同源激光雕刻塑胶印章印文的计算机辅助检验方法
|
摘要
为利用Resnet50网络对三枚同源激光雕刻塑胶印章盖印的印文进行鉴别,通过使用pytorch框架运行Resnet50开源网络模型,制作激光扫描速度分别为210、280、350的三枚同源塑胶印章,盖印15000枚印文,扫描得到印文图像作为训练数据,盖印100枚扫描作为测试数据。研究了以改变训练样本量得到的Resnet50网络模型对同源印章印文鉴别的准确率。结果表明:网络最终收敛能成功学习特征,以150-15000枚的印文图像训练得到的准确率能达到95.66%-99.789%,对300枚印文图像测试得到94.6%以上的准确率。可见利用Resnet50对同源激光雕刻塑胶印章印文进行鉴别是一种可行的方法。
In order to identify the seals stamped by three homologous laser engraved plastic stamps using the Resnet50 network, three reciprocal plastic stamps with laser scanning speeds of 210, 280 and 350 were produced by running the Resnet50 open source network model using the pytorch framework. A stamp of 15,000 stamps was scanned to obtain a printed image as training data, and 100 scans were stamped as test data. The accuracy of the Resnet50 network model obtained by changing the training sample size for the identification of the same-same seal is studied.The results show that the final convergence of the network can successfully learn the characteristics. The accuracy of the training of 150-15 000 printed images can reach 95.66%-99.789%, and the accuracy of 94.6% of the 300 printed images is obtained. It is concluded that it is a feasible method to identify the homologous laser engraving plastic seal stamp by using Resnet50.
In order to identify the seals stamped by three homologous laser engraved plastic stamps using the Resnet50 network, three reciprocal plastic stamps with laser scanning speeds of 210, 280 and 350 were produced by running the Resnet50 open source network model using the pytorch framework. A stamp of 15,000 stamps was scanned to obtain a printed image as training data, and 100 scans were stamped as test data. The accuracy of the Resnet50 network model obtained by changing the training sample size for the identification of the same-same seal is studied.The results show that the final convergence of the network can successfully learn the characteristics. The accuracy of the training of 150-15 000 printed images can reach 95.66%-99.789%, and the accuracy of 94.6% of the 300 printed images is obtained. It is concluded that it is a feasible method to identify the homologous laser engraving plastic seal stamp by using Resnet50.
引文
[1]贾治辉.利用激光、光敏印章机伪造印章印文的鉴定[J].中国司法鉴定, 2009,(5):20-24
[2]王长亮,周光磊,顾会泳, et al.同源激光雕刻橡塑印章印文的鉴别方法[J].刑事技术, 2012, 37(6):42-44
[3]崔岚.同源印章印文鉴别方法研究[J].中国刑警学院学报,2012(3):49-51
[4]王勇.常见印章制作的方法及伪造印章的检验鉴定[J].价值工程, 2012, 31(5):290-291
[5]朱毅,肖华土,李彪.刍议同源印章印文检验[J].广东公安科技, 2018, 26(1):21-22
[6]林红,王长亮,周光磊,等.同源光敏印章印文鉴别方法的研究[J].中国司法鉴定, 2012(6):64-66
[7] He Kaiming,Zhang Xiangyu,Ren Shaoqing,et al. Deep Residual Learning for Image Recognition[J]. 2015
[8]杨文雅,宋广乐,崔超然,等.基于语义感知的图像美学质量评估方法[J].计算机应用, 2018, 38(11):3216-3220
[9]徐子豪,黄伟泉,王胤.基于深度学习的监控视频中多类别车辆检测[J].计算机应用, 2019,39(3):700-705
[10]江和平,沈振康.基于局部交叉熵的图像匹配跟踪算法[J].红外与激光工程,2005,34(6):729-732
[2]王长亮,周光磊,顾会泳, et al.同源激光雕刻橡塑印章印文的鉴别方法[J].刑事技术, 2012, 37(6):42-44
[3]崔岚.同源印章印文鉴别方法研究[J].中国刑警学院学报,2012(3):49-51
[4]王勇.常见印章制作的方法及伪造印章的检验鉴定[J].价值工程, 2012, 31(5):290-291
[5]朱毅,肖华土,李彪.刍议同源印章印文检验[J].广东公安科技, 2018, 26(1):21-22
[6]林红,王长亮,周光磊,等.同源光敏印章印文鉴别方法的研究[J].中国司法鉴定, 2012(6):64-66
[7] He Kaiming,Zhang Xiangyu,Ren Shaoqing,et al. Deep Residual Learning for Image Recognition[J]. 2015
[8]杨文雅,宋广乐,崔超然,等.基于语义感知的图像美学质量评估方法[J].计算机应用, 2018, 38(11):3216-3220
[9]徐子豪,黄伟泉,王胤.基于深度学习的监控视频中多类别车辆检测[J].计算机应用, 2019,39(3):700-705
[10]江和平,沈振康.基于局部交叉熵的图像匹配跟踪算法[J].红外与激光工程,2005,34(6):729-732