基于改进算法的卷积神经网络的单人离线手写签名真伪鉴定  

Single-person Offline Handwritten Signature Authenticity Verification Based on an Improved Algorithm of Convolutional Neural Network

作　　者：谢明风 杨圣[1] 汪后云 周翔 汪琨 Xie Mingfeng

机构地区：[1]中国科学技术大学工程科学学院精密机械与精密仪器系,安徽合肥230031

出　　处：《工业控制计算机》2020年第12期4-5,8,共3页Industrial Control Computer

摘　　要：为了解决深度学习签名鉴定中输入的数据集较少、同时鉴定准确率要求高的问题,提出了单人离线手写签名识别方法,将预处理后的图形输入优化Triplet loss算法后的卷积神经网络并更新参数,最后通过待预测真伪图形与训练中心点的欧氏距离大小来进行鉴定签名的真伪,实现了在少量数据集的情况下获得了95%以上的准确率的目标。In order to solve the small amount of signature sample data and high accuracy requirements,this paper proposes a single-person offline handwritten signature recognition method.The pre-processed graphics are input into the convolutional neural network after the optimization algorithm and the parameters are updated,and finally pass the prediction The Euclidean distance between the authenticity graph and the training center point is used to verify the authenticity of the signature,achieving the goal of obtaining an accuracy of more than 95%in the case of a small number of data sets.

关 键 词：卷积神经网络 手写签名真伪鉴定 Triplet loss算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]