A Steganography Immunoprocessing Framework Against CNN-Based and Handcrafted Steganalysis
Yijing Chen,Hongxia Wang,Wanjie Li,Wenshan Li
DOI: https://doi.org/10.1109/tifs.2024.3409075
IF: 7.231
2024-06-14
IEEE Transactions on Information Forensics and Security
Abstract:Performing post-processing on the stego image has promise for improving the steganography security. Nevertheless, the existing post-processing schemes neglect the characteristics of the stego image, which lack strong theoretical interpretability. Moreover, existing schemes do not fully consider the holistic steganography security against both CNN-based and handcrafted steganalyzers. In this paper, we propose a steganography immunoprocessing (IP) framework based on Artificial Immune System (AIS) that is universal for the stego images from the same steganographic process to further enhance the security. Based on the natural relationship between immune theory and steganography, we regard the immunoprocessing policy as the antibody, and the performance of the anti-steganalysis for stego images protected by antibody as the antibody affinity. By the immune dynamic optimization process, the optimal immunoprocessing policy is dynamically searched and performed on the stego image to achieve further optimization. In addition, we enhance the resistance of the stego against the target CNN-based steganalyzer by limiting the immunoprocessing direction. Performing the optimal immunoprocessing on stego images will enhance the holistic security of steganography. Experimental results demonstrate that the proposed immunoprocessing can significantly improve the holistic security of adaptive steganography against both CNN-based and handcrafted steganalyzers, and achieve better performance than related schemes.
computer science, theory & methods,engineering, electrical & electronic