【作者】 廖剑利；

【导师】 康志伟；

【作者基本信息】 湖南大学 ， 通信与信息系统， 2005， 硕士

【摘要】 根据Marr计算机视觉理论,图像边缘检测在计算机视觉研究中占据着重要地位。但由于问题本身的复杂性和技术手段的限制,图像边缘检测这一经典问题的研究困难重重。小波变换分析是近十年来在工具和数学方法上的重大突破,其卓越的时频分析本领,为这一计算机视觉问题的研究带来了新的契机。但是,由于小波理论产生的时间不长,其理论还算不上很成熟,应用中缺少完全行之有效的应用方法与步骤,这使得小波变换的应用比Fourier变换的应用复杂困难得多。针对这种情况,本文尝试将小波变换分析应用于图像的边缘检测这一计算机视觉中的重要环节,结合问题的需求,研究小波变换的特性,选择恰当的小波基函数和小波变换形式,提出问题解决的新思路和新方法。 本文主要工作如下: (1)深入研究小波变换理论和图像边缘检测理论方法。 (2)根据边缘检测的评价准则,参照最佳边缘滤波器的设计要求,确定用于边缘检测的小波基函数的一般准则,同时得出了“最佳”边缘检测小波——二次B样条小波。本文提出了一种局部自适应阈值选取方法,进而提出一种改进的二次B样条小波图像边缘检测方法。 (3)本文通过提升算法实现了不可分离小波变换,在此基础上提出了一种新的基于“五株形”栅格排列的提升不可分离小波图像边缘检测方法,并通过试验证明了新方法较传统的小波边缘检测方法更具优越性。 (4)根据可操纵小波多分辨率分析的原理,设计实现了基于二维多方向微分的小波滤波器组,利用多方向微分特征,构造了图像边缘检测的新方法。通过仿真试验并将其与Canny算法和传统的B样条小波边缘检测算法进行比较,证明了该方法的有效性和优越性。 本文使用MATLAB作为仿真实验平台对上述研究进行了验证和分析。更多还原

【Abstract】 According to Marr’s theory on computer vision, image edge detection occupies an important position in computer vision research. Research on the typical problem in computer vision, however, is very difficult because of the complexity of the problem and limitation of the available technique. Wavelet transform analysis is a breakthrough in mathematical tools and methods in recent ten years, its prominent ability for time-frequency analysis brings new chances to solve problems in computer vision. However, the theory of wavelet is not so mature since it was very short since wavelet was put forward, and there are lack of effective methods and rules to conduct wavelet applications, so applications of wavelets are much more difficult than applications of Fourier transform. Under this situation, we try to apply wavelets to image edge detection which is the important module in computer vision. Properties of wavelet are studied carefully with respect to the requirements, and then proper mother wavelet and proper wavelet transform formula are selected, and at last new ideals and new methods for the problem are given.The main important work of this paper:1. Deeply studyed wavelet analysis theory and image detection technology.2. According to the criterion of edge detection and consulting the design-aim of optimal edge-filter, we come out with the best edge detection wavelet—Quadratic B-Spline wavelet. The paper presented a new local adaptive threthold way and an adaptive threshold edge detection for image based on Quadratic B-Spline wavelet transform was presented then.3. This paper realized the non-separated wavelet transform using the lifting scheme and then presented a new method of the image edge detection using the lifting non-separate wavelet based on the quincunx lattice. And the superiority of the new method was proved by the results in the experiment compared with the tradional way.4. A new method of the image edge detection was proposed based on the steerable wavelet multi-resolution principle. We designed the two-dimention directional derivative wavelet filters and achieved the multi-directional derivative feature which the new method was based on. Compared with the Canny method and the classical edge detection way using B-spline wavelet, the efficiency and superiority of the new method was proved by the results in the simulated experiment.We used MATLAB as the simulation platform to verify the research mentioned above更多还原