GAB3D-SEVSN： enhanced video steganography model via invertible neural network

doi:10.11772/j.issn.1001-9081.2025050577

Journal of Computer Applications ›› 2026, Vol. 46 ›› Issue (2): 467-474.DOI: 10.11772/j.issn.1001-9081.2025050577

• Cyber security • Previous Articles

GAB3D-SEVSN： enhanced video steganography model via invertible neural network

Qianhui XU¹, Ke NIU¹^,²(), Shunzhe ZHU¹, Lin SHI¹^,², Jun LI¹^,²

^1.College of Cryptography Engineering，Engineering University of PAP，Xi’an Shaanxi 710086，China
^2.Key Laboratory of Network and Information Security，Engineering University of PAP，Xi’an Shaanxi 710086，China

Received:2025-05-28 Revised:2025-08-12 Accepted:2025-08-20 Online:2025-08-22 Published:2026-02-10
Contact: Ke NIU
About author:XU Qianhui， born in 2000， M. S. candidate. Her research interests include information hiding， artificial intelligence.
NIU Ke， born in 1981， Ph. D.， professor. His research interests include multimedia security， information hiding. Email:niuke@163.com
ZHU Shunzhe， born in 2002， M. S. candidate. His research interests include information hiding， artificial intelligence.
SHI Lin， born in 1987， M. S.， lecturer. His research interests include information security， information management systems.
LI Jun， born in 1987， Ph. D.， lecturer. His research interests include multimedia information hiding.
Supported by:
National Natural Science Foundation of China(62272478);Basic Cutting-edge Innovation Project of Engineering University of PAP(WJY202314)

增强型可逆神经网络视频隐写网络GAB3D-SEVSN

徐千惠¹, 钮可¹^,²(), 朱顺哲¹, 石林¹^,², 李军¹^,²

^1.中国人民武装警察部队工程大学密码工程学院，西安 710086
^2.中国人民武装警察部队工程大学武警部队信息安全重点实验室，西安 710086

通讯作者: 钮可
作者简介:徐千惠（2000—），女，山东烟台人，硕士研究生，主要研究方向：信息隐藏、人工智能
钮可（1981—），男，浙江湖州人，教授，博士，主要研究方向：多媒体安全、信息隐藏 Email:niuke@163.com
朱顺哲（2002—），男，湖北武汉人，硕士研究生，主要研究方向：信息隐藏、人工智能
石林（1987—），男，江西九江人，讲师，硕士，主要研究方向：信息安全、信息管理系统
李军（1987—），男，湖南娄底人，讲师，博士，主要研究方向：多媒体信息隐藏。
基金资助:
国家自然科学基金资助项目(62272478);武警工程大学基础前沿创新项目(WJY202314)

Abstract

Abstract:

To address the issues of insufficient long-range motion modeling and over-parameterization caused by channel redundancy in video steganography tasks under small-sample conditions， an enhanced video steganographic network — GAB3D-SEVSN was proposed by integrating 3D Global Attention Block （GAB-3D） and Squeeze-and-Excitation （SE） channel attention. In the model， through the optimized GAB-3D module， key motion trajectories in the 3D spatio-temporal domain were focused on adaptively， thereby enhancing the capability of modeling long-range dependencies. Meanwhile， by embedding the SE module into the reversible architecture， the channel-level adaptive calibration was achieved， which suppressed redundant parameters and alleviated over-parameterization effectively. Experimental results on the UCF101 dataset （13K video samples） demonstrate that， compared to the LF-VSN baseline model， the proposed model achieves improvements of 0.5 dB in Peak Signal-to-Noise Ratio （PSNR） and 2.06% in Structural SIMilarity （SSIM）. Ablation experimental results verify the effectiveness and synergistic effect of various modules. Test results on high-dynamic scene subsets and videos with different attributes show that the model outperforms baseline models in PSNR and SSIM significantly， demonstrating excellent robustness and generalization ability.

Key words: video steganography, Invertible Neural Network (INN), Squeeze-and-Excitation (SE) mechanism, 3D global attention mechanism, channel attention mechanism

摘要：

针对小样本条件下视频隐写任务中存在的长程运动建模不足和通道冗余导致的过参数化问题，提出一种融合三维全局注意力块（GAB-3D）与压缩激励（SE）通道注意力的增强型视频隐写网络GAB3D-SEVSN。该模型通过优化的GAB-3D模块在三维时空域自适应地聚焦关键运动轨迹，从而增强长程依赖的建模能力；同时，通过在可逆架构中嵌入SE模块实现通道级自适应校准，从而有效抑制冗余参数并缓解过参数化现象。在UCF101数据集（13K视频样本）上的实验结果表明，相较于LF-VSN基线模型，所提模型的峰值信噪比（PSNR）和结构相似度（SSIM）分别提升了0.5 dB和2.06%。消融实验结果验证了各模块的有效性和协同效应。而在高动态场景子集和不同属性视频上的测试结果表明，该模型在PSNR和SSIM上均显著优于基线模型，展现出优异的鲁棒性和泛化能力。

关键词: 视频隐写, 可逆神经网络, 压缩激励机制, 三维全局注意力机制, 通道注意力机制

CLC Number:

TP309

Qianhui XU, Ke NIU, Shunzhe ZHU, Lin SHI, Jun LI. GAB3D-SEVSN： enhanced video steganography model via invertible neural network[J]. Journal of Computer Applications, 2026, 46(2): 467-474.

徐千惠, 钮可, 朱顺哲, 石林, 李军. 增强型可逆神经网络视频隐写网络GAB3D-SEVSN[J]. 《计算机应用》唯一官方网站, 2026, 46(2): 467-474.

Figures/Tables 11

Fig. 1 Framework of GAB3D-SEVSN model

Fig. 2 Principle of 3D global attention mechanism

Fig. 3 Architecture of SE channel attention module

Fig. 4 Architecture of spectrally normalized U-Net discriminator

Tab. 1 Weight combination design of loss functions

组合	$λ f o r w$	$λ b a c k$	$λ c e n t e r$	隐写-载体		恢复-秘密
组合	$λ f o r w$	$λ b a c k$	$λ c e n t e r$	PSNR/dB	SSIM	PSNR/dB	SSIM
1	2	1.0	4	42.61	0.964	41.25	0.957
2	1	1.0	5	39.32	0.883	40.84	0.902
3	3	1.0	3	42.94	0.976	38.28	0.849
4	2	0.5	5	40.56	0.931	39.14	0.883

Tab. 1 Weight combination design of loss functions

组合	$λ f o r w$	$λ b a c k$	$λ c e n t e r$	隐写-载体		恢复-秘密
组合	$λ f o r w$	$λ b a c k$	$λ c e n t e r$	PSNR/dB	SSIM	PSNR/dB	SSIM
1	2	1.0	4	42.61	0.964	41.25	0.957
2	1	1.0	5	39.32	0.883	40.84	0.902
3	3	1.0	3	42.94	0.976	38.28	0.849
4	2	0.5	5	40.56	0.931	39.14	0.883

Tab. 2 Comparative analysis of embedding capacity among different video steganography models

模型	嵌入比	相对嵌入容量/BPPC	绝对嵌入容量/B
VStegNET	1∶1	1	435 456.00
En-VStegNET	1∶1	1	435 456.00
LF-VSN	1∶7	7	357 565.44^*
GAB3D-SEVSN	1∶7	7	3 078 336.00

Tab. 3 Comparison results of embedding and extraction performance metrics among different models

模型	载体视频与载密视频对			秘密视频与重建秘密对
模型	PSNR/dB	SSIM	VIF	PSNR/dB	SSIM	VIF
VStegNET	35.57	0.942	0.712	31.88	0.923	0.602
En-VStegNET	36.62	0.950	0.746	33.24	0.946	0.657
Weng等^［10］模型	40.62	0.846	0.828	40.76	0.854	0.836
LF-VSN	42.76	0.973	0.874	44.01	0.988	0.894
GAB3D-SEVSN	43.26	0.993	0.905	45.24	0.995	0.910

Tab. 4 Performance metrics of multi-scenario generalization experiments

测试场景	模型	隐写质量指标		重建质量指标
测试场景	模型	PSNR/dB	SSIM	PSNR/dB	SSIM
高分辨率场景	LF-VSN	42.05	0.981	43.87	0.982
	Weng等^［10］模型	40.04	0.831	40.23	0.840
	GAB3D-SEVSN	42.98	0.991	44.86	0.992
低帧率场景	LF-VSN	38.97	0.831	40.15	0.842
	Weng等^［10］模型	37.42	0.797	38.91	0.828
	GAB3D-SEVSN	40.37	0.924	44.35	0.986
快速运动	LF-VSN	39.86	0.895	41.15	0.962
	Weng等^［10］模型	39.14	0.878	40.22	0.927
	GAB3D-SEVSN	41.98	0.974	43.32	0.981
中速运动	LF-VSN	41.22	0.942	43.28	0.978
	Weng等^［10］模型	39.93	0.915	41.75	0.974
	GAB3D-SEVSN	43.01	0.990	44.91	0.989
慢速运动	LF-VSN	42.74	0.982	44.85	0.991
	Weng等^［10］模型	42.08	0.977	43.16	0.973
	GAB3D-SEVSN	43.60	0.992	45.63	0.994

Fig. 5 Visual comparison of multi-scenario generalization performance

Fig. 6 ROC curves generated by StegExpose

Tab. 5 Ablation study results of different modules

模型	载体视频与载密视频对		秘密视频与重建秘密对
模型	PSNR/dB	SSIM	PSNR/dB	SSIM
LF-VSN（Base）	40.76	0.982 0	41.01	0.987 6
LF-VSN+SE（SEVSN）	40.98	0.992 7	41.52	0.994 0
LF-VSN+GAB3D（GAB3D-VSN）	41.51	0.991 8	43.16	0.995 0
GAB3D-SEVSN	42.06	0.993 3	44.77	0.998 7

References 21

[1]	DINH L， KRUEGER D， BENGIO Y. NICE： non-linear independent components estimation［EB/OL］. ［2025-03-15］..
[2]	DINH L， SOHL-DICKSTEIN J， BENGIO S. Density estimation using real NVP［EB/OL］. ［2025-03-15］..
[3]	MOU C， XU Y， SONG J， et al. Large-capacity and flexible video steganography via invertible neural network［C］// Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2023： 22606-22615.
[4]	KARAMPIDIS K， KAVALLIERATOU E， PAPADOURAKIS G. A review of image steganalysis techniques for digital forensics［J］. Journal of Information Security and Applications， 2018， 40： 217-235.
[5]	RENUKA B， NAIK N M. Secure video steganography technique using DWT and H.264［C］// Proceedings of the 1st International Conference on Advances in Information Technology. Piscataway： IEEE， 2019： 19-23.
[6]	ALY H A. Data hiding in motion vectors of compressed video based on their associated prediction error［J］. IEEE Transactions on Information Forensics and Security， 2011， 6（1）： 14-18.
[7]	ZHAO H B， ZHAO L Y， ZHONG W D. A novel steganography algorithm based on motion vector and matrix encoding［C］// Proceedings of the IEEE 3rd International Conference on Communication Software and Networks. Piscataway： IEEE， 2011： 406-409.
[8]	张弘，尤玮珂，赵险峰. 视频隐写分析技术研究综述［J］. 信息安全学报， 2018， 3（6）： 13-27.
	ZHANG H， YOU W K， ZHAO X F. A survey of video steganalysis［J］. Journal of Cyber Security， 2018， 3（6）： 13-27.
[9]	KHARE R， MISHRA R， ARYA I. Video steganography using LSB technique by neural network［C］// Proceedings of the 2014 International Conference on Computational Intelligence and Communication Networks. Piscataway： IEEE， 2014： 898-902.
[10]	WENG X， LI Y， CHI L， et al. High-capacity convolutional video steganography with temporal residual modeling［C］// Proceedings of the 2019 ACM International Conference on Multimedia Retrieval. New York： ACM， 2019： 87-95.
[11]	MISHRA A， KUMAR S， NIGAM A， et al. VStegNET： video steganography network using spatio-temporal features and micro-bottleneck［C］// Proceedings of the British Machine Vision Conference. Durham： BMVA Press， 2019： No.966.
[12]	JAISWAL A， KUMAR S， NIGAM A. En-VStegNET： video steganography using spatio-temporal feature enhancement with 3D-CNN and hourglass［C］// Proceedings of the 2020 International Joint Conference on Neural Networks. Piscataway： IEEE， 2020： 1-8.
[13]	TAN J， LIAO X， LIU J， et al. Channel attention image steganography with generative adversarial networks［J］. IEEE Transactions on Network Science and Engineering， 2022， 9（2）： 888-903.
[14]	CHEN B， HONG Y， NIE Y. Deep video steganography using temporal-attention-based frame selection and spatial sparse adversarial attack［J］. Journal of Visual Communication and Image Representation， 2024， 104： No.104311.
[15]	XU Y Y， WANG Z M， ZHANG X P. Leveraging spatial residual attention and temporal Markov networks for video action understanding ［J］. Neural Networks， 2024， 169： 378-387.
[16]	LI F， SHENG Y， ZHANG X， et al. iSCMIS： Spatial-channel attention based deep invertible network for multi-image steganography［J］. IEEE Transactions on Multimedia， 2024， 26： 3137-3152.
[17]	JING J， DENG X， XU M， et al. HiNet： deep image hiding by invertible network［C］// Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway： IEEE， 2021： 4713-4722.
[18]	SOOMRO K， ZAMIR A R， SHAH M. UCF101： a dataset of 101 human actions classes from videos in the wild： CRCV-TR-12-01［R/OL］. ［2025-03-15］..
[19]	KINGMA D P， BA J L. Adam： a method for stochastic optimization［EB/OL］. ［2025-03-15］..
[20]	WANG Z， BOUIK A C， SHEIKH H R， et al. Image quality assessment： from error visibility to structural similarity［J］. IEEE Transactions on Image Processing， 2004， 13（4）： 600-612.
[21]	BOEHM B. StegExpose： a tool for detecting LSB steganography［EB/OL］. ［2025-03-15］..

GAB3D-SEVSN： enhanced video steganography model via invertible neural network

增强型可逆神经网络视频隐写网络GAB3D-SEVSN

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 11

References 21

Related Articles 8

Recommended Articles

Metrics

[1]	Meijia LIANG, Xinwu LIU, Xiaopeng HU. Small target detection algorithm for train operating environment image based on improved YOLOv3 [J]. Journal of Computer Applications, 2023, 43(8): 2611-2618.
[2]	Kai ZHANG, Zhengchu QIN, Yue LIU, Xinyi QIN. Multi-learning behavior collaborated knowledge tracing model [J]. Journal of Computer Applications, 2023, 43(5): 1422-1429.
[3]	Ruilin JIANG, Renchao QIN. Multi-neural network malicious code detection model based on depthwise separable convolution [J]. Journal of Computer Applications, 2023, 43(5): 1527-1533.
[4]	Xuedong HE, Shibin XUAN, Kuan WANG, Mengnan CHEN. DeepLabV3+ image segmentation algorithm fusing cumulative distribution function and channel attention mechanism [J]. Journal of Computer Applications, 2023, 43(3): 936-942.
[5]	LIN Yangping, LIU Jia, CHEN Pei, ZHANG Mingshu, YANG Xiaoyuan. Semi-generative video steganography scheme based on deep convolutional generative adversarial net [J]. Journal of Computer Applications, 2023, 43(1): 169-175.
[6]	GAO Shiwei, ZHANG Changzhu, WANG Zhuping. Lightweight real-time semantic segmentation algorithm based on separable pyramid [J]. Journal of Computer Applications, 2021, 41(10): 2937-2944.
[7]	YAO Lu, SONG Huihui, ZHANG Kaihua. Mixed-order channel attention network for single image super-resolution reconstruction [J]. Journal of Computer Applications, 2020, 40(10): 3048-3053.
[8]	. Video steganography based on motion vector [J]. Journal of Computer Applications, 2010, 30(11): 3022-3024.