Development of Intelligent Framework. Image Steganalysis using Deep Level Features

Table of Contents

INTRODUCTION

1.1 Steganography

1.2 Steganalysis

1.3 Deep Learning Overview

1.4 Contributions

1.5 Thesis Roadmap

LITERATURE SURVEY

2.1 Introduction

2.2 Related Literature

2.3 Analysis of image steganalysis frameworks

2.4 Summary

PROBLEM FORMULATION AND RESEARCH METHODOLOGY

3.1 Introduction

3.2 Research Challenges

3.3 Broad problem statement

3.4 Research Objectives

3.5 Research Methodology

3.6 Summary

PROPOSED IMAGE STEGANALYSIS SCHEME

4.1 Introduction

4.2 The CNN based framework for image steganalysis

4.3 Training and testing of proposed image steganalysis framework

4.4 Summary

EXPERIMENTAL ANALYSIS AND RESULTS

5.1 Introduction

5.2 Datasets

5.3 Performance metrics used

5.4 Summary of the model

5.5 Experimental Results

5.6 Comparative Analysis

5.7 Summary

CONCLUSION AND FUTURE WORK

6.1 Introduction

6.2 Conclusion and future direction of the research work

Research Objectives and Themes

The primary goal of this research is to develop an improved deep learning-based framework for image steganalysis that addresses technical challenges such as computational inefficiency, source mismatch, and low embedding ratios. By leveraging Convolutional Neural Networks (CNNs), the thesis provides a comprehensive evaluation of state-of-the-art methods and proposes an optimized architecture to effectively distinguish between cover and stego images.

• Exhaustive analysis of existing CNN-based steganalysis frameworks.
• Development of an improved deep feature-based image steganalysis model.
• Performance evaluation of the proposed framework in binary classification tasks.
• Optimization of training procedures through advanced techniques like batch normalization and ReLU activation.

Auszug aus dem Buch

Summary of Chapters

INTRODUCTION: Provides an overview of steganography and steganalysis, the role of deep learning in modern security, and defines the roadmap of the thesis.

LITERATURE SURVEY: Reviews historical and contemporary research on deep learning-based steganalysis, presenting a tabular summary of various frameworks and their performance.

PROBLEM FORMULATION AND RESEARCH METHODOLOGY: Identifies critical research gaps and challenges, such as execution efficiency and dataset mismatch, while outlining the methodology to solve them.

PROPOSED IMAGE STEGANALYSIS SCHEME: Describes the design and construction of the proposed CNN-based architecture, detailing sub-networks and the utilization of batch normalization and max pooling.

EXPERIMENTAL ANALYSIS AND RESULTS: Presents the setup of the experiment using the BOSSBASE v1.01 dataset, evaluates performance metrics like accuracy and loss, and discusses the findings.

CONCLUSION AND FUTURE WORK: Summarizes the thesis findings and suggests future directions, including the use of larger datasets and more complex model layers.

Keywords

Steganalysis, Steganography, Deep Learning, Convolutional Neural Networks, CNN, Image Security, Feature Extraction, Batch Normalization, Binary Classification, BOSSBASE, Deep Residual Learning, Frequency Domain, Spatial Domain, Model Training, Performance Metrics.

Frequently Asked Questions

What is the core focus of this research?

The research focuses on the field of image steganalysis, aiming to develop an efficient deep learning-based system to detect hidden information within digital images.

What are the primary themes of the work?

Key themes include the mitigation of computational overhead in deep learning models, addressing source and embedder mismatch problems, and improving the accuracy of stego-signal detection.

What is the main objective of the thesis?

The main objective is to design an improved image steganalysis framework that enables training efficiency and robust detection of secret content by utilizing deep feature extraction via CNNs.

Which methodology is employed in the study?

The study utilizes a CNN architecture incorporating high-pass filters, batch normalization, max-pooling layers, and a Softmax activation unit, evaluated through a mini-version experimental setup on the BOSSBASE dataset.

What does the main body of the work cover?

The main body covers a systematic review of existing literature, a detailed formulation of research challenges, the proposal of a custom CNN-based framework, and an experimental results section assessing model accuracy.

Which keywords characterize this work?

The research is characterized by terms such as Steganalysis, Deep Learning, Convolutional Neural Networks, and Image Security, reflecting its focus on modern forensic detection methods.

Why are max-pooling layers significant in the proposed framework?

Max-pooling is used to down-sample image dimensions, effectively reducing noise and extracting important high-level features without losing critical data necessary for steganalysis.

How does the proposed framework handle image noise?

The framework utilizes a high-pass filter (HPF) in the sub-network to isolate frequency changes, which helps in sharpening the image and highlighting potential artifacts caused by steganography.