DCT-domain deep convolutional neural networks for multiple JPEG compression classification
作者：
Highlights：
•
• Novel CNN based system to classify an image according to number of JPEG compressions.
• Design of appropriate pre-processing stage before feeding the image data to CNN.
• Directly utilizes JPEG bit stream and histogram, reduces the content dependence.
• Robust performance even at larger number of compression stages (Section 4.8).
• 91% average accuracy for patches of size 128 × 128 thus potential for forgery detection.
摘要
•Novel CNN based system to classify an image according to number of JPEG compressions.•Design of appropriate pre-processing stage before feeding the image data to CNN.•Directly utilizes JPEG bit stream and histogram, reduces the content dependence.•Robust performance even at larger number of compression stages (Section 4.8).•91% average accuracy for patches of size 128 × 128 thus potential for forgery detection.
论文关键词：Image forensics,Compression forensics,Deep convolutional neural network (CNN),JPEG forensics,Multiple compression,Forgery detection
论文评审过程：Received 3 December 2017, Revised 23 March 2018, Accepted 29 April 2018, Available online 16 May 2018, Version of Record 30 May 2018.
论文官网地址：https://doi.org/10.1016/j.image.2018.04.014
