【作者】 王刚；

【导师】 贺安之；

【作者基本信息】 南京理工大学 ， 光学工程， 2007， 博士

【副题名】基于超小波与分形理论的算法研究

【摘要】 光学无损检测技术涉及光学、图像处理、信息处理、模式识别等多种学科领域。是一种不干扰待测场分布的测量诊断技术。本文提出的光学无损路面病害自动检测算法，针对目前路面病害检测算法存在的精度低、适用范围窄、自动化水平低的缺陷，着重研究了复杂背景下裂纹病害的提取以及病害的自动分类问题。主要工作内容和研究成果简述如下：1．本文提出了基于Ridgelet变换域的模糊自适应图像增强算法，利用在傅里叶变换域给出的Radon变换积分投影与原场分布在频域上的联系，实现离散Radon变换投影切片定理。并提出Radon变换重建原图像的基本条件；利用广义模糊集合概念和最大模糊熵原理，提出一种自适应设置模糊增强函数方法，使得增强后的图像在抑制噪声、增强特征方面达到较好折衷。2．提出基于Curvelet变换域的路面图像去噪算法。Curvelet变换综合了Ridgelet擅长表示直线特征和小波适于表现点状特征的优点，并充分利用了多尺度分析独到的优势，适用于路面中网状裂纹的增强去噪。针对Curvelet系数稀疏矩阵表示出来的局部结构信息设计了一种自适应的消噪处理方法，对局部邻域中心点的高频Curvelet系数进行处理得出新的Curvelet系数，以达到去噪的效果。3．提出了一种基于亚像素多重分形原理求取图像奇异性的新型算法，降低了单纯依靠整数像素位置灰度级梯度信息计算边缘测度所产生的误差。该算法结合CCD成像机理给出在亚像素位置的灰度级梯度分布规律，利用多重分形理论将实际图像分割成一系列具有不同奇异性指数的分形集合，对应着从边缘到纹理各层面的图像内容。采用多重分形理论并结合具体图像的物理和统计特性进行路面裂纹病害图像分割的分析，通过奇异性指数的概率分布以及最奇异指数值可判断图像中有无裂纹病害。4．本文最后介绍了自行研制的路面裂纹病害自动检测系统的框架结构。主要由图像采集和病害检测两大模块组成。详细说明了系统中图像采集模块的设备运行参数，以及病害检测模块的工作流程。目前该系统已经投入使用，并完成多条高等级公路的检测任务。实际使用结果表明本系统的正确率在90％以上。更多还原

【Abstract】 The optical nondestructive examination is involved with many kinds of disciplinedomains, such as optics, image processing, information processing, and pattern recognition.The technology can measure the observation object without interferencing the field. In thisdissertation, the pavement damage automatic detection arithmetic is proposed based on theoptical nondestructive examination. In the current pavement damage detection arithmetics,there exist the defects such as low accuracy, narrow scope of application and lowautomation level. Then we lay stress on the method of abstracting and classifying thecracks under the complicated background. The main work is described as follows:1. The implement algorithm of adaptive fuzzy image enhancement is proposed inRidgelet transform domain. The discrete Radon projection slice theorem is brought basedon the integral projection of Radon transform and original field distribution in the Fourierdomain. The basic condition of reconstructing the original image through Radon transformis proposed. The algorithm of adaptive fuzzy image enhancement is put forward based onthe generalized fuzzy set and the maximum fuzzy entropy. The processed image is thebetter compromise between enhancing the characteristics and inhibiting the noise.2. The pavement image denoise algorithm is proposed based on the Curvelet transform.Curvelet transform generalized the Ridgelet feature representing the straight line and theWavelet trait characterizing the point. And it takes full advantage of the multiscale analysis.It is applicable to enhance the alligator cracks. The method of adaptive denoise is proposedbased on the local information denoted by the Curvelet transform coefficient matrix.Derived from the Curvelet high frequency coefficients of local central point, the processedCurvelet coefficients can enhance the object.3. In order to analyze image singularity and the features of the different sections, a newmultifractal algorithm is proposed based on sub-pixel edge measure. The greylevelgradient areal density function and edge-measure of random subsets (radii can reach theprecision of sub-pixel) are obtained by the square aperture sampling law on the position ofsub-pixel. Utilized the multifractal frame, the image can be segmented into a series fractalsets of the different singularity exponents. The existing cracks can be judged by probabilitydistribution of singularity exponents and the most singular exponent.4. In the end, the automatic detection system for pavement damage is presented. It isconsisted of image collection and pavement damage detection. The working parameters of the apparatus and the workflow of the detection have been presented in details.Now thesystem has been gone to service and completed several highway detection tasks. Theresults show that the accuracy of the indexes of the experiments has reached above 90%.更多还原