机织物纹理识别方法研究

发布时间：2019-06-13 14:32

【摘要】：机织物图像组织结构识别的研究,在机织物组织结构识别方面具有重要的应用价值,同时在纹理分析与识别算法研究方面具有很好的理论意义。本文在机织物图像组织结构识别相关知识基础上,对纹理分析与识别方法进行研究,改进了机织物组织结构识别的流程和框架,提出了基于纱线边界特征的组织结构分类和基于S-Gabor特征和分类矫正的组织图识别两步机织物识别策略,在机织物图像组织结构识别过程中将两个不同特征分层次融合使用。组织点纱线边界特征计算过程中,采用亮度变化信息确定组织点纱线边界信息,首先进行局部组织点归一化处理,进一步定义“相邻组织点图像间亮度绝对变化”以及“相邻组织点图像间亮度相对变化”两步操作实现组织点纱线边界特征提取。在组织循环纱线数计算操作中,利用平均海明距离表示两条纱线间的相似度,并在利用极小值点得出组织循环纱线数后,以不同长度分割所得相邻等长子序列平均相关系数进行修正。组织结构分类操作中,提出同时利用机织物组织循环纱线数和预识别组织图斜向相关性对机织物组织结构进行分类的方案。组织图识别过程中,利用Steerable Filter和Gabor推导得出S-Gabor变换,讨论了S-Gabor变换的性质,并给出了S-Gabor变换生物学原理,在理论上验证S-Gabor变换对图像梯度信息具有更好的特征提取效果。进一步使用S-Gabor变换提取非斜纹组织点特征,并进行PCA降维,而后利用SVM分类组织点属性,最后采用不同组织点误检方案分类矫正平纹组织、斜纹及其变化组织、缎纹组织预识别组织图。本文基于天津工业大学机织物图片数据库进行实验,该数据库样本组织结构种类全面、图像信息多变、干扰因素较多,研究价值和参考价值较高,实验证明本文提出S-Gabor变换能很好地提取相关纹理特征;本文组织循环纱线数计算方法、组织结构分类方法和组织图识别方法在数据库中具有极高的准确率,分别为99.25%、99.62%、96.98%,鲁棒性较高;本文组织图识别方法很好地平衡了基于纱线边界特征组织结构识别的高效率和基于S-Gabor特征组织结构识别的高准确率。
[Abstract]:The research of woven fabric image tissue structure recognition has important application value in woven fabric tissue structure recognition, and has good theoretical significance in texture analysis and recognition algorithm research. In this paper, based on the knowledge of woven fabric image organizational structure recognition, the texture analysis and recognition methods are studied, the flow and framework of woven fabric tissue structure recognition are improved, and the organization structure classification based on yarn boundary features and the recognition strategy of two-step woven fabric based on S-Gabor features and classification correction are proposed. In the process of tissue structure recognition of woven fabric image, two different features are used in hierarchical fusion. In the process of calculating the boundary characteristics of tissue point yarn, the yarn boundary information of organization point is determined by using brightness change information. Firstly, the local organization point is normalized, and the two steps of "absolute change of brightness between adjacent organization point images" and "relative change of brightness between adjacent organization point images" are further defined to realize the extraction of yarn boundary feature of organization point. In the calculation operation of the number of tissue circulating yarns, the average hamming distance is used to represent the similarity between the two yarns, and after the number of tissue circulating yarns is obtained by using the minimum point, the average correlation coefficient of adjacent isometric subsequences obtained by different length segmentation is modified. In the classification operation of fabric structure, a scheme is proposed to classify the microstructure of woven fabric by using the number of recycled yarns in woven fabric and the oblique correlation of pre-recognized tissue diagram at the same time. In the process of organization diagram recognition, the S-Gabor transform is deduced by Steerable Filter and Gabor, the properties of S-Gabor transform are discussed, and the biological principle of S-Gabor transform is given. It is theoretically verified that S-Gabor transform has better feature extraction effect on image gradient information. Furthermore, S-Gabor transform is used to extract the characteristics of non-twill tissue points, and then PCA is used to reduce the dimension of non-twill tissue points, and then SVM classification is used to organize point attributes. Finally, different tissue point misdetection schemes are used to correct plain tissue, twill and its changing tissue, and satin tissue pre-recognizes tissue diagram. In this paper, the experiment is carried out based on the woven fabric picture database of Tianjin University of Technology. The sample structure of the database is comprehensive, the image information is changeable, there are many interference factors, and the research value and reference value are high. The experiment proves that the S-Gabor transform can extract the relevant texture features very well. In this paper, the calculation method of organization circulation yarn number, the classification method of organization structure and the recognition method of organization chart have very high accuracy in the database, which are 99.25%, 99.62% and 96.98%, respectively, and the robustness is high. The organization chart recognition method in this paper balances the high efficiency based on yarn boundary feature organization structure recognition and the high accuracy rate based on S-Gabor feature organization structure recognition.
【学位授予单位】：河北工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TS101.923;TP391.41

【参考文献】