← Back to projects

AI · DIGITAL FORENSICS · 2026

FortifyAI

An explainable hybrid deep-learning framework for facial image forgery detection.

FortifyAI project image 1 FortifyAI project image 2 FortifyAI project image 3 FortifyAI project image 4 FortifyAI project image 5 FortifyAI project image 6 FortifyAI project image 7 FortifyAI project image 8 FortifyAI project image 9 FortifyAI project image 10 FortifyAI project image 11 FortifyAI project image 12 FortifyAI project image 13 FortifyAI project image 14

Overview

FortifyAI addresses the growing threat of deepfakes by combining AI-based detection with digital forensic evidence rather than relying on a single method. This improves generalisation across unseen manipulation techniques and datasets.

Key features

  • EfficientNet-B3 backbone with perturbation subtraction and FFT frequency analysis.
  • Multi-signal forensic pipeline using ELA, PRNU, landmark geometry, and texture analysis.
  • Grad-CAM heatmaps for transparent and explainable predictions.
  • Django REST API with a full-stack dashboard and OTP-based authentication.

Results

97.6% Mean accuracy
0.9916 AUC-ROC
90.9% Unseen GAN accuracy

Evaluated across FaceForensics++, Celeb-DF, and GAN-generated image datasets. Published at SCAD 2026, the International Conference on Smart Computing, AI and Data Engineering.

Tech stack

  • Deep learning: PyTorch, TensorFlow, EfficientNet-B3, Grad-CAM
  • Forensics: OpenCV, MTCNN, FFT, PRNU, ELA, GLCM texture analysis
  • Backend: Python and Django REST Framework
  • Frontend: Next.js, HTML, CSS, and JavaScript
  • Database: MongoDB
  • Tools: NumPy, Pandas, Scikit-learn, Timm, and Git

Explore the project

View project on GitHub ↗

Research paper

Read our FortifyAI research paper below.

Our team

FortifyAI project team

A SMALL WIN

With the same team that built our mini project, we took on FortifyAI as our main project and secured an S grade, guided by Dr. Joby P. P., our Head of Department, and Dr. Nabeel Koya, Scientist at C-DAC.