摘要
为了进一步提高篡改检测率,提出了一种基于卷积神经网络(CNN)的图像篡改检测算法(SCNN)。虽然CNN能够直接从数据中学习分类特征,但是在其标准形式中,它倾向于学习与图像内容相关的特征。为了克服图像取证任务中的这一问题,提出了一种新的图像预处理层来共同抑制图像内容并自适应地学习特征。使用CASIA V2.0中75%的图像对SCNN进行训练和验证,并使用CASIA V2.0中的其余图像和哥伦比亚未压缩数据集中的所有图像进行测试。实验结果表明,本文SCNN框架明具有一定有效性及鲁棒性。
In order to further improve the tamper detection rate,this paper proposes an image tampering detection algorithm based on Convolutional Neural Network(CNN),which is called the spatial domain CNN model(SCNN). Although CNNs can learn classification features directly from data,in their standard form they tend to learn features related to image content. In order to overcome this problem in image forensics tasks,a new image preprocessing layer is proposed to jointly suppress image content and adaptively learn features. The SCNN is trained and validated using 75% of the images in CASIA v2.0,and all images in the uncompressed data set in Colombia are tested using the remaining images in CASIA v2.0.A large number of experiments show that the proposed SCNN framework has certain effectiveness and robustness.
作者
钟辉
李红
李振建
欧阳若川
闫冬梅
ZHONG Hui;LI Hong;LI Zhen-jian;OUYANG Ruo-chuan;YAN Dong-mei(Management Center of Big Data and Network,Jilin University,Changchun 130012,China)
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2020年第4期1428-1434,共7页
Journal of Jilin University:Engineering and Technology Edition
基金
吉林省省级产业创新专项资金项目(2017C031-4)
赛尔网络下一代互联网技术创新项目(NGII20180104,NGII20181202)。
关键词
多媒体取证
卷积神经网络
篡改检测
拼接图像
特征提取
multimedia forensics
convolutional neural networks
image manipulation detection
splicing image
feature extraction
