基于Gabor和视觉信息的布匹瑕疵检测方法研究

发布时间：2018-07-01 19:19

本文选题：布匹瑕疵检测 + 人眼视觉注意机制　；参考：《江苏大学》2017年硕士论文

【摘要】：传统布匹瑕疵检测方法是依赖于检测工人肉眼对成品布匹进行人工瑕疵检测,而现存的计算机视觉布匹瑕疵检测方法主要基于对无瑕疵布匹图像提取的特征进行学习,从而进行布匹瑕疵检测。该类方法需要对无瑕疵布匹图像进行对比学习,且对特定角度下采集的布匹实验图像有严格要求。上述方案依赖于无瑕疵布匹图像的先验知识学习,对不同尺度的布匹或多种布匹同时检测缺乏考虑,无法同时适用于不同尺度的布匹检测,从而导致误检,鲁棒性低。本论文针对上述问题提出了基于Gabor和视觉信息的布匹瑕疵检测方法。该方案构建了布匹瑕疵数据库,根据人眼能够快速准确的从不同尺度与不同纹理布匹图像中鉴别瑕疵区域的现象,将人眼视觉引入布匹瑕疵检测中,从而提出了改进的多通道Gabor布匹瑕疵检测方法与基于改进视觉显著性的布匹瑕疵检测方法。本文的主要研究内容如下:⑴针对现阶段布匹检测方法依赖于无瑕疵布匹图像的先验知识学习,无法同时对不同尺度的纺织工厂线采集到的布匹图像进行有效的瑕疵检测的问题,提出一种基于改进多通道Gabor的布匹瑕疵检测方法。该方法采用改进多通道Gabor滤波器对布匹图像进行滤波,通过子块计分方法对多通道Gabor滤波结果进行选择,将选择出的多通道进行融合,通过阈值分割,得到瑕疵区域。实验结果表明该方法与传统的布匹瑕疵检测方法形态学方法、MRF方法比较,可以有效提高对典型的布匹瑕疵检测的准确率。⑵针对改进多通道Gabor的布匹瑕疵检测检测效率低的问题,提出一种改进视觉显著性的布匹瑕疵检测方法。本文所述方法在经典的视觉显著性模型基础上得到布匹图像的自底向上的显著性特征,并提出一种自顶向下的熵、能量显著性特征计算方法。该方法通过对显著性特征图进行融合,使用最大类间方差法进行分割显著性特征图,得到布匹图像视觉显著性区域。实验结果表明该方法与⑴中比较,在保证识别率的基础上,能够快速检测典型布匹瑕疵。
[Abstract]:The traditional method of fabric defect detection depends on the manual flaw detection of finished fabric with the naked eyes of workers, while the existing computer vision fabric defect detection method is mainly based on the features of the image extraction of unblemished fabric. Thus the fabric defect detection is carried out. This kind of method needs to compare and study the image of the blemeless cloth, and has strict requirements for the experimental images of cloth collected at a specific angle. The above scheme relies on the prior knowledge learning of the imperfections of fabric images, and it lacks consideration of the simultaneous detection of different sizes of cloth or fabrics, and can not be applied to fabric detection of different scales at the same time, which leads to false detection and low robustness. In this paper, a cloth defect detection method based on Gabor and visual information is proposed. According to the phenomenon that the human eyes can quickly and accurately distinguish the defective areas from different scales and different textures of cloth images, the human visual system is introduced into fabric defect detection. An improved multi-channel Gabor fabric defect detection method and a fabric defect detection method based on improved visual significance are proposed. The main research contents of this paper are as follows: 1. Aiming at the present fabric detection method, which depends on the prior knowledge learning of the image of the flawless cloth, The defect detection method based on improved multi-channel Gabor is proposed, which can not detect the defects of fabric images collected from textile factory lines of different scales at the same time. The improved multi-channel Gabor filter is used to filter the fabric image, the sub-block scoring method is used to select the multi-channel Gabor filtering results, the selected multi-channel is fused, and the defect region is obtained by threshold segmentation. The experimental results show that this method is compared with the traditional cloth defect detection method, morphology method and MRF method. It can effectively improve the accuracy of the typical fabric defect detection. 2 in order to improve the efficiency of fabric defect detection based on multi-channel Gabor, an improved visual significant fabric defect detection method is proposed. Based on the classical visual saliency model, the bottom-up salience features of cloth images are obtained, and a top-down entropy and energy salience feature calculation method is proposed. In this method, the significant feature map is fused and the significant feature map is segmented by using the maximum inter-class variance method, and the visual significance region of the cloth image is obtained. The experimental results show that the proposed method can quickly detect typical fabric defects on the basis of guaranteed recognition rate.
【学位授予单位】：江苏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TS101.97;TP391.41

【参考文献】