Research Projects

Improving Deepfake Detection in Unseen Scenarios

TUBITAK 1001 Funding Program, Principal Investigator, 2025- (external funding)

This project extends the techniques developed in our earlier TUBITAK-funded work to tackle even more challenging “unseen” face manipulation scenarios, where test data may come from completely novel DeepFake generation models. The aim is to push the limits of detection generalisation by combining one-class classification, adaptive multiple-kernel learning, and ensemble fusion with robust regularisation techniques. The project focuses on scalability, real-time deployment readiness, and adaptability to emerging manipulation methods.

 —  —  —  —  —  —  —  —  —  —  —  —  —  —  —  — 

Deepfake Detection Using One-Class and Open-Set Classification

TUBITAK 1001 Funding Program, Grant No: 121E465, Principal Investigator, 2022-2024 (external funding) Project webpage

This project developed novel one-class and open-set classification methods to detect DeepFakes, especially in the “unseen” manipulation scenario where conventional binary classifiers fail. Key innovations include:

• One-class classifier fusion with sparsity control

• lp-norm multiple kernel learning

• Localised multiple-kernel learning with matrix-norm constraints

• Large-margin novelty detection

These approaches were rigorously evaluated on the FaceForensics dataset, showing superior performance to existing state-of-the-art methods.

 —  —  —  —  —  —  —  —  —  —  —  —  —  —  —  — 

Unconstrained Face Recognition

Iranian National Elites Foundation, Principal Investigator, 2011-2012 (external funding)

This project developed face recognition algorithms capable of operating under unconstrained conditions, including wide variations in pose, illumination, and expression. The research introduced pose-invariant methods using graphical models and multi-descriptor fusion, significantly improving robustness for surveillance and biometric applications.