印刷品自动光学检测伪缺陷去除方法

Algorithm of Pseudo-defects Elimination in Automatic Optical Inspection of Printing Product

目的为了解决使用差影法进行印刷品表面缺陷自动光学检测存在伪缺陷从而造成误检的问题,提出一套通过提高配准精度和空间滤波来减少伪缺陷的方法。方法使用连续域蚁群算法求全局最优解,改进基于轮廓的模板匹配算法配准精度,减少伪缺陷的产生;配准时的仿射变换会产生轮廓伪影,根据伪影的分布特点将图像切分成轮廓区和非轮廓区。分别使用不同的空间滤波方法削弱伪缺陷,再使用阈值分割将其剔除。结果在实验环境中,连续域蚁群算法改进的基于轮廓的模板匹配配准方法,可精确到亚像素级,配准率为92%;使用空间滤波剔除伪缺陷后进行缺陷检测,缺陷误检率为0,漏检率为3%;印刷品表面缺陷平均检测时间为1.05s,最长时间小于1.5s。结论该研究改善了差影法在自动光学检测中的效果,方法快速、有效、易于实现;在降低了缺陷误检率的同时不会造成大量漏检;满足在线检测的要求,可用于实际工业生产。

The work aims to propose a set of methods to eliminate the pseudo-defects by improving the registration accuracy and spatial filtering to reduce false inspection during automatic optical inspection(AOI)of printing product by difference image method.Firstly,the registration accuracy of contour-based template matching was improved by ant colony optimization for continuous domains for global optimal solution to reduce the pseudo-defects.Secondly,the image was cut into two parts:contour zone and non-contour zone according to the distribution characteristics of the artifacts caused by affine transformation during automatic registration.After weakened by corresponding spatial filtering,the pseudo-defects were eliminated with threshold segmentation.Under experimental environment,the contour-based template matching improved by ant colony optimization for continuous domains could be accurate to the sub-image level,and the registration rate was 92%.The false inspection rate was 0 and the missed inspection rate was 3%after the pseudo-defects were eliminated with spatial filtering.The average inspection time was 1.05 s and the maximum was less than 1.5 s.The proposed method improves the effect of image difference in AOI and can be realized fast,effectively and easily.It reduces the false inspection rate of defects without causing missed inspection and can be used in actual industrial production while meeting the requirements of online inspection.