【作者】 郑宇乐；

【导师】 耿兆丰；

【作者基本信息】 东华大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 目前在纺织行业中,对针织物结构的分析与识别主要还是凭经验或借助专业工具来完成。这种由专家人工进行针织物组织结构参数的分析和提取虽然具有权威性,但是操作要求高,不易掌握。同时分析与识别的时间周期也相对较长,分析过程单调而乏味。因此,有必要研制能代替人工对针织物的结构参数、组织编织结构等进行有效的自动检测与分析的算法和技术路线。本文研究了利用基于形状特征参数统计、PCA算法和能量统计的图像处理与模式识别技术对针织物进行自动参数提取和组织结构识别的三种方法,提出了针织物自动识别的流程与技术路线。然后通过分析针织物图像亮度分布的特点,给出提取针织物结构参数——织物密度、线圈距离的自动提取算法,并将它们的计算结果与手工测量结果进行了比较和分析。本文提出了针织物组织结构识别的几种方法。首先我们通过扫描仪把样品图像输入到计算机,获得针织物的采样图,然后再通过图像处理对针织物组织结构进行特征提取和模式识别。第一种方法是利用链码和形状数提取针织物组织的形状特征参数:形状数、周长、面积、形状因子等特征参数,再通过这些形状特征参数识别未知针织物的组织结构。第二种方法是利用PCA算法对输入的针织物图像数据进行降维,提取图像数据的全局特征,然后通过测试样本的全局特征与训练样本的全局特征之间的欧氏距离来识别未知针织物的组织结构。第三种方法是利用Mallat和Coif1小波变换提取针织物在水平方向,垂直方向和斜方向上的能量特征值,然后通过未知组织针织物采样图的能量比例特征值与1+1罗纹组织、纬编平针组织和1+1双反面组织的标准能量比例特征值之间的距离,来识别未知针织物的组织结构。本文所提出的上述技术路线和算法,都得到了可行性的验证,取得了较好的效果,在该领域具有一定的理论意义和较大的参考价值。选用形状数为特征参数可以克服针织物在采样过程中样品平移、旋转带来的误差;选用PCA算法可以对针织物图像数据进行降维获取全局特征,大大减少计算量,加快识别速度,而又不失正确性;选用二维小波变换提取的针织物水平、垂直和斜方向上的能量特征具有普遍性,且无需前期的复杂的预处理过程,选用能量特征更具实用性。本文所提出的方法在对于针织物组织结构的识别上具有一定的新意,可以适用于诸如:纬编平针组织、罗纹组织和双反面组织等简单的纬编针织组织,具有一定的推广价值。更多还原

【Abstract】 Nowadays, in the field of Textile industry, analysis and recognition of knitted fabric mainly depend on manual work or special equipment. Though this way is authoritative, it is not easy to manipulate, and hard to master. Moreover, it is time consuming and tedious. So it has become necessary to research on how to get and analyze knitted fabric structure with computer automatically.In this work, we provide three algorithms, the first of which is based on character statistics, the second of which is based on PCA algorithm and the third of which is based on energy statistics to get and analyze knitted fabric construction parameters and structure using Digital Image Processing and Pattern Recognition technology, and develop the flow and technical route of automatic analysis of knitted fabric parameters and structure recognition. Then we depict particularly the algorithm of getting knitted fabric construction parameters-vertical and horizontal densities and coil distance, according to the character that knitted fabric’s structure is periodic on space. Finally, the results of calculating are compared to the results of manual measuring.In this work, We provided the flow of weft knitted fabric structure analysis and recognition. First, the image of the sample is inputted to computer by scanner. Then abstract the characters from the image through image processing. Finally, the unknown image can be recognized by these characteristics. The first algorithm is that abstract the shape characters from the image based on chain code and shape number, and then recognize the variety of the kintt in the unknown image by the shape characters. The two algorithm is that reduce the dimension of the image data and abstract the global character, based on PCA algorithm, and then then recognize the variety of the kintt in the unknown image based on the distance between the test sample and the training samples. The third algorithm is that abstract the energy characters of the horizontal, vertical and oblique directions based on the Mallat algorithm and Coifl wavelet transforms, and then recognize the variety of the kintt in the unknown image.The technical route and algorithms we provided above, have gained feasibility validation and achieve applicable results, have certain theoretical value and use for reference in the domain. Because the shape number has no changes when the fabric sample is translated and revolved, we choose the shape number as the characteristic to come over the error during the sampling. Because the PCA algorithm can reduce the calculation by reducing the dimensiong of the image data, we choose the algorithm to speed up the process of the recognization. We choose the energy characters, because it can be used in the images of different sizes and color without complex pre-processing. The pattern draft recognition method has some innovation in this field, especially analyzing the weft plain knitted fabric, rib knitted fabric and pearl knitted fabric.更多还原