[Paper Review] Deep Learning for Steganalysis of Diverse Data Types: A review of methods, taxonomy, challenges and future directions

Hamza Kheddar, Mustapha Hemis+3 more|arXiv (Cornell University)|Aug 8, 2023

Advanced Steganography and Watermarking Techniques244 references4 citations

TL;DR

This paper presents a comprehensive review of deep learning-based steganalysis techniques for detecting hidden data in diverse digital media, including images, audio, video, and text. It evaluates deep transfer learning (DTL) and deep reinforcement learning (DRL) for improved detection performance, offering a systematic taxonomy, analysis of datasets and metrics, and insights into future research directions in secure digital communication.

ABSTRACT

Steganography and steganalysis are two interrelated aspects of the field of information security. Steganography seeks to conceal communications, whereas steganalysis is aimed to either find them or even, if possible, recover the data they contain. Steganography and steganalysis have attracted a great deal of interest, particularly from law enforcement. Steganography is often used by cybercriminals and even terrorists to avoid being captured while in possession of incriminating evidence, even encrypted, since cryptography is prohibited or restricted in many countries. Therefore, knowledge of cutting-edge techniques to uncover concealed information is crucial in exposing illegal acts. Over the last few years, a number of strong and reliable steganography and steganalysis techniques have been introduced in the literature. This review paper provides a comprehensive overview of deep learning-based steganalysis techniques used to detect hidden information within digital media. The paper covers all types of cover in steganalysis, including image, audio, and video, and discusses the most commonly used deep learning techniques. In addition, the paper explores the use of more advanced deep learning techniques, such as deep transfer learning (DTL) and deep reinforcement learning (DRL), to enhance the performance of steganalysis systems. The paper provides a systematic review of recent research in the field, including data sets and evaluation metrics used in recent studies. It also presents a detailed analysis of DTL-based steganalysis approaches and their performance on different data sets. The review concludes with a discussion on the current state of deep learning-based steganalysis, challenges, and future research directions.

Motivation & Objective

• To provide a systematic review of deep learning-based steganalysis methods across multiple data types, including image, audio, video, and text.
• To analyze the role of advanced deep learning techniques such as deep transfer learning (DTL) and deep reinforcement learning (DRL) in enhancing steganalysis performance.
• To establish a structured taxonomy of steganalysis methods based on carrier type and technique type for improved research navigation.
• To evaluate existing datasets, benchmark metrics, and performance evaluation protocols used in recent steganalysis research.
• To identify key challenges and propose future research directions for advancing robust, generalizable, and interpretable steganalysis systems.

Proposed method

• The study employs a systematic literature review methodology with defined criteria for study selection, quality assessment, and bibliometric analysis.
• It classifies steganalysis approaches based on carrier type (image, audio, video, text) and technique type (DTL, DRL, hybrid, other deep learning models).
• The paper evaluates deep neural networks (DNNs), convolutional neural networks (CNNs), residual networks (ResNets), and attention mechanisms such as CBAM and SPP for feature extraction and detection.
• It examines deep transfer learning (DTL) for knowledge transfer across domains and steganographic techniques to improve detection with limited labeled data.
• Deep reinforcement learning (DRL) is analyzed for adaptive detection strategies that learn optimal decision policies through interaction with steganographic environments.
• The methodology includes a critical review of datasets (e.g., BOSSBase, AMR, TID2013), evaluation metrics (AUC, DR, FAR, FPR, PSNR, SSIM), and benchmarking protocols.

Figure 1 : Steganalysis process.

Experimental results

Research questions

• RQ1 How do deep learning models perform across different carrier types—images, audio, video, and text—in steganalysis tasks?
• RQ2 To what extent can deep transfer learning (DTL) improve steganalysis performance when generalizing across different steganographic techniques and data domains?
• RQ3 How can deep reinforcement learning (DRL) enhance the adaptability and robustness of steganalysis systems in dynamic or unknown steganographic environments?
• RQ4 What are the most effective deep learning architectures and loss functions for detecting subtle steganographic signals in digital media?
• RQ5 What are the key challenges in data availability, model interpretability, adversarial robustness, and privacy that hinder the deployment of real-world steganalysis systems?

Key findings

• Deep learning-based steganalysis models significantly outperform traditional methods in detection accuracy, with state-of-the-art models achieving high AUC scores and low detection error rates.
• Deep transfer learning (DTL) enables effective detection in low-data regimes by transferring knowledge from source domains to target steganographic techniques, improving generalization.
• Deep reinforcement learning (DRL) shows promise in learning adaptive detection policies that respond dynamically to evolving steganographic strategies and cover types.
• Hybrid models combining attention mechanisms (e.g., CBAM) with CNNs or ResNets demonstrate superior feature representation and detection performance on complex steganographic signals.
• The use of large language models like ChatGPT can assist in generating synthetic steganographic content and analyzing linguistic patterns, aiding in dataset creation and algorithm evaluation.
• Despite progress, challenges remain in model interpretability, adversarial robustness, hardware constraints, and privacy-preserving training, which limit real-world deployment.

Figure 2 : Interpretation of steganograms in AWGN channel.