摘要
为提高真伪卷烟商标纸鉴别的准确性和效率,降低鉴别的经验要求和主观性,提出一种基于特征点配准的真伪卷烟商标纸鉴别方法。使用一致的标准扫描采集卷烟样品图像,基于尺度不变特征转换算法提取图像特征点,通过特征匹配和基于单应性变换的图像配准获取判别预测变量。采用逻辑回归、梯度提升分类决策树算法构建二元分类模型对图像样本进行训练和评估。在64个卷烟规格、2 918个样本数据集上进行实验,该方法准确率高于95%。通过对比实验验证了该方法的稳定性和有效性。
In order to improve the accuracy and efficiency of the identification of genuine and fake cigarette label paper,and reduce the experience requirements and subjectivity of identification,this paper proposes a method for identifying cigarette label paper based on feature point registration.Sample images were acquired based on a unified image acquisition standard,and discriminant predictors were acquired through feature point extraction and description based on SIFT algorithm,feature matching and image registration based on homograph transformation.Logistic regression algorithm and gradient boosting classification decision tree algorithm were selected to construct a binary classification model for training and evaluation.We performed experimental evaluation on 64 cigarette specifications and 2918 sample data sets.The results show that the accuracy of the proposed identification method is higher than 95%.The stability and effectiveness of the proposed identification method are verified through a comparative experiment.
作者
冯伟华
王锐
宗国浩
赵志成
罗泽
周明珠
李晓辉
邢军
Feng Weihua;Wang Rui;Zong Guohao;Zhao Zhicheng;Luo Ze;Zhou Mingzhu;Li Xiaohui;Xing Jun(Zhengzhou Tobacco Research Institute of CNTC,Zhengzhou 450001,Henan,China;China National Tobacco Quality Supervision&Test Center,Zhengzhou 450001,Henan,China;Computer Network Information Center,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《计算机应用与软件》
北大核心
2024年第3期194-201,共8页
Computer Applications and Software
基金
国家烟草专卖局科技重大专项(110201901029(SJ-08))。
关键词
卷烟商标纸
真伪鉴别
特征点
图像配准
模型算法
机器学习
Cigarette label paper
Identification of genuine and fake
Feature point Image registration
Model algorithm
Machine Learning
