首页|期刊导航|网络与信息安全学报|基于二值特征压缩与超分辨率重建的以图藏图隐写方法

基于二值特征压缩与超分辨率重建的以图藏图隐写方法OA

Hiding image within image steganography based on binary feature compression and super resolution reconstruction

中文摘要英文摘要

以图藏图是一种新型图像隐写技术,旨在将秘密图像隐藏于载体图像中,并在尽可能保持载密图像视觉外观不变的前提下,实现隐蔽传输.现有方法通常采用端到端神经网络完成秘密图像的嵌入与提取,主要关注网络结构设计,而对传统隐写领域中成熟的安全嵌入机制利用不足,导致载密图像在隐写安全性方面仍存在提升空间.为此,提出一种基于金字塔视觉变换器(pyramid vision transformer,PVT)编码、二值特征压缩、传统隐写嵌入与基于Swin变换器的图像恢复(Swin transformer for image restoration,SwinIR)网络重建的解耦式图像隐写框架.该框架首先利用PVT编码器对秘密图像进行特征提取与压缩,并通过二值化模块将其转换为二值码流;随后采用传统彩色图像隐写中的空时编码(space-time code,STC),将二值码流嵌入载体图像,生成载密图像;在提取阶段,通过STC解码恢复二值码流,并利用基于窗口注意力机制的SwinIR网络重建秘密图像.与现有依赖嵌入神经网络的以图藏图方法不同,所提方法使用STC替代嵌入神经网络,使深度特征表示能够与传统彩色图像隐写的代价建模机制相结合,从而降低载密图像的可检测性.实验结果表明,所提方法在保持较高载体/载密图像对视觉质量的同时,能够有效重建秘密图像,并在彩色空间富模型(color spatial rich model,Color-SRM)和CovNet隐写分析器下表现出一定的相对安全性改善.

Hiding image within image is an emerging image steganography technique that aims to hide a secret im-age within a cover image and enable covert transmission while preserving the visual appearance of the stego image as much as possible.Existing methods usually employ end-to-end neural networks to perform secret image embed-ding and extraction,focusing mainly on network architecture design while making insufficient use of mature secure embedding mechanisms from traditional steganography.As a result,there remains room for improvement in the steganographic security of stego images.To address this issue,a decoupled image steganography framework based on pyramid vision transformer(PVT)encoding,binary feature compression,traditional steganographic embedding,and Swin Transformer for image restoration(SwinIR)reconstruction was proposed.Specifically,the framework first used a PVT encoder to extract and compress features from the secret image,and then converted them into a bi-nary bitstream through a binarization module.The binary bitstream was subsequently embedded into the cover im-age using space-time code(STC)from traditional color image steganography to generate the stego image.In the ex-traction stage,the binary bitstream was recovered through STC decoding,and the secret image was reconstructed using SwinIR based on a window-attention mechanism.Unlike existing methods for hiding image within image that rely on neural embedding networks,the proposed method replaced the neural embedding network with STC,enabling deep feature representation to be combined with the distortion modeling mechanism of traditional color image steganography,thereby reducing the detectability of stego images.Experimental results show that the pro-posed method can effectively reconstruct the secret image while maintaining high visual quality between the cover and stego images,and shows a certain relative security improvement against color spatial rich model(Color-SRM)and CovNet steganalyzers.

刘明林;王瑜琪;郭苏娴;孙俊旗;钟智超;罗向阳;骆伟祺

郑州大学,河南 郑州 450001郑州大学,河南 郑州 450001陆军工程大学,江苏 南京 210000郑州大学,河南 郑州 450001郑州大学,河南 郑州 450001河南省网络空间态势感知重点实验室,河南 郑州 450001中山大学,广东 广州 510006

信息技术与安全科学

图像隐写以图藏图二值特征压缩超分辨率重建

image steganographyhiding image within imagebinary feature compressionsuper resolution recon-struction

《网络与信息安全学报》 2026 (4)

1-16,16

国家自然科学基金资助项目(No.62302458,No.62472458,No.U23A20305)广东省自然科学基金资助项目(No.2025A1515012823)河南省创新科技人才团队建设项目(No.254000510007)The National Natural Science Foundation of China(No.62302458,No.62472458,No.U23A20305),The Natural Science Foundation of Guangdong Province(No.2025A1515012823),Innovation Scientists and Technicians Troop Con-struction Projects of Henan Province(No.254000510007)

10.11959/j.issn.2096-109x.AQ26102

评论