基于计算机视觉和机器学习的真伪卷烟包装鉴别

Authentication of packeted cigarettes based on computer vision and machine learning

为解决人工鉴别真伪卷烟效率低、主观性强等问题,基于计算机视觉和机器学习建立了一种真伪卷烟包装鉴别模型。利用计算机视觉对卷烟包装进行图像处理和特征向量提取,分别以相似性度量模型、机器学习模型对特征向量进行分类并判定卷烟真伪。相似性度量模型采用曼哈顿距离模型进行分类,并对高斯双边滤波函数进行了参数优化;机器学习模型则以图像分块为基础,确定最优分块数量和面积。以“中华(软)”“玉溪(软)”“钻石(荷花)”3个卷烟品牌共603个真伪样品为对象,分别采用两种模型进行判定,结果表明:相似性度量模型在“玉溪(软)”样品测试集的准确率为96.17%;机器学习模型在“中华(软)”“玉溪(软)”“钻石(荷花)”3个样品测试集的准确率分别为98.99%、96.61%和100%。机器学习模型与相似性度量模型相比较,具有较好的迁移能力和鲁棒性,适用于卷烟真伪鉴别样品量大、品类多、图像复杂等情况。该方法可为提高真伪卷烟鉴别效率和准确率提供技术支持。

To pursue high efficiency and high accuracy in authentication of cigarette packets,an authentication model was established based on computer vision and machine learning.Computer vision was adopted to process the images of cigarette packet and extract feature vectors,a similarity measurement model and a machine learning model were used separately to classify the feature vectors and to discriminate between genuine and fake cigarettes.The similarity measurement model made classification with model of Manhattan distance,and optimized the parameters in Gauss bilateral filter function.Machine learning model was based on image segmentation,and determined the optimal amount and area of segment.A total of 603 samples of three cigarette brands,“CHUNGHWA”,“YUXI”and“HEHUA”were identified by the two models separately.The results showed that the accuracy of the similarity measurement model for the test set of brand“YUXI”was 96.17%,and the accuracies of the machine learning model for the test sets of brands“CHUNGHWA”,“YUXI”and“HEHUA”reached 98.99%,96.61%and 100%respectively.Comparing with the similarity measurement model,the machine learning model has better migration ability and robustness,it is suitable for the authentication of cigarette samples of large amount,multiple categories and complex images.This method provides a technical support for improving the efficiency and accuracy of cigarette authentication.