Smart Stegoguard an AI Powered System for Steganography Detection, Prevention, Secure Image Generation and Transmission

Smart Stegoguard an AI Powered System for Steganography Detection, Prevention, Secure Image Generation and Transmission

Ms.Vedanti Manoj Patekar  –  Exam No: B400400370

Ms.Apurva Bharat Thorat   –  Exam No: B400400405

Ms.Shreya Yogesh Khadilkar   –  Exam No: B400400347

Ms.Diksha Dipak Jadhav   –  Exam No: B400400335

Under the Guidance of – Prof. Radhika Balmuri

Department of Computer Engineering

JSPM’s Jayawantrao Sawant College of Engineering, Pune Hadapsar, Tal-Haveli, Pin- 411028, Tndia

ABSTRACT

Smart StegoGuard is an Al-powered cybersecurity system developed to detect, pre- vent, and secure steganographic content hidden within digital images. As steganog- raphy techniques become increasingly sophisticated, hidden information can be used for unauthorized communication and malicious activities, posing significant security threats. The proposed system provides an intelligent framework to identify such hidden content and ensure secure image transmission.

The system utilizes image processing and machine learning techniques for effec- tive steganography detection. Discrete Wavelet Transform {DWT} is employed to analyze image frequency components and detect subtle modifications caused by data embedding. Histogram Analysis is used to examine pixel intensity distributions and identify abnormalities that may indicate the presence of hidden information. These methods enable accurate and efficient analysis of digital images.

To enhance performance, the Salp Swarm Optimization Algorithm {SSOA} is used for selecting the most relevant features, while the XGBoost classifier accurately distinguishes between clean and steganographic images. This combination improves detection accuracy and reduces false classifications.

Additionally, Smart StegoGuard supports secure image generation and transmis- sion through a user-friendly interface. The system is implemented using Python along with libraries such as NumPy, Pandas, scikit-learn, Matplotlib, and XGBoost. With applications in law enforcement, defense, healthcare, education, and corporate sec- tors, Smart StegoGuard provides a reliable, scalable, and efficient solution for secure digital communication and steganography prevention.

INDEX

ACKNOWLEDGEMENT Page No

ABSTRACT Page No

CHAPTER 1: INTRODUCTION 1

1. lntroduction 1

CHAPTER 2: LITERATURE SURVEY 2

CHAPTER 3: PROBLEM DEFINITION AND SCOPE 3

1. Problem Statement 3

2. Project Scope 3

CHAPTER 4: SOFTWARE REQUIREMENT AND SPECIFICATION

1. lntroduction 5

2. Algorithm Selection 5

3. Design Constraints 9

4. System Features 9

5. Hardware Requirements 10

6. Software Requirements 10

1. Software Quality Attributes 11

2. Analysis Model 12

CHAPTER 5: PROJECT PLAN 13

1. Project Plan 13

2. SDLC 15

CHAPTER 6: BLOCK DIAGRAM 16

CHAPTER 7: SYSTEM DESIGN 18

1. DFD 18

2. Class Diagram 20

3. Use Case Diagram 21

4. Activity Diagram 22

CHAPTER 8: RESULT 23

CHAPTER 9: TECHNICAL SPECIFICATION 25

1. Advantages 25

2. Disadvantages 26

3. Applications 27

CHAPTER 10: CONCLUSION 28

CHAPTER 11 :BIBLIOGRAPHY 29

Chapter 1

Introduction

1. Introduction

Steganography is the practice of concealing information within digital media such as images, audio, or video files in a manner that hides the existence of the secret data. Although it is widely used for privacy protection and secure communication, it can also be misused for unauthorized data exchange, cybercrime, and malicious activities. With the rapid growth of digital communication and the availability of advanced steganography tools, detecting hidden information has become a major challenge in the field of cybersecurity.

To address this issue, Smart StegoGuard is proposed as an Al-powered system designed to detect, prevent, and secure against steganographic content in digital images. The system employs image processing techniques such as Discrete Wavelet Transform

{DWT} and Histogram Analysis, along with machine learning models like XGBoost and Salp Swarm Optimization Algorithm {SSOA}, to accurately identify hidden data. lt also provides secure image generation and transmission, ensuring safe, reliable, and trustworthy digital communication across various domains.

Chapter 2

Literature Survey

Ta b I e 2.1: Literature Survey

Chapter 3

ProbIem Defination and Scope

1. ProbIem Statement

   To design and develop an Al-based system that can detect and prevent steganog- raphy in multimedia files using advanced machine learning techniques. The system should analyze and identify hidden data patterns, prevent unauthorized data trans- mission, and ensure secure communication channels through intelligent monitoring and adaptive learning. i want to this code in the uploaded image

2. Project Scope

   An AI-powered steganography detection, prevention, and secure trans- mission system. The project aims to identify hidden data within multimedia files and prevent unauthorized information exchange using intelligent machine learning techniques.

   In-Scope:

   Steganography Detection:

   Development of an Al-based detection module capable of analyzing image files to identify hidden or suspicious data patterns using machine learning and optimization techniques.

   Prevention Mechanism:

   lntegration of a real-time prevention system that blocks or isolates files containing detected steganographic content before transmission or storage.

   Secure Transmission:

   lmplementation of secure data transmission protocols ensuring that only verified and clean files are allowed for communication, minimizing data leakage risks.

   ImpIementation PIatform:

   • Programming Languages: Python for Al model development and system integration.

   • Frameworks/Libraries: NumPy, Pandas, Scikit-learn, Matplotlib, XGBoost, and Flask for data analysis, machine learning, and web application development.

   • Front-end/Interface: Web-based dashboard for uploading images, viewing analysis results, and secure file transmission.

   • Testing and VaIidation: Verification of system accuracy, detection rate, false- positive rate, and efficiency using benchmark steganographic datasets.

   Out of Scope:

   1. Development of new multimedia encryption algorithms {existing encryption standards will be used}.

   2. Detection of steganography in live network traffic or cloud-based communication systems.

   3. Physical security measures or device-level protection mechanisms.

   4. Analysis of audio and video steganography {current implementation focuses primarily on image files}.

      Chapter 4

      Software Requirement and Specification

      1. Introduction

         The Software Requirement and Specification {SRS} defines the overall functionality, performance, and constraints of the proposed system. lt acts as a blueprint for sys- tem development, validation, and deployment, ensuring that every module operates according to the defined objectives. The primary goal of this project is to design and implement a secure Al-based steganography system that ensures data confidentiality and controlled access through the integration of cryptographic authentication and intelligent image processing. This chapter outlines the algorithms, hardware-software requirements, and quality parameters that govern the operation and performance of the system. The Smart StegoGuard system is developed using Python with Flask for Al-based image embedding and extraction, integrated with cryptographic modules for authentication and security validation. The implementation is carried out on a Windows/Linux platform, providing a secure and scalable environment for testing and analysis.

      2. AIgorithm SeIection

         1. AIgorithm – I : DWT (Discrete WaveIet Transform)

            Purpose:

            Used to extract frequency-domain features from images to detect hidden stegano- graphic data.

            Working PrincipIe:

            • The input image is resized and converted to grayscale for uniform processing.

            • A single-level 2D Discrete Wavelet Transform {DWT} is applied using the Haar wavelet.

            • The image is decomposed into LL, LH, HL, and HH sub-bands.

            • Each sub-band captures unique spatial-frequency information useful for steganog- raphy detection.

              Advantages:

            • High security and imperceptibility.

            • Robust against compression and noise.

            • Higher embedding capacity than LSB methods.

         2. AIgorithm – II : Histogram AnaIysis

            Purpose:

            Used to identify irregularities in pixel intensity distribution caused by hidden data embedding.

            Working PrincipIe:

            • Compute pixel intensity histograms from images.

            • Analyze frequency distribution patterns.

            • Detect anomalies that indicate hidden information.

            • Generate statistical features for classification.

              Advantages:

            • Simple and computationally efficient.

            • Effective for detecting steganographic modifications.

            • Provides useful statistical features.

         3. AIgorithm – III : SSOA (SaIp Swarm Optimization AIgorithm)

            Purpose:

            Used to select the most relevant features for steganography detection.

            Working PrincipIe:

            • lnitialize a population of salps representing feature subsets.

            • Evaluate each subset using a fitness function.

            • Leader salps guide the search toward optimal solutions.

            • Follower salps update their positions accordingly.

            • The process continues until the best feature subset is obtained.

              Advantages:

            • Reduces feature redundancy.

            • lmproves detection accuracy.

            • Reduces computational complexity.

         4. AIgorithm – IV : XGBoost (Extreme Gradient Boosting)

            Purpose:

            Acts as the primary classifier for distinguishing between clean and stego images.

            Working PrincipIe:

            • Start with an initial prediction.

            • Calculate prediction errors {residuals}.

            • Train a decision tree to correct errors.

            • Update predictions iteratively.

            • Repeat until optimal performance is achieved.

              Advantages:

            • High classification accuracy.

            • Handles large datasets efficiently.

            • Reduces overfitting through regularization.

         5. AIgorithm – V : DNA-Based Encryption and Decryption

            Purpose:

            Used to secure secret information before transmission.

            Working PrincipIe:

            1. Convert binary data into DNA sequences using mapping rules.

            2. Apply DNA operations such as complement, substitution, and XOR.

            3. Generate encrypted DNA sequences.

            4. Reverse the operations during decryption.

            5. Recover the original data securely.

               Advantages:

               • Provides strong security.

               • Resistant to cryptanalysis attacks.

               • Suitable for secure data transmission.

      3. Design and ImpIementation Constraints:

         • ComputationaI Overhead: Al-based image generation and authentication code embedding introduce additional computation, which may slightly increase processing time for large multimedia files.

         • Key Storage Security: Storing access keys or authentication metadata re- quires secure storage to prevent unauthorized extraction, tampering, or loss.

         • Integration CompIexity: lntegrating DWT, SSOA, and XGBoost models with feature extraction, feature selection, and access control layers requires an in-depth understanding of machine learning workflows and system design.

         • Performance Trade-off: Real-time verification and extraction introduce a minor delay during data retrieval, although it remains acceptable for secure multimedia applications.

      4. System Features:

         • Secure Data Embedding: Ensures that secret information is securely hidden within Al-generated photorealistic images, preventing unauthorized detection or extraction.

         • Authentication Code Verification: Uses CNN-based embedded authentica- tion codes to verify the integrity and authenticity of stego images before data retrieval.

         • Key-Based Access ControI: Allows only authorized users with valid access keys to extract hidden data, preventing unauthorized access.

         • Lightweight and ScaIabIe Design: Designed to efficiently handle images, audio, or video files without heavy computational overhead, making it scalable for real-world applications.

      5. Hardware Requirements (Minimum Requirement):

         Component	Requirement
         Processor	lntel Core i5 or Higher
         RAM	8 GB Minimum {16 GB Recommended}
         Storage	256 GB SSD or Higher
         Display	1366 × 768 Resolution or Above
         Network	Stable lnternet Connection
         lnput Devices	Keyboard and Mouse

      6. Software Requirements (Minimum Requirement):

         Component	Requirement
         Programming Language	Python 3.10+
         Deep Learning Framework	TensorFlow / PyTorch
         Machine Learning Algo- rithm	XGBoost
         Feature Selection Technique	SSOA {Salp Swarm Optimization Algo- rithm}
         lmage Processing Tech- niques	DWT {Discrete Wavelet Transform}, His- togram Analysis
         Steganography Technique	LSB {Least Significant Bit}
         Libraries	NumPy, Pandas, Matplotlib, Scikit-learn
         Database	PostgreSQL
         Version Control	Git and GitHub

      7. Software QuaIity Attributes:

         • Security: Ensures strong protection against unauthorized data embedding and extraction through DNA-based encryption and Al-driven detection mechanisms.

         • ReIiabiIity: Maintains consistent performance and accuracy in steganographic detection under varying image types and noise conditions.

         • Integrity: Guarantees that the embedded or extracted data remains unaltered during transmission and processing.

         • MaintainabiIity: Modular design allows easy updates of encryption keys, fea- ture extraction modules, or classifiers without affecting the overall system.

         • Performance Efficiency: Utilizes optimized algorithms like DWT and XG- Boost to ensure fast execution with minimal computational overhead.

         • ScaIabiIity: Supports extension to large-scale datasets and integration with cloud-based image security systems.

      8. AnaIysis ModeI

      Figure 4.1: Analysis Model

      This system analysis diagram, titled Smart StegoGuard, illustrates a data security and steganography process. lt begins with an lnput lmage that undergoes Prepro- cessing and Feature Extraction. Separately, Data Security is applied through DNA Encryption. The extracted features and a Secret Data are combined in Feature Opti- mization before being processed by ClassiScoop. An additional Feature Optimization step then feeds into a decision point, which checks if the data is Clean. lf clean, it proceeds to Authentication and produces the Output. lf not clean, it is sent for Logging Monitoring. The second feature optimization also directly feeds into an XG- Boost model for classification. The process integrates image feature analysis, secret data embedding, DNA encryption for security, and a classification/authentication mechanism to produce a secured output

      Chapter 5 Project PIan

      1. Project PIan for Project

      2. Project PIan

         A well-structured project plan ensures systematic development, testing, and successful deployment of the Smart StegoGuard System. The project is executed through a series of well-defined stages, ensuring reliability, scalability, and robust security against steganographic threats. lt follows standard software engineering methodolo- gies to achieve accurate detection, prevention, and secure data transmission.

         Project Overview

         The project aims to develop and integrate an AI-powered security framework ca- pable of detecting, preventing, and securing data against steganography-based at- tacks. The Smart StegoGuard system combines machine Iearning aIgorithms, optimization techniques, and cryptographic moduIes to analyze digital images, identify hidden data patterns, and ensure secure communication across networks.

         The development is divided into six major phases:

         1. Requirement AnaIysis: Gathering project requirements, studying steganog- raphy techniques, and identifying system objectives.

         2. System Design: Designing the architecture, workflow, database structure, and user interface of the proposed system.

         3. ImpIementation: Developing modules for DWT feature extraction, Histogram

            Analysis, SSOA optimization, XGBoost classification, and DNA-based encryp- tion.

         4. Integration: Combining all modules into a unified system and ensuring smooth communication between components.

         5. Testing & VaIidation: Evaluating system performance, detection accuracy, false-positive rate, and overall reliability using benchmark datasets.

         6. Documentation and Maintenance: Preparing project documentation, user manuals, and ensuring future scalability and maintenance of the system.

      3. SDLC

The AgiIe ModeI

Figure 5.1: SDLC Diagram

The diagram illustrates the complete SDLC model and workflow of the Smart StegoGuard system. The development process follows six phases: Requirement Anal- ysis, System Design, Development, Testing, Deployment, and Maintenance to ensure a reliable and secure steganography detection solution. The workflow begins with im- age upload and preprocessing, followed by feature extraction and classification using the XGBoost machine learning model. Based on the analysis, the system determines whether the image is normal or contains hidden steganographic content and displays the result to the user. The model can also be retrained with new data to improve detection accuracy and adapt to emerging steganography techniques.

Chapter 6 BIock Diagram

1. BIock Diagram for the Project

Figure 6.1: Block Diagram

Description

The proposed Smart StegoGuard system consists of three major phases: Sender, Transmission, and Receiver. The system is designed to detect, prevent, and secure steganographic content during image transmission.

At the Sender side, the user first selects an image for transmission. Data hiding is optional and may be performed if secret information needs to be embedded within the image. Before transmission, the image undergoes Prior SteganaIysis, where the system analyzes the image for suspicious or malicious steganographic content. lf ma- licious content is detected, the image is immediately blocked from further processing. lf the image is verified as legal and safe, it proceeds to the transmission phase.

During the Transmission phase, the image is protected using DNA-based En- cryption, which converts the image data into a secure encrypted format. This en- cryption enhances confidentiality and prevents unauthorized access during communi- cation. At the receiving end, DNA-based Decryption is performed to restore the original image data for further analysis.

At the Receiver side, the decrypted image undergoes SteganaIysis to determine whether steganographic content is present. lf no hidden data is detected, the image is considered safe and is directly available for download. lf steganographic content is identified, the image is passed to the DigitaI Signature Verification module to verify its authenticity and legitimacy.

Based on the threat detection results, the system takes appropriate action. lmages with a threat level of 90% or above are automatically bIocked, while images with a threat level between 60% and 90% are moved to a quarantine area (Safe FoIder) for further inspection. Verified and trusted images are allowed for download. This workflow ensures secure image transmission, effective steganography detection, and protection against unauthorized data communication.

Chapter 7 System Design

1. DFD LeveI 2Diagram

Figure 7.1: DFD Level 2 Diagram

The DFD Level-2 of the Smart StegoGuard System illustrates the complete workflow for secure image transmission with steganography detection and threat prevention. The process begins at the sender side, where the selected image is validated by check-

ing its size and dimensions. lf the image exceeds the allowed limits, a warning message is displayed; otherwise, the image proceeds to the encryption and data embedding stage, where LSB steganography and DNA encryption are applied. The image then undergoes Prior Steganalysis using techniques such as DWT Analysis, Histogram Analysis, SSOA Feature Selection, and XGBoost Classification to identify malicious hidden content. lf a threat is detected, the image is blocked; otherwise, it is stored in the Stego lmage Store {D1} and transmitted through a secure DNA-encrypted com- munication channel. At the receiver side, the image is analyzed for hidden data, and if steganographic content is found, the system decrypts and extracts the embedded information. A digital signature verification process ensures data integrity and au- thenticity before allowing the verified image to be downloaded. Finally, the system performs threat detection and risk assessment, where high-risk files are blocked and medium-risk files are quarantined in a safe folder, with all incidents recorded in the Threat Log {D2} for monitoring and security reporting.

1. CIass Diagram for Project

   Figure 7.2: Class Diagram

   The class diagram of this system illustrates the structural design and interactions among various components involved in threat prevention and steganography. The lmageHandler class is responsible for receiving, loading, and preprocessing images using techniques such as DWT and Histogram analysis. The StegoProcessor class performs steganographic operations, including LSB overwriting, JPEG compression, and extraction of hidden data from images. The ThreatDetector class analyzes image content using deep learning frameworks like TensorFlow or PyTorch to identify mali- cious patterns and potential threats. The LogManager class manages the generation and maintenance of system logs, reports, and threat visualizations using tools such as SQLite and Chart.js. The TransmissionManager ensures the secure transmission of analyzed or sanitized data through HTTPS and other communication protocols. Finally, the OutputGenerator class produces the processed results in multiple for- mats, including PDF, DOCX, and TXT, providing flexible and user-friendly output options.

2. Use Case Diagram

   Figure 7.3: Use Case Diagram

   The use case diagram illustrates the interaction between external actors and the steganography system, focusing on how data is hidden and retrieved securely. There are two main actors in this system: the Sender and the Receiver, along with the Sys- tem which manages all operations. The Sender initiates the process by selecting an image, selecting the file to be hidden, and entering a security key. The system calcu- lates the file size, image size, and image dimensions to verify if embedding is possible. Once validated, the file is encrypted and embedded into the image, converting it into BMP format for better stability. lf the file size is larger than the image capacity, the system shows a message alerting the sender. After successful embedding, the sender can save the stego-image and send it to the receiver

3. Activity Diagram for Project

Figure 7.4: Activity Diagram

The activity diagram represents the step-by-step workflow of the steganography threat prevention and data sanitization system. The process starts with the user uploading an image, which is then received by the system for further processing. The uploaded image undergoes LSB {Least Significant Bit} overwrite and JPEG compression, both of which are techniques used to detect or remove any hidden information that might be embedded inside the image. After preprocessing, the system uses deep learning-based threat detection methods implemented with DWT, SSOA,XGBoost to identify any malicious or hidden steganographic content.

Chapter 8 ResuIt

Figure 8.1: User 1 Chat Window

Figure 8.2: User 2 Chat Window

The above figures illustrate the chat interface of the proposed DNAMesh secure com- munication system for two different users, User 1 and User 2. The interface enables secure image sharing and message exchange while integrating steganography-based threat detection mechanisms. When an image is sent, the system automatically an- alyzes it and generates a Steganalysis Report displaying details such as detection status, confidence score, threat level, and verification results. The first figure shows User 1 {Vedanti} receiving an image that has been scanned and verified by the sys- tem, while the second figure shows User 2 {Apurva} viewing another image along with its corresponding analysis report. The chat window also includes options for secure transmission, image encryption status, and threat notifications. These screenshots demonstrate the system’s ability to provide real-time threat detection, secure image communication, and detailed analysis reports to ensure safe and trustworthy data exchange between users.

Chapter 9 TechnicaI Specification

1. Advantages:

   • High Detection Accuracy: Al-driven classification using XGBoost ensures precise identification of steganographic images.

   • Enhanced Security: DNA-based encryption safeguards image data from unau- thorized access or tampering.

   • Optimized Performance: SSOA {Seagull Search Optimization Algorithm} improves feature selection, reducing computation time.

   • Robust Data Protection: Ensures secure image transmission with end-to- end encryption.

   • ScaIabiIity: Can be extended to handle large datasets and integrated into various digital communication systems.

   • Automation: Al-based detection and alert generation operate without manual intervention.

   • User-FriendIy Interface: lntegrated WhatsApp-like module allows easy com- munication and instant alert notifications.

2. Disadvantages:

   • High ComputationaI Demand: Al training and encryption require powerful hardware for faster processing.

   • Storage Requirement: Large datasets and model files increase storage needs.

   • CompIex ImpIementation: lntegration of multiple modules such as DWT, SSOA, XGBoost, and DNA-based encryption increases design complexity.

   • Limited ReaI-Time CapabiIity: Processing speed may decrease when han- dling high-resolution images or large volumes of image data.

   • Dependency on Dataset QuaIity: Detection accuracy and model perfor- mance depend heavily on the quality, diversity, and size of the training dataset.

3. AppIications:

   • Messaging AppIications: Detect and block hidden malicious data embedded within images shared through messaging platforms.

   • SociaI Media PIatforms: Prevent covert communication, unauthorized in- formation exchange, and illegal data sharing through image files.

   • Media PIatforms: Detect copyright violations, unauthorized hidden water- marks, and manipulated multimedia content.

   • Twitter (X) Memes: Prevent attackers from embedding hidden instructions, malicious payloads, or secret messages within viral meme images.

Chapter 10

1. ConcIusion

Smart StegoGuard is an intelligent security solution developed to address the in- creasing threats of steganography in digital communication. By integrating artificial intelligence with advanced image processing techniques such as DWT, Histogram Analysis, SSOA, XGBoost, and LSB steganography, the system effectively detects hidden information, identifies malicious content, and sanitizes suspicious images. lt also ensures secure image generation and transmission, providing users with a re- liable platform for safe digital communication. The system’s modular architecture allows it to operate efficiently across various domains, including cybersecurity, law enforcement, healthcare, education, and social media platforms.Furthermore, Smart StegoGuard overcomes the limitations of traditional steganalysis methods by offering automated, real-time threat detection and comprehensive security reporting. The generated steganalysis reports help users understand the security status of images before sharing or receiving them, thereby reducing the risk of data leakage and covert communication. With its scalable design, high detection accuracy, and secure trans- mission capabilities, the proposed system contributes significantly to enhancing digital security and lays a strong foundation for future advancements in Al-driven steganog- raphy detection and cyber threat prevention.

Chapter 11

BibIiography

1. References

   1. Khan, N., Haan, R., Boktor, G., McComas, M., Daneshi, R. {2020}. Steganogra- phy GAN: Cracking Steganography with CycleGANs. arXiv:2006.04008. Demon- strates using CycleGANs to detect and break classical LSB steganography tech- niques.

   2. Liu, M., Zhang, M., Liu, J., Yang, X. {2018}. Generative Steganography Based on GANs. ln Cloud Computing and Security {lCCCS 2018}, Lecture Notes in Computer Science. Proposes GAN architectures for embedding secret data securely within synthesized images.

   3. Guan, Y., Tan, S., Li, Q. {2023}. Binary Steganography Based on Generative Adversarial Nets. Multimedia Tools and Applications, 82, 6687-6706. Applies GANs for binary data embedding using syndrome-trellis codes and distortion minimization.

   4. Kaneria, S., Jotwani, V. {2024}. Comparative Performance of U-Net, V-Net, and U-Net++ for Deep Learning-Based Image Steganography. Journal of Ad- vanced Zoology, 45{3}. Evaluates different encoder architectures for embedding and reconstruction quality.

   5. Edge-Guided Dual-Stream U-Net for Secure Image Steganography. {2023}. Ap- plied Sciences, 15{8}, 4413. Uses a dual-stream U-Net architecture to improve imperceptibility and resistance against steganalysis.

   6. Yangjie, Z., Jia, L., Meiqi, L., Ke, Y., Zhang, M. {2024}. Generative Image

      Steganography Based on Point Cloud. arXiv preprint arXiv:2410.11673. lntro- duces a point-cloud-based representation for secure steganographic embedding.

   7. Wang, X., Chen, K., Qi, Y., Liu, R., Zhang, W., Yu, N. {2025}. GIFDL: Gener-

      ated Image Fluctuation Distortion Learning for Enhancing Steganographic Se- curity. arXiv preprint arXiv:2504.15139. Proposes a distortion-learning method using generated image fluctuations to improve resistance to steganalysis.

   8. Zhang, K.A., Cuesta-lnfante, A., Xu, L., Veeramachaneni, K. {2019}. arXiv preprint arXiv:1901.03892. This paper introduces a GAN-based approach achiev- ing high embedding capacity and imperceptibility in stego-images.