Steganalysis based on awareness of selection-channel and deep learning
Document Type
Conference Proceeding
Publication Date
1-1-2017
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.
Identifier
85028448682 (Scopus)
ISBN
[9783319641843]
Publication Title
Lecture Notes in Computer Science Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics
External Full Text Location
https://doi.org/10.1007/978-3-319-64185-0_20
e-ISSN
16113349
ISSN
03029743
First Page
263
Last Page
272
Volume
10431 LNCS
Grant
61379155
Fund Ref
National Natural Science Foundation of China
Recommended Citation
Yang, Jianhua; Liu, Kai; Kang, Xiangui; Wong, Edward; and Shi, Yunqing, "Steganalysis based on awareness of selection-channel and deep learning" (2017). Faculty Publications. 9983.
https://digitalcommons.njit.edu/fac_pubs/9983
