文献类型：期刊文献

英文题名：Provably Secure and Robust Training-Free GIF Steganography

作者：Wang, Huazhong[1]; Sun, Linhao[1]; Zhu, Zhiying[1]; Jiang, Qingchao[1]; Zhang, Xinpeng[2]; Qian, Zhenxing[2]

机构：[1] East China University of Science and Technology, School of Information Science and Engineering, Shanghai, 200237, China; [2] Fudan University, College of Computer Science and Artificial Intelligence, Shanghai, 200433, China

年份：2026

外文期刊名：IEEE Transactions on Multimedia

收录：EI(收录号：20263121214013);Scopus(收录号：2-s2.0-105046090266)

语种：英文

外文关键词：Animation - Cryptography - Extraction - Iterative methods

摘要：GIF steganography utilizes animated GIF images as carriers for transmitting secret messages. Conventional GIF steganography techniques rely on modifications of index values. However, such methods alter statistical features, thereby compromising security, and their extraction accuracy degrades when the stego-GIF undergoes compression. In this paper, we propose a robust, training-free GIF steganography framework with provable security guarantees. The sender duplicates and encrypts the secret message and maps it into latent variables, which are subsequently integrated with controlled text and motion modules for iterative denoising. Finally, it is fed into a decoder to generate a GIF animation. The receiver then splits the GIF into individual frames and extracts the secret message from the reconstructed latent space. Specifically, we use an embedding module that ensures that the secret message is effectively mapped to the latent space and aligns with the standard Gaussian distribution. Meanwhile, we propose an extraction module that retrieves the secret message from each individual frame, while the validation module enhances robustness by calibrating inter-frame comparisons. By preserving the original latent distribution, our method is theoretically undetectable. Extensive experimental results show that our method can achieve an accuracy of near 100% under various attacks, which is significantly better than existing GIF steganography methods. ? 1999-2012 IEEE.