【作者】 王胜先；

【导师】 袁晔；

【作者基本信息】 北京服装学院 ， 机械电子工程， 2012， 硕士

【摘要】 织物组织识别是纺织测试分析及设计领域的前沿课题,具有重要的现实意义。旨在通过分析织物的外观纹理结构、纹理走向及色调分布提取纱线组织结构参数并识别织物的纹理组织规律。对各种织物面料类别及不同的组织结构进行正确的识别与分析,可以为面料设计、仿样和创新提供科学依据。织物组织结构识别的研究对于纺织及面料行业的产品生产、设计创意和仿样改样进而提升我国纺织产业的国际竞争力具有重要意义。纺织品CAD技术的出现在很大程度上简化了产品的设计和织造过程。但由于织物组织结构复杂多样,生产过程受纺织工艺的影响,最终形成的纺织品具有许多不确定性,从而增加了织物组织自动识别的难度。传统织物组织识别方法无论是统计方法、几何方法,还是谱方法和模型方法都假定织物图像为平稳信号,即经纬纱线排列理想,水平方向和垂直方向上的灰度信号能很好的体现纱线的间隙位置。然而实际织物图像为非平稳信号,纱线排列与理想模型存在较大差异。传统的织物组织识别方法很难适应实际织物图像的非平稳性特点。结合织物图像的非平稳性特点,突破现有织物组织识别技术瓶颈,发展新的织物参数分析与识别技术成为织物组织智能分析与设计领域的前沿核心问题。本文研究基于图像的空间-频率域分析理论,针对织物的组织类型与结构参数进行了深入的研究,主要工作包括:(1)研究了Hilbert边际谱、Hilbert-Huang Hough变换在织物结构参数估计中的应用,分别基于Hilbert边际谱和Hilbert-Huang Hough变换提出了平纹组织纱线密度和斜纹组织纱线密度估计算法。将织物图像转换为一维信号,对一维信号分别求取Hilbert边际谱和Hilbert-Huang Hough变换,据此求得织物图像在经向、纬向和斜纹方向的主频率,并将主频率换算为织物纱线密度。仿真实验表明,该方法能精确估计纱线密度。(2)提出了一种新的基于B样条小波变换极大模的织物多组织识别算法。由B样条小波对织物图像进行分解,利用分解后的纬向和经向的子图像建立正常织物图像和待检测织物图像的极大模边缘图像;然后由二者的差分结果提取检验织物组织识别准确性的判定参数;最后由组织矩阵绘制织物的组织图。仿真结果表明:本文方法能够精细准确的刻画织物组织点位置并具有效率高,稳定性能好等优点。(3)提出了一种基于四元数-Gabor滤波和四元数-PCA彩色织物组织识别新方法。将彩色织物图像表示为四元数形式,并应用四元数-Gabor滤波器进行滤波处理,对滤波后的图像进行聚类处理实现初始分割,在初始分割的基础上运用四元数-PCA建立彩色织物组织结构的精确纹理分割模型。更多还原

【Abstract】 Recognition for fabric weave pattern is one of the key issues in fabric analysis and design field. The parameters of yarn can be extracted and fabric weave pattern can be recognized by analyzing the surface, weave pattern and color distribution of the fabric sample. Scientific basis of fabric innovative design is established with the accurate weave pattern recognition, which will promote competitiveness of textile industry.Although the emergence of textile CAD system greatly simplifies the product design and weaving, automatic recognition for weave pattern has a great of difficulty. Due to the complexity of manufacturing, the final form of the fabric has much uncertain information. The recognition performances of classical methods aren’t satisfying. No matter the statistical method, geometric method or spectral method and models method are all assume the fabric images are the ideal arrangement of warp and weft yarns, that the yarn gap position can be well represented. These methods are presented in condition of stationary and spatial-invariance signals. In practice, real fabric images are mostly non-stationary and spatial-variance. Considering the disadvantages of classical methods, this dissertation focuses on the key part of parameters of yarn and recognition of weave pattern.In this dissertation, several theoretical issues based on space-frequency domain are focused. The main contributions are given as follows:(1) On the study of the marginal spectrum of Hilbert-Huang transform a method is proposed to calculate fabric yarn density of tabby by analyzing Hilbert marginal spectrum of fabric sample image, and the yarn density of twill will be calculated by analyzing the Hilbert-Huang Hough transform. Experiments results show that this method is infinitely close to the value measured by manual.(2) A new fabric multi-pattern recognition proposed based on B-spline wavelet transform modulus maximum theory to recognize fabric sample point effectively, and capture gray-scale changes accurately in the fabric image. Therefore, the exact position of the yarns achieved clearly. This method is proposed to help simplify the implementation, and improve the robustness of the pattern detection, whether the fabric is single or multi-pattern.(3)A method of recognition for color fabric weave pattern is proposed based on the theory of Quaternion-Gabor and Quaternion-PCA. The color fabric weave pattern image is pre-processed using Quaternion-Gabor filter firstly, and clustering method is applied to approach initial segmentation, then color fabric weave pattern segmentation is build based on Quaternion-PCA method to implement accurate segmentation of different weave patterns and different color models.更多还原