【作者】 姚芳；

【导师】 李立轻；

【作者基本信息】 东华大学 ， 纺织工程， 2010， 硕士

【摘要】 在织物生产过程中,织物的质量控制如疵点检测、起毛起球等级评定、织物褶皱评定等是非常重要的,然而传统的检测丰要依靠人工检测,由于受人为因素影响,误检率和漏检率较高。随着计算机技术的广泛应用,织物的纹理分析开始实现自动化、智能化,其理论基础是织物纹理的数字化表征。织物图像纹理表征分析直接决定了织物组织参数和疵点自动检测等研究的应用效果。在将计算机视觉应用于纺织品的外观分析和检测方面,织物纹理表征和分析在值得进行进一步深入的研究,对数字化纺织的应用具有重要应用价值。织物纹理属于周期性纹理,小波变换是周期性纹理图像的特征表示方法中最主要的方法之一。本文正是基于自适应小波对纹理图像进行特征提取,改善了传统小波基非自适应的缺点,从织物图像的疵点检测的角度验证了该织物纹理表征算法的可行性和有效性。本文采用自适应小波三层分解的方法来表征纹理。首先,本文简单介绍了小波分析在机织物纹理表征上的应用及其在疵点检测方面的验证,经过比较发现自适应小波分析适合织物纹理的表征。其次,对织物图像进行预处理后,通过逼近条件寻找与织物纹理相匹配的滤波器,提出两种新的约束条件即小波系数纹理方向的波动性和灰度共生矩阵,与常用的能量和小波系数差等逼近条件相比,本文方法效果更佳。接着,关于自适应小波分解层数的确定,突破以往自适应小波单层分解,采用熵值减小程度作为分解终止信号,确定分解层数为三层。在确定了表征纹理的方法为自适应小波三层分解,以织物疵点检测为例,判断该方法的有效性。最后,将这种纹理表征的方法运用于平纹和斜纹类织物的疵点检测,对织物图像进行分解后提取特征值,通过特征值判断疵点的存在性,根据比较几种阈值分割方法,找到适合疵点二值化的方法为一维最大熵值方法。运用椭圆的方法计算目标疵点部分面积、大小、与水平方向夹角和长短轴比值,据此分析该疵点类型及大小等特征。试验结果表明本文所运用的表征织物纹理的方法是可行的,有效的。更多还原

【Abstract】 In the fabric production process, the fabric quality control is very important, such as defect detection, pilling rating, puckering assessment, and so on. However, the traditional fabric detection mainly relied on manual testing, due to human factors, whose miss rate and fall-alarm rate were high. With the extensive application of computer technology, it begins to realize the automation and intelligence in fabric texture analysis, which is based on the digital characterization theory of fabric texture. The application effect of the research on fabric parameters and defect study of auto-detection is directly determined by analysis on characterization of the fabric image texture. As to applying computer vision to the analysis and detection of the appearance of textiles, fabric texture characterization and analysis are worth further studying which is important to the application of digital textile.Fabric texture is periodic. Wavelet transforming is one of the most important methods to describe the periodic features. In this paper it extracts the features of texture image using adaptive wavelet, which improves the traditional non-adaptive wavelet methods. This paper verifies the feasibility and effectiveness of the characterization algorithm of fabric texture in terms of fabric defect detection.In the paper it describes the fabric features using adaptive wavelet with the three-ply resolution. Firstly it gives a brief introduction of the application of this method in characterization of woven fabric texture and its verification in defect detection. Adaptive wavelet decomposition can be proved to be a better method in describing the characterization of fabric texture, suitable for fabric defect detection. Secondly, after the image’s pre-processing, we search for the filter to match fabric texture through the approximation conditions. This paper offers two new constraint conditions, the fluctuations in the direction and gray-scale of co-occurrence matrix of wavelet coefficient texture. The method in this paper is proved better than the approximation condition about energy and scope of the wavelet coefficient. Thirdly, Breaking through the single adaptive wavelet decomposition, it shows a stop signal is the extent of some of the decrease of entropy in Approximate Image. It applies the three-ply resolution method, which is more appropriate for fabric defect detection. In the paper the method to describe the periodic features is the adaptive wavelet with the three-ply resolution, which is shown effective using defect detection. At last, this texture characterization method is applied to the defect detection of the twill fabric and the plain fabric, it shows on how to decide if the defect exists by the characteristic value. According to comparing several thresholding methods, it finds a suitable method of binarization, which is one-dimension maximum entropy. It calculates the target’s area, size, the angle between the long axis and horizontal and the long axis and short axis ratio, using ellipse method. And it gives an analysis of defect characteristics, such as size and type, and so on. The result is shown feasible and effective.更多还原