【作者】 刘正；

【导师】 黄战华；

【作者基本信息】 天津大学 ， 光学工程， 2009， 博士

【摘要】 本文提出并新建了一种轮胎X光多纹理图像的特征提取模型和系统方法。该方法首先采用极坐标频谱分析法分析了多纹理图像的频谱特征;然后构建最优的滤波方法对纹理频谱分量进行分割;最后综合纹理分析结构法和统计法提取各纹理分量的特征信息,如方向、周期、边缘、形状、规则度等,并根据这些特征检测轮胎X光缺陷。针对轮胎X光多纹理图像方向分解过程中的Gibbs效应和频谱混叠等局限性,首次设计了一种抗频谱混叠的正交可分离方向滤波器,采用Gauss函数对方向滤波器频率响应进行调制,提高了多纹理图像方向分解的抗频谱混叠和失真的能力。结合轮胎X光图像的频谱特征,提出了采用多个通道Gabor滤波器组代替Gabor变换进行空频域联合分析的新方法,并研究了滤波器组的设计方法,有效地降低了纹理特征提取的复杂度,最后根据纹理特征实现了轮胎X光图像的多纹理分割和提取。结合轮胎X光图像纹理结构特点,首次提出并实现了一种有效的轮胎X光缺陷综合检测方法。该方法首先采用局部一阶统计量提取纹理灰度分布特征并分割纹理异常区域;然后,在纹理异常区域提出了一种基于邻域灰度特征点搜索的纹理基元骨架快速提取方法,并采用定向搜索算法对骨架进行矢量化描述,建立纹理结构矢量模型,提出了7种矢量化参数对纹理结构进行描述,并给出了计算方法;最后根据矢量化结构描述参数实现轮胎X光缺陷的检测和识别。根据轮胎X光自动检测系统——AXIS系统的应用需求分析,采用基于C#语言和MATLAB COM组件的联合编程方法实现该系统的模块化编程。通过现场试验测试表明AXIS系统能有效地检测出轮胎X光缺陷,正确率达到99.95%,并且系统功能满足现场检测的自动化和智能化要求。更多还原

【Abstract】 Textural defects analysis is one of the main research topics in computer vision inspection field. Image defects can be detected and recognized by analyzing anomaly of texture. However, in complex multi-texture images, for example, in tyre X-ray image, it is difficult to represent characteristics of textures and each texture layer must be segmented and extracted firstly. Now the most common methods of texture analysis are structural analysis method, statistical method, model method and signal processing method, etc. However, it is unable to acquire texture features in complex multi-textural images directly. In this dissertation, a new multi-textural features extraction model and system method has been proposed to analysis tyre X-ray images. In this model, each texture component has been extracted and segmented by spectrum filtering method. Then using textural statistical and structural method, some textural features, such as direction, period, edge, shape, consistency, had been extracted.A new method has been proposed to improve the performance of directional decomposition process in tyre X-ray images. In this method, orthogonal-separable directional filters have been innovatively designed to decompose tyre X-ray images based on thire texture characteristics. Gauss function has been used to modulate the frequency response of filters. So that Gibbs effect has been removed and the capability of anti-spectrum-aliasing has been improved.According to the spectrum characteristics of tyre X-ray images, a multi-channel Gabor filtering method is proposed to substitute Gabor transform to implement joint time-frequency analysis of images. The construction method of filter bank has been researched. So that the extraction complexity of textural features has been decreased and based on these features, the multi-texture segmentation and extraction of tyre X-ray images has been reached.According to the structural characteristics of texture in tyre X-ray images, an effective systematic method has been proposed to inspect X-ray defects. First, first-order statistics has been used to extract texture features and segment the region of texture anomaly. Second, skeleton of texture elements has been extracted by searching of feature points in local region and the vectoriazation of skeleton has been reached by directional searching arithmetic. Based on vectors of texture skeletons, 7 parameters are proposed to describe structure of texture and X-ray texture. At last, according to parameters, tyre X-ray defects have been detected and recognized.According to the analysis of application requirements in tyre Automatic X-ray Inspection System (AXIS), the joint modularization program method of C# and Matlab COM has been used to implement the AXIS. The experimental results showed that tyre X-ray defects had been inspected effectively by AXIS, the accurate rate had reached 99.95% and AXIS function had met the command of automation and intelligence.更多还原