理是图像工程的低层次操作，其处理效果对图像的高层次操作，即图像分析和计算机视觉的影响很大。但由于图像处理问题本身的复杂性和学科交叉性，几十年来该问题一直是研究的热点。 图像本质上具有模糊性，这是由于(1)三维目标投影在二维图像平面上带来的信息丢失。(2)定义边界、区域和纹理等图像特征时存在模糊性。(3)对图像底层处理结果的解释带有模糊性。因此，模糊信息处理技术在图像处理中的使用有其内在的合理性和必然性。 基于上述原因，本文提出了许多基于模糊信息技术的图像处理新方法和新思想，并取得了良好的效果。本文共分为七章，主要内容如下： 第一章为绪论，阐述了图像处理是计算机视觉和机器智能研究中的基本问题，论述了模糊信息技术在解决图像处理问题时有其必然性和合理性。最后列出了本文取得的主要研究成果。 第二章简要介绍了和本论文相关的模糊数学知识，作为本论文的数学预备。 第三章提出了一些图像空域增强算法，即几种局部运算方法。该章对图像中不同类型加性脉冲噪声的去噪问题进行了系统地研究和分析，提出了三种基于模糊信息处理的图像平滑去噪滤波器，分别为基于模糊熵的多值图像去噪滤波器、基于直方图的图像椒盐噪声滤波器、基于直方图和区域信息的自适应图像脉冲噪声滤波器。实验表明这些方法滤波效果优于传统的滤波器和其它模糊滤波器。 第四章研究了图像空间域增强的另一类方法，即点运算法。提出了一种新的基于模糊信息处理的自适应图像灰度级修正法。新方法首先通过模糊熵的引入，在图像灰度级修正算法中利用了图像的邻域统计信息、模糊信息和人眼视觉特性；其次利用灰度值的统计特性达到了图像自适应增强的目的；接着补充了一个有用的非线性变换，增加了这类算法的普适性：最后解决了原算法在实现图像增强时出现的灰度值过调问题。实验结果表明新算法不仅能够有效地增强整个图像的对比度，提高图像的视觉效果，而且能够在增强图像边缘的同时有效地抑制噪声。 第五章提出了一种基于模糊信息处理的图像变换域增强算法。该章给出了一种性能更好的模糊增强变换算子，并将该算子成功地引入到多层次图像模糊增强算法中，取得了较好的效果；另外，新算法针对以往算法需要人工介入设置阈值参数方可进行图像增强的缺点，通过模糊熵的引入，可以自动地选取最佳的阈值参数而不需人为的介入，因而成功的解决了这一难题；最后，将新的多层次模糊增强算法应用于图像边缘检测中，取得了优于现有模糊增强方法的效果。 第六章通过模糊嫡的引入，构造出了几个边缘检测性能更好的信息测度，利用输入图像提取这些信息测度，组成一个反映图像边缘特征的数据集，接着用加权FCM聚类算法将该数据集分为两类，即边缘点数据和非边缘点数据，从而达到图像边缘检测的目的。该方法无需确定阐值;对弱边界检测较敏感。 第七章对受噪声干扰的图像、目标和背景相互交迭的图像、边界不明确的图像等降质图像的分割问题进行了较深入的研究，并考虑到这类图像中所具有的不确定性往往本质上是模糊性的特点，合理地将模糊信息处理技术引入到降质图像的分割中去，取得了较好的效果。关键词:图像处理模糊信息处理图像增强图像分割边缘检测更多还原

【Abstract】 Image processing is low level operation in image engineering, it plays an important role in image analyzing and computer vision. Because of image processing problem possesses high complexity and the intercross characteristic of subject, the problem has not been good solved for several decades.It occurs information loss when the three dimensions target is projected on the two dimensions image plane; There is fuzzy character in defining boundary, region and texture of image; It brings fuzzy character that explaining the result of the low level image processing. Thus the use of fuzzy information processing technique in image processing is reasonable and inevitability.Based above reasons, a lot of new ideas and methods based fuzzy information processing technique is proposed in this paper, and achieve better results. This dissertation is classified seven chapters, which is organized as follows:In chapter 1, it is illustrated that image processing is basic technique problem in computer vision and intelligent information processing, it is illustrated that the use of fuzzy information processing technique in image processing is reasonable and inevitability. Finally, the main achievements of this dissertation are summarized.In chapter 2, fuzzy mathematics knowledge relating with this dissertation is briefly introduced as mathematics preparation of this dissertation.In chapter 3, a lot of image enhancement local algorithms in space field are proposed. Removing different kinds of impulsive noise in image is systematically researched, and proposes three image removing noise filter based on fuzzy information processing technique, i e., fuzzy entropy-based filter for removing noise in multilevel image, histogram-based filter for removing salt and pepper noise in image, and histogram- based adaptive filter for removing image impulsive noise. Experimental results show that these filters give superior performance compared with conventional filters and other fuzzy filter.In chapter 4, an image enhancement point algorithm in space field is researched. And proposes a new adaptive modifying image grey value method based on fuzzy information processing technique. Through quoting fuzzy entropy, the new method utilizes image neighborhood statistic information, fuzzy information and human visualproperties. The experimentation results show that new algorithm can effectively improve the overall contrast of an image, can increase image visual effect, can sharpen the image edges and avoid much enhancement of noise.In chapter 5, we propose an image enhancement algorithm in transform field based on fuzzy information processing technique. A better performance image fuzzy enhancement arithmetic operator is proposed, the enhancement arithmetic operator is induct into multi-level image fuzzy enhancement algorithm, and get better effect; In addition, through quoting fuzzy entropy, make the select of threshold value in image enhancement have a certainty purpose; Finally, we use our new algorithm to extract image edges, with better result than that of now-available image fuzzy enhancement methods.In chapter 6, through quoting fuzzy entropy, and constructs a lot of better performance information measures. These measures are gotten via input image, and compose a data set that can reflect the natural characters of image edge, and divide the data set into two clusters of the edge data set and not edge data set by weighed FCM. Our method does not need any threshold; more sensitive for weak edge detection.In chapter 7, the damnified image segmentation is deeply researched, the damnified image include three, i e., the noise-corrupted image, the objective and background-overlapped image, the boundary-fuzzy image, and consider that the uncertainty in these image is essentially fuzzy uncertainty, thus, the fuzzy information processing technique is reasonably used in image segmentation, and get better effect.更多还原