| 摘要: |
| 目的 研究印刷标志套准机器快速和高精度的识别方法。方法 提取印刷标志图像的灰度共生矩阵表达其纹理特征， 采用Adaboost分类器对印刷标志套准图像进行识别， 以判断印刷是否套准。结果提取出了印刷标志图像的能量、 熵、 惯性矩、 相关度等的均值和标准差的8维图像纹理特征。为了比较不同分类器的分类性能， 分别得出了Adaboost、 K近邻、 贝叶斯、 支持向量机、 Fisher和决策树对印刷标志图像纹理特征的分类准确率和分类时间。结论 采用文中方法， 印刷标志图像套准识别准确率达到97.5%， 分类时间达到0.022 377 s， 优于其他的分类方法。 |
| 关键词: 印刷套准 纹理特征 熵 Adaboost |
| DOI： |
| 分类号: |
| 基金项目:国家基础研究项目 （2011CB013104）；国家自然科学联合基金 （U1134004， 51275093）；高等教育学科建设重点项目 （2012CXZD0020）；广 东 省 专 项 研 发 项 目 LED 行 业 基 金（2011A081301001， 2012A08030300）；省 部 产 学 研 结 合 项 目（2012B091000028）；数控一代项目 （2012B011300027） |
|
| Printing Registration Recognition |
|
JIAN Chuan-xia, GAO Jian, LI Ke-tian, AO Yin-hui
|
|
Guangdong University of Technology， Guangzhou 510006， China
|
| Abstract: |
| The aim of this work was to study the rapid and highly accurate recognition method for printing registration based on machine vision. Gray level co-occurrence matrix of the printing marks image was extracted to represent its texture features. Adaboost classifier was used to recognize printing marks images, to check the accuracy of printing registration. 8-dimentioal texture vectors including the means and standard deviations of energy, entropy, moment of inertia and correlation in the printing marks images were extracted. To compare the classification performance of different types of classifiers, the accuracy and runtime of classifying these 8-dimentioal vectors were obtained using Adaboost, K-Nearest Neighbor, Naivebayes, Support Vector Machine, Fisher and Decision Tree. The recognition rate of 97.5 % and the classification runtime of 0.022 377 s could be achieved using the proposed method, superior to other classification methods. |
| Key words: printing registration texture entropy Adaboost |
