基于机器视觉的印刷品在线检测方法研究
文献类型：学位论文
| 作者 | 黄为 |
| 学位类别 | 工学博士 |
| 答辩日期 | 2010-11-25 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 王云宽 |
| 关键词 | 印刷在线视觉检测 错帖检测系统 纹理图像分析 Gabor滤波器 条码定位 图像配准 Vision based printing defect detection Page-checking system texture image analysis Gabor filter bar-code location image registration |
| 其他题名 | Research on Method of Vision Based Printed Matter Online Detection |
| 学位专业 | 控制理论与控制工程 |
| 中文摘要 | 随着印刷行业飞速发展，印刷技术的不断进步，传统的印刷品检测方法已无法满足印刷生产的高度自动化、印刷品种类和图样多样化的要求。基于机器视觉的检测技术因其检测速度快、检测精度高、非接触式等特性，将会是未来印刷在线检测的必然趋势。机器视觉检测系统的好坏，关键在于检测算法的性能，近年来，尽管国内外的许多学者和科研机构提出了许多关于印刷的检测理论和实时检测算法，但是在实际应用中，仍存在一些问题有待解决。 本文对基于机器视觉印刷品在线检测领域中的检测算法进行了研究和探索，主要研究内容包括：不变性特征提取，鲁棒性的检测与定位算法等，并将这些算法应用于错帖检测系统和电子监管码检测系统中，论文所完成的主要工作有： 1) 针对书刊自动装订中传统的错贴检测方法的不足，并考虑到在线实时检测的实际工程需求，提出了一种检测精度高而且运行速度快的书帖检测算法。该算法首先提取图像的定尺度SIFT特征向量，然后结合广义霍夫变换和最小二乘法对书帖的位置进行定位，最后通过计算模板书帖图像与待检测书帖图像的相关系数实现对书帖的检测。实验表明，在针对实际图像的检测中本文算法可以检测出书帖图像的有无和摆放方向是否正确，满足了实际工业系统的实时性和准确度要求。 2) 针对书页图像在检测时，存在偏移、旋转等刚性变化，提出了一种结合形态学与Fast Corner-9检测技术的书页图像快速配准方法。该方法依据FastCorner-9技术快速提取图像特征点，然后通过梯度直方图建立每个特征点的方向，针对特征点的位置和方向精度不够的问题，设计一种基于形态学滤波的多模板局部特征快速匹配方法。该方法结合特征点的正负极性，特征点主方向，以及多模板的局部特征，实现了快速的特征点匹配，然后利用RANSAC去除错误的匹配点，最终完成书页图像的快速配准。实验表明，该方法在采样图像偏移、角度旋转时，能快速配准模板图像和采样图像，快速建立采样图像和模板 图像的映射关系。 3) 针对印刷在线检测中条形码的定位问题，提出了利用条码结构特征有效且快速定位、分割图像中条形码的方法。该方法对含有条形码的印刷图像进行多个方向上的Gabor滤波得到二维信号的幅值纹理图像，二值化每个幅值纹理图像，利用形态学膨胀原理，得到一些条码的候选区域，然后通过连通域分析和条码区域几何特征分析得到条形码区域，最后利用投影分析精确分割条码区域。实验结果表明,该算法可以在复杂背景下快速定位和精确分割条码。 |
| 英文摘要 | With the rapid development of the printing industry and the printing technology, the traditional printing detection methods have been unable to meet the requirements for the high automation level of the printing instruments and the diversity of the printing types and patterns. Because of the advantages, such as the high detection speed and precision and the non-contact property, the detection based on machine vision will be the inevitable trend of the future for the printed matter online detection. The key of a successful inspection system based on machine vision is its detection performance. Although many researchers and related research institutes proposed various online detection methods for the printed matter area, there are still many difficult problems. This thesis is focused on some key techniques in the fields of printing detection system based on machine vision, such as the invariant features, the robust algorithms and so on. The main work including: 1) Page-checking plays an important role in the automatic bookbinding process, which is essential for both the efficiency of the automatic bookbinding and the quality of the printed books. We propose a fast and accurate page-checking algorithm to meet the requirements of the on-line inspection, which can be summarized as follows: the constant scale SIFT features of the page image are firstly extracted, then the sampling objects are localized based on the General Hough Transform and Least Square Methods, and finally the cross correlation between the sampling objects and the model objects is calculated to determine the inspection result. Experimental results illustrate that the proposed algorithm can achieve good performance on inspecting the content and orientation of the page image in real time. 2) Considering the problem of the translation and rotation of the page picture when detection, a registration method based on gray scale morphology combined with the Fast Corner-9 is proposed. The first step of the proposed method is to detect the corner points with the Fast Corner-9, and build the main direction for each corner point. To deal with the bias of the corner points direction and other issues, a fast matching method based on morphological multi-template local descriptor is designed in the second step. At last, the registration is finished through the map of the right matched corner points filtered by the RANSAC. Experiment results illustrate that the proposed algorithm could quickly build the m... |
| 语种 | 中文 |
| 其他标识符 | 200718014628008 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/6308]  |
| 专题 | 毕业生_博士学位论文 |
| 推荐引用方式 GB/T 7714 | 黄为. 基于机器视觉的印刷品在线检测方法研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2010. |
入库方式： OAI收割
来源：自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。