Hybrid Transformer-Based Steganography Using Frequency Embedding and Adversarial Robustness

Abstract

Steganography is the practice of hiding secret information within digital media such as images, audio, or video. It ensures confidential communication by concealing the existence of data itself, unlike encryption. Modern research focuses on improving invisibility, security, and resistance to detection using AI and deep learning. This project aims to design a secure and intelligent image steganography system using a hybrid Transformer model. It focuses on increasing data hiding capacity while maintaining image quality and reducing detectability. The system will be capable of resisting steganalysis attacks, compression, and noise distortions in real-world usage. Recent studies show that deep-learning steganography models still suffer from low robustness and poor scalability. Most approaches fail under compression or noise, and their hidden data can be detected by advanced AI models. High computational cost, limited payload capacity, and dataset dependency further affect their reliability. This project introduces a hybrid Transformer integrated with Discrete Cosine Transform (DCT) for frequency embedding. By combining spatial and frequency domains, it ensures better concealment and robustness. Adversarial training with a steganalysis discriminator will enhance security against modern detection models. The system will achieve higher PSNR and SSIM scores, proving superior imperceptibility and accuracy. Implementation will use Python, PyTorch, and OpenCV for model training and image processing. Datasets like COCO, BOSSBase, and ImageNet will be used for evaluation. Performance metrics such as PSNR, SSIM, MSE, and BER will measure quality and accuracy. Development and testing will be carried out in Jupyter Notebook or Google Colab environments.

Introduction

Steganography is the technique of hiding secret information within digital media such as text, images, audio, or video so that the existence of the message remains concealed. Unlike cryptography, which encrypts data into an unreadable format, steganography disguises the presence of the information itself. Its effectiveness is measured by three key factors: imperceptibility (minimal distortion), payload capacity (amount of hidden data), and security (resistance to detection and extraction).

Types of Steganography

The study discusses four major forms of steganography:

• Text Steganography: Hides data using spaces, punctuation, font styles, or word substitutions. It is simple but has limited storage capacity and robustness.
• Image Steganography: The most widely used method, employing techniques such as Least Significant Bit (LSB) embedding and transform-domain methods like Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT). It offers high payload capacity while maintaining image quality.
• Audio Steganography: Embeds data within digital audio using techniques such as LSB modification, phase coding, spread spectrum, and echo hiding. It provides good audio quality but may be affected by compression.
• Video Steganography: Combines spatial and temporal redundancy across video frames to hide large amounts of data. It offers higher capacity and security but faces challenges from video compression and steganalysis attacks.

Recent advances in deep learning, Transformers, and adversarial learning have further improved the robustness and intelligence of modern steganographic systems.

Literature Review

The reviewed studies highlight several improvements in steganography:

• Enhanced LSB techniques that improve image quality by minimizing pixel modifications.
• Image segmentation methods that embed data in texture-rich regions, increasing resistance to steganalysis.
• GUI-based image steganography systems that combine LSB embedding with encrypted messages for secure communication.
• Improved phase coding algorithms for audio steganography that enhance robustness while preserving audio quality.

Although these approaches improve performance, many focus on a single media type rather than providing a unified solution.

Proposed System

The study proposes a secure web-based multi-modal steganography platform capable of hiding and retrieving secret information using images, audio, video, and text files. The system combines AES encryption with steganography to provide two layers of security and includes user authentication, activity tracking, and a service-oriented architecture.

Steganography Modules

Image Steganography

• Uses adaptive LSB embedding to hide more data in textured image regions.
• Employs random pixel selection based on a user key.
• Supports DCT-based embedding for JPEG images.
• Evaluates quality using PSNR, SSIM, MSE, and embedding capacity.

Audio Steganography

• Supports WAV audio files.
• Uses LSB encoding, phase coding, and echo hiding.
• Includes synchronization information for accurate extraction.
• Measures quality using Signal-to-Noise Ratio (SNR).

Video Steganography

• Utilizes both video frames and audio tracks to maximize hiding capacity.
• Applies frame-selective LSB embedding and DCT-based techniques.
• Improves robustness through motion vector modification.
• Evaluates quality using PSNR, SSIM, VMAF, and bitrate analysis.

Text Steganography

• Hides information using invisible Unicode characters, whitespace manipulation, synonym substitution, capitalization, and punctuation patterns.
• Preserves the original meaning of the text while embedding hidden information.

System Architecture

The proposed platform follows a five-layer service-oriented architecture:

1. Client Layer – Provides browser-based access through secure HTTPS connections.
2. Presentation Layer – Handles user interaction, input validation, file preview, and downloads.
3. Application Layer – Manages authentication, REST APIs, user dashboards, and steganography requests.
4. Service Layer – Performs embedding, extraction, AES encryption/decryption, key generation, and activity logging.
5. Data Layer – Stores user accounts, authentication details, activity logs, cover files, and generated stego files.

Working Process

The system operates in two stages:

Encoding

1. User authentication.
2. Secret message encrypted using AES.
3. Encrypted data embedded into the selected cover file using LSB techniques.
4. Generated stego file stored and returned to the user.

Decoding

1. User uploads the stego file and enters the password.
2. Hidden encrypted data extracted using the LSB extraction process.
3. AES decryption recovers the original message.
4. The recovered message is displayed.

This combination of encryption and steganography ensures that even if hidden data is detected, it remains unreadable without the correct decryption key.

Performance

The proposed system demonstrates efficient performance with very low encoding and decoding times, making it suitable for practical secure communication across multiple media formats.

Conclusion

In this work, a secure multi-modal steganography platform was designed and implemented for hiding confidential information within text, image, audio, and video files. The system combines steganography techniques with AES encryption to provide an additional layer of security, ensuring that hidden data remains protected from unauthorized access. Different embedding methods were employed for each media type to achieve effective data hiding while preserving the quality of the cover media. Experimental results demonstrated that the proposed system successfully performs both encoding and decoding operations with high accuracy and acceptable processing time. Image and audio steganography maintained good perceptual quality, while video steganography provided higher payload capacity for hiding larger amounts of information. The use of authentication, activity logging, and secure storage further enhanced the reliability and usability of the platform. Overall, the proposed system provides a secure, efficient, and flexible solution for covert communication across multiple digital media formats. The combination of steganography and cryptography improves confidentiality and makes the system suitable for applications requiring secure information exchange. Future enhancements may include the integration of deep learning-based steganography techniques, improved robustness against steganalysis attacks, and support for cloud-based and real-time communication environments.

References

[1] A. Rehman, T. Saba, T. Mahmood, Z. Mehmood, M. Shah, and A. Anjum, \"Data hiding technique in steganography for information security using number theory,\" Journal of Information Security and Applications, vol. 78, p. 103618, 2023. J. Breckling, Ed., The Analysis of Directional Time Series: Applications to Wind Speed and Direction, ser. Lecture Notes in Statistics. Berlin, Germany: Springer, 1989, vol. 61. [2] R. Meng, Q. Cui, Z. Zhou, Z. Fu, and X. Sun, \"A steganography algorithm based on CycleGAN for covert communication in the Internet of Things,\" IEEE Access, vol. 11, pp. 12345–12358, 2023. [3] S. Kaur, S. Singh, M. Kaur, and H.-N. Lee, \"A systematic review of computational image steganography approaches,\" Archives of Computational Methods in Engineering, vol. 30, no. 7, pp. 4775–4805, 2023. [4] N. Subramanian, O. Elharrouss, S. Al-Maadeed, and A. Bouridane, \"Image steganography: A review of the recent advances,\" IEEE Access, vol. 11, pp. 23409–23423, 2023. [5] M. M. Hashim, A. A. Mahmood, and M. S. M. Rahim, \"A comparative study among different techniques of audio steganography,\" Indonesian Journal of Electrical Engineering and Computer Science, vol. 33, no. 1, pp. 456–467, 2024. [6] P. Singh and R. Kumar, \"An improved LSB image steganography technique using bit-inversion in spatial domain,\" Multimedia Tools and Applications, vol. 83, pp. 18211–18233, 2024. [7] A. K. Sahu, M. Hassaballah, R. Saraswathi Rao, and G. Suresh, \"Logistic-map based fragile image watermarking scheme for tamper detection and localization,\" Multimedia Tools and Applications, vol. 82, pp. 24069–24100, 2023. FLEXChip Signal Processor (MC68175/D), Motorola, 1996. [8] H. Kheddar, M. Hemis, Y. Himeur, D. Megías, and A. Amira, \"Deep learning for steganography and steganalysis: Recent advances and future perspectives,\" Neurocomputing, vol. 581, p. 127495, 2024. A. Karnik, “Performance of TCP congestion control with rate feedback: TCP/ABR and rate adaptive TCP/IP,” M. Eng. thesis, Indian Institute of Science, Bangalore, India, Jan. 1999. [9] M. A. Majeed, R. Sulaiman, Z. Shukur, and M. K. Hasan, \"A review on text steganography techniques,\" Mathematics, vol. 11, no. 4, p. 920, 2023. [10] D. Megías, W. Mazurczyk, and M. Kuribayashi, \"Data hiding and its applications: Digital watermarking and steganography,\" Applied Sciences, vol. 13, no. 21, p. 11924, 2023.

Copyright

Copyright © 2026 Aakash Bonagiri, N Naveen Kumar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.