Steganalysis based on awareness of selection-channel and deep learning

Jianhua Yang, Kai Liu, Xiangui Kang, Edward Wong, Yunqing Shi

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Abstract

    Recently, deep learning has been used in steganalysis based on convolutional neural networks (CNN). In this work, we propose a CNN architecture (the so-called maxCNN) to use the selection channel. It is the first time that the knowledge of the selection channel has been incorporated into CNN for steganalysis. The proposed method assigns large weights to features learned from complex texture regions while assigns small weights to features learned from smooth regions. Experimental results on the well-known dataset BOSS-base have demonstrated that the proposed scheme is able to improve detection performance, especially for low embedding payloads. The results have shown that with the ensemble of maxCNN and maxSRMd2+EC, the proposed method can obtain better performance compared with the reported state-of-the-art on detecting WOW embedding algorithm.

    Original languageEnglish (US)
    Title of host publicationDigital Forensics and Watermarking - 16th International Workshop, IWDW 2017, Proceedings
    PublisherSpringer Verlag
    Pages263-272
    Number of pages10
    Volume10431 LNCS
    ISBN (Print)9783319641843
    DOIs
    StatePublished - 2017
    Event16th International Workshop on Digital Forensics and Watermarking, IWDW 2017 - Magdeburg, Germany
    Duration: Aug 23 2017Aug 25 2017

    Publication series

    NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    Volume10431 LNCS
    ISSN (Print)0302-9743
    ISSN (Electronic)1611-3349

    Other

    Other16th International Workshop on Digital Forensics and Watermarking, IWDW 2017
    CountryGermany
    CityMagdeburg
    Period8/23/178/25/17

    Fingerprint

    Steganalysis
    Neural Networks
    Neural networks
    Assign
    Network Architecture
    Network architecture
    Texture
    Ensemble
    Textures
    Experimental Results
    Learning
    Awareness
    Deep learning

    Keywords

    • Adaptive steganography
    • Convolutional neural networks (CNN)
    • Selection-channel
    • Steganalysis

    ASJC Scopus subject areas

    • Theoretical Computer Science
    • Computer Science(all)

    Cite this

    Yang, J., Liu, K., Kang, X., Wong, E., & Shi, Y. (2017). Steganalysis based on awareness of selection-channel and deep learning. In Digital Forensics and Watermarking - 16th International Workshop, IWDW 2017, Proceedings (Vol. 10431 LNCS, pp. 263-272). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10431 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-64185-0_20

    Steganalysis based on awareness of selection-channel and deep learning. / Yang, Jianhua; Liu, Kai; Kang, Xiangui; Wong, Edward; Shi, Yunqing.

    Digital Forensics and Watermarking - 16th International Workshop, IWDW 2017, Proceedings. Vol. 10431 LNCS Springer Verlag, 2017. p. 263-272 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10431 LNCS).

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Yang, J, Liu, K, Kang, X, Wong, E & Shi, Y 2017, Steganalysis based on awareness of selection-channel and deep learning. in Digital Forensics and Watermarking - 16th International Workshop, IWDW 2017, Proceedings. vol. 10431 LNCS, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10431 LNCS, Springer Verlag, pp. 263-272, 16th International Workshop on Digital Forensics and Watermarking, IWDW 2017, Magdeburg, Germany, 8/23/17. https://doi.org/10.1007/978-3-319-64185-0_20
    Yang J, Liu K, Kang X, Wong E, Shi Y. Steganalysis based on awareness of selection-channel and deep learning. In Digital Forensics and Watermarking - 16th International Workshop, IWDW 2017, Proceedings. Vol. 10431 LNCS. Springer Verlag. 2017. p. 263-272. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-64185-0_20
    Yang, Jianhua ; Liu, Kai ; Kang, Xiangui ; Wong, Edward ; Shi, Yunqing. / Steganalysis based on awareness of selection-channel and deep learning. Digital Forensics and Watermarking - 16th International Workshop, IWDW 2017, Proceedings. Vol. 10431 LNCS Springer Verlag, 2017. pp. 263-272 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
    @inproceedings{52c8b6bdd3dc4ad48780e8796b865d8e,
    title = "Steganalysis based on awareness of selection-channel and deep learning",
    abstract = "Recently, deep learning has been used in steganalysis based on convolutional neural networks (CNN). In this work, we propose a CNN architecture (the so-called maxCNN) to use the selection channel. It is the first time that the knowledge of the selection channel has been incorporated into CNN for steganalysis. The proposed method assigns large weights to features learned from complex texture regions while assigns small weights to features learned from smooth regions. Experimental results on the well-known dataset BOSS-base have demonstrated that the proposed scheme is able to improve detection performance, especially for low embedding payloads. The results have shown that with the ensemble of maxCNN and maxSRMd2+EC, the proposed method can obtain better performance compared with the reported state-of-the-art on detecting WOW embedding algorithm.",
    keywords = "Adaptive steganography, Convolutional neural networks (CNN), Selection-channel, Steganalysis",
    author = "Jianhua Yang and Kai Liu and Xiangui Kang and Edward Wong and Yunqing Shi",
    year = "2017",
    doi = "10.1007/978-3-319-64185-0_20",
    language = "English (US)",
    isbn = "9783319641843",
    volume = "10431 LNCS",
    series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
    publisher = "Springer Verlag",
    pages = "263--272",
    booktitle = "Digital Forensics and Watermarking - 16th International Workshop, IWDW 2017, Proceedings",
    address = "Germany",

    }

    TY - GEN

    T1 - Steganalysis based on awareness of selection-channel and deep learning

    AU - Yang, Jianhua

    AU - Liu, Kai

    AU - Kang, Xiangui

    AU - Wong, Edward

    AU - Shi, Yunqing

    PY - 2017

    Y1 - 2017

    N2 - Recently, deep learning has been used in steganalysis based on convolutional neural networks (CNN). In this work, we propose a CNN architecture (the so-called maxCNN) to use the selection channel. It is the first time that the knowledge of the selection channel has been incorporated into CNN for steganalysis. The proposed method assigns large weights to features learned from complex texture regions while assigns small weights to features learned from smooth regions. Experimental results on the well-known dataset BOSS-base have demonstrated that the proposed scheme is able to improve detection performance, especially for low embedding payloads. The results have shown that with the ensemble of maxCNN and maxSRMd2+EC, the proposed method can obtain better performance compared with the reported state-of-the-art on detecting WOW embedding algorithm.

    AB - Recently, deep learning has been used in steganalysis based on convolutional neural networks (CNN). In this work, we propose a CNN architecture (the so-called maxCNN) to use the selection channel. It is the first time that the knowledge of the selection channel has been incorporated into CNN for steganalysis. The proposed method assigns large weights to features learned from complex texture regions while assigns small weights to features learned from smooth regions. Experimental results on the well-known dataset BOSS-base have demonstrated that the proposed scheme is able to improve detection performance, especially for low embedding payloads. The results have shown that with the ensemble of maxCNN and maxSRMd2+EC, the proposed method can obtain better performance compared with the reported state-of-the-art on detecting WOW embedding algorithm.

    KW - Adaptive steganography

    KW - Convolutional neural networks (CNN)

    KW - Selection-channel

    KW - Steganalysis

    UR - http://www.scopus.com/inward/record.url?scp=85028448682&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=85028448682&partnerID=8YFLogxK

    U2 - 10.1007/978-3-319-64185-0_20

    DO - 10.1007/978-3-319-64185-0_20

    M3 - Conference contribution

    SN - 9783319641843

    VL - 10431 LNCS

    T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

    SP - 263

    EP - 272

    BT - Digital Forensics and Watermarking - 16th International Workshop, IWDW 2017, Proceedings

    PB - Springer Verlag

    ER -