Face Deepfakes -- A Comprehensive Review

Authors: Tharindu Fernando, Darshana Priyasad, Sridha Sridharan, Arun Ross, Clinton Fookes

Published: 2025-02-13 23:08:05+00:00

AI Summary

This paper offers a comprehensive theoretical review of state-of-the-art face deepfake generation and detection methods, providing in-depth algorithmic insights, training paradigms, loss functions, and evaluation metrics. It systematically evaluates the implications of deepfakes on face biometric recognition, outlines key applications, discusses research gaps, and proposes future research directions to advance the field. The study emphasizes the critical need for structured analysis of deepfake technology given its rapid advancements and societal impact.

Abstract

In recent years, remarkable advancements in deep-fake generation technology have led to unprecedented leaps in its realism and capabilities. Despite these advances, we observe a notable lack of structured and deep analysis deepfake technology. The principal aim of this survey is to contribute a thorough theoretical analysis of state-of-the-art face deepfake generation and detection methods. Furthermore, we provide a coherent and systematic evaluation of the implications of deepfakes on face biometric recognition approaches. In addition, we outline key applications of face deepfake technology, elucidating both positive and negative applications of the technology, provide a detailed discussion regarding the gaps in existing research, and propose key research directions for further investigation.

Key findings

Current face deepfake generation methods, particularly for reenactment, often have limited resolution (mostly ≦256 × 256) and struggle with gaze adaptation, extreme poses, and preserving source facial features. There is no universal deepfake detection methodology that can generalize across evolving generation techniques, unseen datasets, or diverse real-world contexts, with interpretability remaining a major challenge. Furthermore, deepfakes, including those from methods like Wav2Lip and SimSwap, are capable of fooling state-of-the-art and lightweight face biometric recognition systems, highlighting significant security vulnerabilities.

Approach

The authors conduct a comprehensive theoretical analysis of state-of-the-art face deepfake generation and detection methods, including detailed algorithmic insights, training paradigms, loss functions, and evaluation metrics. They systematically evaluate deepfake implications for face biometric recognition, categorize applications, identify research gaps, and propose future research directions.

Datasets

FaceForensics++, CelebA, DISFA, VoxCeleb2, CREMA-D, UADFV, DF-TIMIT, Celeb-DF, DF1.0, DFD, VidTIMID, VoxCeleb

Model(s)

UNKNOWN

Author countries

Australia, United States

← Previous