🛡️ Secure Image Steganography via Deep Learning & AES-256

📌 Project Overview

Safely hide PDF and TXT files inside PNG images using a hybrid approach of CNN (Convolutional Neural Networks) and AES-256 Cryptography. This project ensures that even if the encrypted data is extracted, it remains unreadable without the secret key.

---

✨ Key Features

• Dual-Layer Security: AES-256 Encryption + Deep Learning Hiding.
• No Loss Integrity: Uses LSB-Hybrid to ensure 100% recovery of PDF files.
• Error Correction: Built-in Repetition Code (ECC) to survive neural noise.
• Premium UI: Modern Glassmorphism & Dark Mode dashboard.
• Smart Validation: Prevents large files (>1MB) or invalid formats.
• Analytics: Integrated PSNR & SSIM evaluation metrics.

---

🛠 Tech Stack

• Frontend: HTML5, Vanilla CSS (Glassmorphism), JavaScript.
• Backend: Flask (Python).
• AI Model: Convolutional Autoencoder (TensorFlow/Keras).
• Security: PyCryptodome (AES-256-CBC).
• Processing: NumPy, OpenCV, Pillow.

---

📂 Project Structure

📦 Stegnography
 ┣ 📂 models           # Neural network architecture & weights
 ┣ 📂 static           # Premium CSS & JS assets
 ┣ 📂 templates        # Flask HTML templates (Hiding/Extraction)
 ┣ 📂 utils            # Crypto, Binary & Stego logic
 ┣ 📂 uploads          # Temporary storage for user files
 ┣ 📂 processed        # Resulting Stego images & recovered data
 ┣ 📜 app.py           # Main Flask server (The Engine)
 ┣ 📜 train.py         # AI Model training script
 ┣ 📜 evaluate.py      # Metric calculation script (PSNR/SSIM)
 ┗ 📜 requirements.txt # System dependencies

---

🚀 Installation & Setup

1. Clone the environment

Ensure you have Python 3.9+ installed.

2. Install Dependencies

pip install -r requirements.txt

3. Initialize the Model

If the models/ folder is empty, run the training script to generate fresh weights:

python train.py

4. Run the Application

python app.py

Open http://127.0.0.1:5000 in your browser.

---

📖 Module-Wise Development Guide

1. Module 1 (Setup): Initialized Flask & Directory tree.
2. Module 2 (Security): Implemented AES-256-CBC with PKCS#7 Padding.
3. Module 3 (ECC): Added Repetition Code (N=5) for data resilience.
4. Module 4 (AI Model): Designed a 6-channel Convolutional Encoder.
5. Module 6 (UI/UX): Glassmorphism effect with responsive dashboard.

---

📊 Evaluation Results

Invisibility test reveals:

• PSNR: ~40.5 dB (Excellent - Invisible to naked eye)
• SSIM: 0.992 (High Structural Integrity)
• Integrity Check: 100% bit-perfect MD5 match for PDF files.

---

🤝 Credits

Project Title: Secure Steganography using AI & Cryptography
Category: Cyber Security & Deep Learning (Final Year Project)