A JPEG image steganalysis method based on depth extraction of steganalytic noise.

CN115410037BActive Publication Date: 2026-05-26Chinese People's Liberation Army Cyberspace Force Information Engineering University
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Chinese People's Liberation Army Cyberspace Force Information Engineering University
Filing Date
2022-08-19
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing deep learning steganalysis methods are insufficient in JPEG image detection accuracy, and traditional high-pass filters rely on human experience and cannot completely suppress the influence of image content on detection.

Method used

A deep steganalysis noise extraction network is constructed and trained using supervised learning. A steganalysis image detection network is built using the Swing Transformer backbone network. Through high-dimensional feature extraction and steganalysis noise learning, the loss and model evaluation index P are calculated by combining L1_Loss and the network is used to extract and classify steganalysis noise.

Benefits of technology

It improves the accuracy of JPEG image steganalysis, reduces the impact of image content on detection, and maintains detection performance even under carrier mismatch conditions, outperforming existing methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115410037B_ABST
    Figure CN115410037B_ABST
Patent Text Reader

Abstract

This invention discloses a JPEG image steganalysis method based on deep extraction of steganalytic noise. The method first designs a deep steganalytic noise extraction network structure and trains it using supervised learning. Then, it selects the optimal network using a designed model evaluation metric. Finally, it fuses the noise extraction network with an image classification network to train a hidden image detection network, and performs hidden image detection based on this network. The steganalysis method proposed in this invention can extract steganalytic noise more accurately, reducing the impact of image content on steganalysis, and achieves better detection results compared to typical deep learning-based JPEG image steganalysis methods.
Need to check novelty before this filing date? Find Prior Art