🔬 VeriLens Forensics

Production-Grade Multi-Engine Forensic Framework for Detecting Synthetic Media, Facial Deepfakes, and Digital Tampering.

---

📌 Executive Summary

Standard deepfake and AI-generated image detectors frequently fail when confronted with compression artifacts, modern diffusion architectures (Midjourney, Flux, Stable Diffusion), or localized Photoshop splicing.

VeriLens Forensics addresses this vulnerability by deploying a Multi-Engine Hybrid Forensic Architecture:

1. Frequency-Domain Spectral Profiling (2D-FFT): Identifies anomalous high-frequency grid residue and unnatural intra-band variance introduced by generative upsampling layers.
2. Error Level Analysis (ELA): Localizes spatial tampering, copy-move operations, and compression discrepancies across image planes.
3. Biometric Landmark & Corneal Reflection Verification: Analyzes ocular specular reflections, pupil light consistency, and facial boundary blending gradients.
4. Digital Provenance & Metadata Engine: Validates camera sensor profiles, detects editing software tags (Photoshop, GIMP), and inspects C2PA / generative prompts.

---

🧠 System Architecture & Mathematical Foundations

1. 2D Fast Fourier Transform (FFT) Decomposition

Input imagery $f(x, y)$ of dimensions $M \times N$ is mapped to the spatial frequency domain:

$$F(u, v) = \sum_{x=0}^{M-1} \sum_{y=0}^{N-1} f(x, y) \cdot e^{-j 2\pi \left(\frac{ux}{M} + \frac{vy}{N}\right)}$$

The centered magnitude spectrum in decibels is extracted via:

$$S(u, v) = 20 \cdot \log_{10}(|F_{\text{shift}}(u, v)| + 10^{-8})$$

2. Error Level Analysis (ELA) Quantization

Analyzes the absolute reconstruction difference against a predetermined compression baseline ($Q = 90$):

$$\Delta(x, y) = |I_{\text{original}}(x, y) - I_{\text{recompressed}}(x, y)|$$

Extrema amplification normalizes localized compression anomalies, exposing non-uniform compression grids indicative of spatial tampering.

---

📂 Repository Layout

├── core/
│   ├── frequency.py      # 2D-FFT Spectral Engine & Radial Decomposition
│   ├── forensics.py      # Error Level Analysis (ELA) & Tampering Masking
│   ├── biometrics.py     # Facial Localization & Corneal Reflection Engine
│   ├── metadata.py       # EXIF & Generative Software Fingerprinting
│   └── pipeline.py       # Consolidated Multi-Engine Forensic Orchestrator
├── api/
│   └── index.py          # High-Performance Serverless Inference Endpoint
├── public/
│   └── index.html        # Enterprise Dark-Mode Forensic Dashboard
├── vercel.json           # Serverless Routing & Static Asset Configuration
├── requirements.txt      # Production Serverless Dependency Manifest
└── README.md             # Theoretical Architecture & Documentation

---

🗺️ Project Roadmap & Completed Milestones

• Phase 1: Frequency Forensics Engine
  • Dynamic spatial scaling & 2D-FFT extraction.
  • Azimuthal radial integration & high-frequency ratio benchmarking.
• Phase 2: Spatial & Compression Forensics
  • Error Level Analysis (ELA) with dynamic scale quantization.
  • Forensic heatmap mask generation for spatial tampering localization.
• Phase 3: Metadata & Provenance Engine
  • EXIF tag validation & AI tool signature parsing (C2PA / Generative chunks).
• Phase 4: Biometric & Multi-Engine Pipeline
  • Haar cascade facial bounding box localization & corneal lighting reflection verification.
  • Unified forensic decision pipeline synthesizing composite risk scores.
• Phase 5: Cloud Deployment & User Interface
  • Serverless FastAPI backend microservice with Base64 telemetry streaming.
  • Responsive cybersecurity dashboard deployed for public access.

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.