文献类型：期刊文献

中文题名：基于深度卷积神经网络的图像哈希认证方法

英文题名：Image Hashing Authentication Method Based on Deep Convolution Neural Network

作者：蒋翠玲[1];庞毅林[1];林家骏[1];康周茂[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2018

卷号：46

期号：5

起止页码：53

中文期刊名：华南理工大学学报（自然科学版）

外文期刊名：Journal of South China University of Technology(Natural Science Edition)

收录：CSTPCD;;EI(收录号：20185206305301);Scopus;北大核心:【北大核心2017】；CSCD:【CSCD2017_2018】；

基金：国家自然科学基金资助项目(61371150)~~

语种：中文

中文关键词：信息安全;卷积神经网络;图像哈希;区分性;鲁棒性

外文关键词：information security;convolutional neural network;image hashing;discrimination;robustness

摘要：提出了一种基于深度卷积神经网络的图像哈希认证方法.首先构建深度卷积神经网络AlexNet模型,训练该网络得到预定的网络性能;然后由训练好的卷积神经网络提取图像的特征,最后生成图像哈希序列用于图像内容的篡改认证.实验结果表明,相比同类方法,文中提出的图像哈希认证方法具有较高的区分性,同时对随机攻击、JPEG压缩、加性高斯噪声等具有可接受的鲁棒性. ROC曲线表明,文中提出的方法实现了区分性与鲁棒性的均衡.
This paper presents a scheme of deep convolution network for image hashing authentication.First,the AlexNet model of deep convolution network is constructed and the given network performance is achieved through training.Then,the trained network is used to extract image features and generate image-hashing series for content authentication.The experimental results show that in comparison with other methods,the proposed method has a higher discrimination and an acceptable robustness against content-preserving operations such as random attack,rotation,JPEG compression,and additive Gaussian noise.Receiver operating characteristics(ROC)curve comparison demonstrates that the proposed method is able to attain a desirable compromise between the robustness and discrimination.