%0 Journal Article
%T 基于深度学习的密钥控制多图隐写技术研究
Deep Learning-Based Key-Controlled Multi-Image Steganography
%A 朱嘉伟
%A 建一飞
%J Modeling and Simulation
%P 1225-1239
%@ 2324-870X
%D 2025
%I Hans Publishing
%R 10.12677/mos.2025.144368
%X 本研究提出了一种基于可逆神经网络的密钥控制的多图像隐写方案(SDRNN)，以提升安全性和视觉质量。采用私钥加密确保信息安全性，即便算法公开也能保护秘密信息，增强抗攻击能力。针对高容量隐写常见的视觉伪影问题，设计了SCDense模块，通过选择性通道密集连接优化信息嵌入，有效减少轮廓阴影和颜色失真。实验结果表明，相比现有方法，本方案在峰值信噪比(PSNR)和结构相似性(SSIM)等指标上有显著提升，提高了隐写图像的质量和鲁棒性。这显示了该方法不仅理论上有价值，在实际应用中也更可靠、适应性更强。
This study proposes a deep learning-enabled, key-controlled multi-image steganographic framework (SDRNN) to enhance security and visual quality. Private key encryption ensures information security, protecting secret information even with public algorithm disclosure, thereby strengthening attack resistance. Addressing common visual artifacts in high-capacity steganography, we design an SCDense module that optimizes information embedding through selective channel dense connections, effectively reducing contour shadows and color distortion. Experimental results demonstrate significant improvements in peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) compared to existing methods, enhancing both the quality and robustness of stego-images. This indicates the method’s theoretical value and superior practical reliability/adaptability. The research presents a novel effective solution for multi-image steganography, particularly demonstrating notable advantages in security enhancement and visual quality preservation.
%K 图片隐写，
%K 密钥，
%K 可逆神经网络，
%K 多图
Image Steganography
%K Secret Key
%K Reversible Neural Network
%K Multi-Image
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=113546