DeepScan AI — Deepfake Detection System

An intelligent multimodal deepfake detection system that analyzes images, videos, and audio for signs of AI generation or manipulation.

Built for Vulnuris Security Solutions Pvt. Ltd. Hackathon Challenge.

---

Features

🔍 Multi-Modal Detection

• Image Analysis — Error Level Analysis (ELA), FFT Frequency Domain Analysis, Face Region Analysis, Noise Pattern Analysis, Metadata Inspection
• Video Analysis — Per-frame forensic analysis, Temporal Consistency checks, Face Stability tracking, Blink Pattern analysis
• Audio Analysis — Mel-Spectrogram analysis, MFCC Feature extraction, Pitch Consistency (F0), Audio Artifact detection

🎯 Key Capabilities

• Confidence score with detailed breakdown
• Visual forensic heatmaps and spectrograms
• Downloadable analysis reports
• Analysis history tracking
• Drag-and-drop file upload
• Support for 20+ file formats

---

Quick Start

Prerequisites

• Python 3.8+
• pip

Installation

# Navigate to project directory
cd VERSANIX

# Install dependencies
pip install -r requirements.txt

# Run the application
python app.py

Access

Open your browser to http://localhost:5000

---

Supported Formats

Type	Formats
Image	PNG, JPG, JPEG, GIF, BMP, WebP, TIFF
Video	MP4, AVI, MOV, MKV, WMV, FLV, WebM
Audio	MP3, WAV, OGG, FLAC, AAC, M4A, WMA

---

API Endpoints

Method	Endpoint	Description
GET	/	Web interface
GET	/api/health	Health check
POST	/api/analyze	Analyze uploaded media file
GET	/api/history	Get analysis history

Example API Usage

curl -X POST -F "file=@photo.jpg" http://localhost:5000/api/analyze

---

Detection Techniques

Image Detection

1. Error Level Analysis (ELA) — Re-saves image at known quality and compares pixel-level differences
2. Frequency Domain Analysis — Applies FFT to detect GAN spectral fingerprints
3. Face Region Analysis — Checks for blending artifacts and blur inconsistencies around faces
4. Noise Pattern Analysis — Extracts and analyzes noise residuals for synthetic patterns
5. Metadata Analysis — Examines EXIF data for AI tool signatures

Video Detection

1. Per-Frame Analysis — Runs image forensics on sampled keyframes
2. Temporal Consistency — Checks for score variations between frames
3. Face Stability — Tracks face position jitter across frames
4. Blink Analysis — Detects unnatural blink patterns

Audio Detection

1. Spectral Analysis — Analyzes mel-spectrogram for TTS smoothness
2. MFCC Analysis — Checks for statistical anomalies in cepstral coefficients
3. Pitch Consistency — Measures F0 steadiness (TTS tends to be unnaturally steady)
4. Artifact Detection — Finds clicks, phase discontinuities, and splice artifacts

---

Project Structure

VERSANIX/
├── app.py                      # Flask application
├── requirements.txt            # Python dependencies
├── README.md                   # This file
├── detectors/
│   ├── __init__.py
│   ├── image_detector.py       # Image forensic analysis
│   ├── video_detector.py       # Video deepfake detection
│   └── audio_detector.py       # Audio synthesis detection
├── static/
│   ├── css/style.css           # UI stylesheet
│   └── js/app.js               # Frontend logic
├── templates/
│   └── index.html              # Web interface
└── uploads/                    # Temporary upload directory

---

Technology Stack

• Backend: Python, Flask, Flask-CORS
• Image Processing: OpenCV, Pillow, NumPy, SciPy
• Audio Processing: Librosa, SoundFile
• Frontend: HTML5, CSS3, JavaScript (Vanilla)
• Design: Glassmorphism, Dark Theme, Inter Font

---

Disclaimer

This tool uses heuristic-based forensic analysis techniques. Results are indicative and should be used alongside other verification methods for critical decisions. Detection accuracy may vary depending on the sophistication of the deepfake.

---

Team VERSANIX | Vulnuris Hackathon 2026