【作者】 彭冬亮；

【导师】 吴铁军；

【作者基本信息】 浙江大学 ， 模式识别与智能系统， 2003， 博士

【摘要】 随着信息技术和计算机科学的迅速发展，图像处理、自动目标识别以及计算机视觉技术已经被广泛于工业、国防、航空和航天等诸多领域。这篇文章重点讨论了如下的三个问题： 1．灰度等级少、对比度低的降质图像增强问题。图像增强是图像预处理的重要组成。通过不同的图像增强手段可以得到适用于不同用途的增强图像。传统的图像增强方法有灰度变换、直方图修正、直方图均衡、图像平滑和维纳滤波。由于图像中存在许多不确定性和不精确性，对于这种类型的图像，与传统的图像增强手段相比，基于模糊集理论的图像增强技术在一定范围内可以取得较好的增强效果。但是由Pal提出的经典模糊增强方法并没有改变图像灰度等级的上界，所以这种增强方法不适合于灰度等级少、对比度低的图像增强问题，并且这种经典模糊增强方法的隶属度函数值域不具有通常的规范形式，即[0，1]区间。为了解决上述问题，本文在经典模糊增强方法和灰度变换基础上提出了一种广义的迭代模糊增强算法。这种新的图像增强方法包括三个阶段，即图像滤波、模糊增强和灰度级变换，扩大了原始图像灰度等级的范围，在保留了原有模糊增强方法和灰度变换优点的同时，规范化了模糊增强环节中灰度隶属度函数的形式。同时为了结束迭代增强，基于图像灰度直方图分布的统计特性，提出了一种客观的图像质量评价指标。计算机仿真结果表明这种新的图像增强方法比模糊增强和灰度变换方法更适合于处理灰度等级少、对比度低的图像增强问题。 2．具有方向可调节性的边缘检测器的设计问题。在图像处理和图像分析中，图像中的边缘包含了重要的图像信息，因此如何有效快速地提取图像中的边缘是一个十分重要的问题。边缘检测器的性能直接影响着图像分析和物体识别的结果。通常的边缘检测器只能有效检测几个特定方向的边缘，如果需要检测图像中各个不同方向的边缘，使用通常的卷积方法在计算上是十分费时的。为了解决这个问题，本文基于变换群理论提出了一种方向自适应滤波器的设计方法；给出了这个滤波器与图像卷积的算法。这种自适应滤波器经过参数变换后的结果可以用一组固定的、有限数目的基滤波器的线性组合表示。当不同方向的滤波器与图像进行卷积时，计算效率可以得到显著提高。一个仿真实例表明这种用于边缘检测的自适应滤波器是有效的。 3．六轴机械手视觉系统的设计和应用问题。工件的自动识别和定位是实现加工工业现代化的前提，可以极大提高生产效率和减轻工作人员的工作负担。本11 摘 要文从实际应用角度出发，设计和开发了六轴机械手的视觉系统，从硬件和软件方面详细描述了这个系统的实现过程。在图像识别阶段，提出了一种有效的图像分割方法，即改进标号算法；还提出了一种有效确定图像中物体位置和方向的算法，解决了物体方向表示的奇异性问题；为了确定图像上物体的绝对坐标，本文详细分析了像素坐标系、图像平面坐标系和机械手参考坐标系之间的关系，讨论了几个与视觉系统应用有关的实际问题。最后的实际应用结果验证了视觉系统的实用性。更多还原

【Abstract】 With the quick development of information technique and computer science, the techniques about image processing, automatic target recognition and computer vision have been widely applied in many fields, such as industries, national defence, aeronautics and aerospace. In this thesis, the following three problems are discussed in detail:1. The image enhancement problem for degraded images with less gray levels and low contrasts. Image enhancement is an important part of image preprocessing, and enhanced images suitable for different applications can be obtained using different image enhancement methods. The traditional image enhancement approaches include gray-scale transformation, histogram modification, histogram equalization, image smoothing and Wiener filtration. The theory of fuzzy sets has been used to deal with image enhancement problems for degraded images in which the image edges are uncertain and inaccurate. For those kinds of images, to some extent, the good enhancement effect can be obtained using the fuzzy sets-based image enhancement method instead of the traditional image enhancement approaches. The gray level maximum has not been changed in the classical fuzzy enhancement method proposed by S. K. Pal, so this kind of image enhancement method is not fit for the enhancement problem of degraded images with less gray levels and low contrasts; the fact that the value domain of membership function of gray levels is not normalization form, i.e. [0,1], is another disadvantage of the traditional fuzzy enhancement means. To deal with the problems mentioned above, a generalized iterative fuzzy enhancement algorithm is proposed in this thesis that consists of a three-stage procedure, i.e., image filtering, fuzzy enhancement and gray-level transformation. The generalized fuzzy enhancement method extend the gray level range of the original image, and a canonical form of membership function in the stage of fuzzy enhancement is presented which remains the advantages of the original fuzzy enhancement and the gray level transformation while transforming the membership function of the gray scale to [0,1]. A new image quality assessment criterion is suggested on the basis of the statistical features of the gray-level histogram of images to control the iterative procedure of the proposed image enhancement algorithm. Computer simulation results showedthat this new enhancement method is more suitable than fuzzy enhancement and gray-level transformation for handling the enhancement problems of images with less gray levels and low contrasts.2. The design problem of edges detector steering orientation. Since edges in images possess very important information about objects in image processing and analyzing, how to extract edges effectively and rapidly becomes an imperative issue. The results of image analyzing and object recognition have a close relationship with the performance of edges detector. Generally, edges in some specified directions can be only detected by most of edges detectors, and detecting edges in many different orientations using usual convolution means is very costly computationally. To handle this problem, a novel adaptive filter for orientation parameter design technique is proposed in this thesis according the theory of One-parameter Transformation Group. An algorithm for the convolution of this filter and an image is also given. The filter, after this parameter changes, can be represented in form of the linear combination of a fixed, finite set of basis filters. When this kind of filters are convoluted with an image in different orientations, the computing efficiency is improved remarkably. A simulation example for edge-detection is given to demonstrate validity of the adaptive filter.3. The design and application problem of the vision system for a six-joint robot manipulator. The automatic recognition of a workpiece and determining its position and orientation (pose) is a prerequisite for implementing production modernization, and thereby can improve productive effici更多还原