PixelGuard — Forensic Image Analysis Platform

A production-grade forensic image analysis system that combines deep neural networks with classical signal processing to detect image tampering and AI-generated content.

Features

• Error Level Analysis (ELA) — Detects JPEG compression inconsistencies from editing
• AI Generation Detection — GAN spectral analysis, channel correlation, and gradient pattern detection
• CNN Classifier — ConvNeXt-based binary classifier trained on ELA-preprocessed images
• Noise Analysis — Wiener filtering and statistical noise pattern analysis
• EXIF Metadata — Extracts and flags suspicious intrinsic file properties
• Copy-Move Detection — ORB feature matching to find cloned image regions

Classification Categories

Category	Description
Authentic	No significant manipulation or AI generation patterns detected
Tampered	Evidence of post-capture modification (splicing, cloning, retouching)
AI Generated	Image likely produced by generative AI (GANs, diffusion models)

Architecture

PixelGuard2/
├── app.py                    # Flask REST API (auth, analysis, history)
├── analysis/                 # Forensic analysis modules
│   ├── ela.py                # Error Level Analysis
│   ├── deepfake.py           # AI generation detection (spectral/statistical)
│   ├── cnn_classifier.py     # ConvNeXt CNN binary classifier
│   ├── forensics.py          # Frequency, noise, and copy-move analysis
│   └── metadata_analysis.py  # EXIF metadata extraction
├── models/
│   └── database.py           # SQLAlchemy models (User, AnalysisRecord)
├── training/
│   └── train_model.py        # Model training script
├── frontend/                 # React + Vite + Tailwind + Shadcn UI
│   └── src/
│       ├── main.tsx           # Entry point + auth routing
│       ├── layouts/           # DashboardLayout (sidebar)
│       └── pages/             # Dashboard, Analyze, History, Settings
├── model_tampering.keras     # Pre-trained tampering detection model
└── model_deepfake.keras      # Pre-trained AI detection model

Quick Start

Prerequisites

• Python 3.10–3.12 (TensorFlow requirement)
• Node.js 18+

Backend Setup

python -m venv venv
venv\Scripts\activate        # Windows
# source venv/bin/activate   # Linux/Mac

pip install -r requirements.txt
python app.py

Frontend Setup

cd frontend
npm install
npm run dev

Development (Both)

npm run dev    # Runs Flask + Vite concurrently

The app runs on http://localhost:5173 (frontend) proxied to http://localhost:5002 (API).

Model Training

Train your own models using the included training script:

# Prepare dataset structure:
# dataset/
#   authentic/     ← original images
#   manipulated/   ← tampered or AI-generated images

# Train tampering model
python training/train_model.py \
  --dataset path/to/dataset \
  --output model_tampering.keras \
  --type tampering \
  --epochs 20

# Train AI detection model
python training/train_model.py \
  --dataset path/to/dataset \
  --output model_deepfake.keras \
  --type ai_detection \
  --epochs 20 \
  --fine-tune

If models are not present, the system falls back to mathematical heuristics (ELA variance, spectral analysis).

API Endpoints

Method	Endpoint	Description
POST	/api/v1/auth/signup	Create account
POST	/api/v1/auth/login	Login
POST	/api/v1/auth/logout	Logout
GET	/api/v1/auth/me	Current user
GET	/api/v1/dashboard/stats	Dashboard statistics
POST	/api/v1/analyze	Upload & analyze image
GET	/api/v1/history	Paginated analysis history

Tech Stack

• Backend: Flask, SQLAlchemy, TensorFlow/Keras, NumPy, SciPy, Pillow
• Frontend: React 19, TypeScript, Vite 6, Tailwind CSS v4, Shadcn UI
• Database: SQLite
• Models: ConvNeXt-based binary classifiers (transfer learning)

License

MIT