基于图像处理的花边布瑕疵检测算法研究

发布时间：2018-02-02 20:19

本文关键词： 瑕疵检测图像分块二值化中值滤波特征提取基元　出处：《吉林大学》2017年硕士论文　论文类型：学位论文

【摘要】：花边布又称抽纱、蕾丝,是指有花纹图案的、用于装饰的带状织物.我国生产花边的历史虽然晚于欧洲,但生产技术水平提高速度很快,已形成了较大的生产规模.在纺织品的生产过程中,瑕疵检测对于提升织物的品质起着十分重要的作用.目前我国的纺织品主要依赖于人工对其进行质检,检测过程容易受到人为因素影响,误检情况时有发生.为降低生产成本、提高检测准确率,纺织品瑕疵自动检测应逐步替代人工检测.本文的目的是设计有实际应用价值的花边布自动瑕疵检测算法.以横W纹花边布为主要研究对象,瑕疵检测算法分为三个步骤:图像预处理、布面的特征提取、瑕疵检测.图像预处理主要包括图像二值化和图像去噪两部分.由于织物生产环境复杂,拍摄的花边布图像存在着亮度不均的问题,这给下一步的特征提取带来了很大困难.为解决这个问题,本文对灰度图像二值化的方法进行研究,提出了一种基于大津法的图像分块二值化方法,先在矩形网格下利用传统的大津法求小块矩形图像的阈值,再在三角形网格下,将阈值做平滑处理,最终得到更为理想的二值化图像.然后利用中值滤波,去除二值图像中对图案纹理产生影响的噪声,为下一步的布面特征提取做好准备工作.在布匹中,其图案纹理是有周期性的,存在一个尺寸最小的区域,整个布匹的图案纹理可以通过这个区域做平移操作得到,这个最小区域称为格子.格子可以被分解为更精密的组成成分,称为基元,通过基元的平移、旋转、镜面反射等操作可以得到格子.在布面的特征提取部分,借鉴人脸识别特征提取方法中的基于局部特征的方法,基本思想是利用图案纹理的局部几何特性,找出每个基元的关键点;然后利用关键点对花边布进行分块处理,得到花边布的格子和基元图像.第三步为瑕疵检测,利用Ngan等人提出的运动差能量和方差的方法.这种方法可以忽略基元图像的轻微变形和不对齐,同时运动差能量放大了有瑕疵基元的瑕疵信息,使得瑕疵更容易被检测出.通过确定无瑕疵格子中任意两个基元的运动差能量和方差的取值范围,判断一幅格子图像中是否存在瑕疵.若能量和方差的计算结果在确定的范围内,则图像无瑕疵,反之则有瑕疵.这种方法可以判断出较大面积的花纹错乱、油污等瑕疵.为了可以进一步判断出较小面积的瑕疵如孔洞、划痕等,本文利用上述的方法,对从基元中分割出的子窗口图像继续进行瑕疵检测.对现有的217张图像进行了实验,结果表明本文提出的花边布瑕疵检测算法具有可行性.
[Abstract]:Lace fabric, also called lace, refers to the ribbon fabric with pattern pattern, which is used for decoration. Although the history of lace production in China is later than that in Europe, the level of production technology improves rapidly. In the process of textile production, defect detection plays a very important role in improving the quality of fabrics. At present, Chinese textiles mainly rely on manual quality inspection. The detection process is easy to be affected by human factors and false detection occurs from time to time. In order to reduce production costs and improve the accuracy of detection. Automatic detection of textile defects should be replaced by manual detection step by step. The purpose of this paper is to design an automatic flaw detection algorithm for lace cloth with practical application value. Defect detection algorithm is divided into three steps: image preprocessing, fabric feature extraction, defect detection. Image preprocessing mainly includes image binarization and image denoising. In order to solve this problem, the method of binarization of gray image is studied in this paper. In this paper, a method of image segmentation based on Otsu method is proposed. Firstly, the threshold value of small rectangular image is obtained by using the traditional Otsu method under rectangular mesh, and then the threshold value is smoothed under triangular mesh. Finally, a more ideal binary image is obtained, and then the median filter is used to remove the noise that affects the pattern texture in the binary image, so as to prepare for the next step of fabric feature extraction. Its pattern texture is periodic, there is a minimum size of the region, the entire fabric pattern texture can be obtained by this region for translation operation. This minimum region is called a lattice. The lattice can be decomposed into more precise components called primitives. The feature extraction part of the fabric can be obtained by the translation rotation and mirror reflection of the elements. The basic idea is to find out the key points of each element by using the local geometric characteristics of pattern texture for reference to the method of feature extraction based on local features of face recognition. Then we use the key points to divide the lace cloth into blocks to get the lattice and basic image of the lace cloth. The third step is defect detection. Using the method of motion difference energy and variance proposed by Ngan et al. This method can ignore the slight distortion and misalignment of the primitive image and amplify the defective information of the defective primitive by moving difference energy. By determining the range of the difference energy and variance of the motion of any two elements in the blemeless lattice, the defect can be detected more easily. If the calculation results of energy and variance are within a certain range, the image has no defects, otherwise there are defects. This method can determine the large area of pattern disorder. In order to further judge the smaller areas of defects such as holes, scratches and so on, this paper uses the above method. The defect detection of the sub-window image from the primitive is carried out. 217 images are tested, and the results show that the algorithm proposed in this paper is feasible.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TS101.97;TP391.41

【参考文献】