基于样本生成机制的包装印刷品真伪防御性判别  

Defensive Identification of Authenticity of Packaging Printed Matter Based on Sample Generation Mechanism

作　　者：乔俊伟 宛东 QIAO Junwei;WAN Dong(Key Lab of Intelligent and Green Flexographic Printing,Shanghai Publishing and Printing College,Shanghai 200093,China;College of Communication and Art Design,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海出版印刷高等专科学校智能与绿色柔版印刷重点实验室,上海200093 [2]上海理工大学出版印刷与艺术设计学院,上海200093

出　　处：《包装学报》2023年第5期31-37,共7页Packaging Journal

基　　金：上海市东方学者特聘教授基金资助项目(TP2022126);国家新闻出版署智能与绿色柔版印刷重点实验室开放招标课题(ZBKT202301)。

摘　　要：借助手机实现假包装印刷品的识别,实现主动确权,从而达到更加精确的真假包装印刷品判别。针对造假样品极度缺乏的情况,提出基于记忆增强DCGAN样本生成算法来实现假样本扩增,再利用孪生并行注意力卷积神经网络(S-PA-CNN)实现包装印刷品真伪的防御性判别。在1种印刷纸张、2种拍摄光源和2种拍摄手机组合的4种开放场景中,拍摄多个真包装印刷品和少量的假包装印刷品图像,用基于记忆增强DCGAN样本生成算法扩增假包装印刷品样本,建立数据集。实验结果表明:数据扩增后,假样本和真样本数量差不多时,S-PA-CNN的检测准确率在97%以上。本文数据扩增方法能够提升网络模型的真样本特征识别能力、细粒度判别精度和泛化能力。With the help of mobile phones,the identification of forged packaging printed matter can be realized,and the active confirmation can be realized,so as to achieve more accurate identification of true and fake packaging printed matter.In the case of extremely lack of fake samples,a sample generation algorithm based on DCGAN memory enhancement was proposed to realize the amplification of fake samples,and the attention convolutional neural network was used to distinguish the authenticity of packaged printed matter defensively.Several groups of authentic packaging printed matter and a very small number of counterfeit packaging printed matter images were shot in four open scenes,which consisted of one kind of printing paper,two kinds of shooting light source and two kinds of shooting mobile phone.The sample of counterfeit packaging printed matter was generated based on DCGAN memory enhancement algorithm,and the algorithm research data set was established.The experimental results show that the discriminant accuracy of S-PA-CNN twin attention convolutional neural network based on amplified data set is more than 97%.The experimental results show that the proposed data amplification method can further improve the authenticity feature recognition ability of the network model,improve thefine-grained discrimination accuracy,and enhance the generalization ability.

关 键 词：包装印刷品 DCGAN 编解码器 卷积神经网络 微小篡改 真伪判别 

分 类 号：TB482[一般工业技术—包装工程] TP391[自动化与计算机技术—计算机应用技术]