The state-of-the-art of visual deepfake detection

Thumbnail Image

Files

URL

Journal Title

Journal ISSN

Volume Title

School of Science | Bachelor's thesis
Electronic archive copy is available locally at the Harald Herlin Learning Centre. The staff of Aalto University has access to the electronic bachelor's theses by logging into Aaltodoc with their personal Aalto user ID. Read more about the availability of the bachelor's theses.
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Date

2025-12-11

Department

Major/Subject

Data Science

Mcode

Degree programme

Aalto Bachelor's Programme in Science and Technology
Aalto Bachelor's Programme in Science and Technology
Aalto Bachelor's Programme in Science and Technology

Language

en

Pages

25

Series

Abstract

Contemporary advances in artificial intelligence (AI) and machine learning (ML) have enabled the creation of highly realistic synthetic deepfake media. While deepfakes could have their applications in educational and creative contexts, they also carry significant risks. Identity manipulation, disinformation, and the erosion of public trust are among the most serious potential consequences of deepfake abuse. This thesis provides a comprehensive review of the current state of deepfake detection research. It analyzes detection methods in terms of their architectural design and their assumptions about the generation techniques. It evaluates the reported benchmarking results in light of the characteristics of the datasets used in the experiments. The analysis reveals that most detection models are trained on large, synthetic, and homogeneous datasets, which greatly limits their generalization to real-world deepfakes. Due to strong assumptions that are embedded in the datasets and in the model architectures, no current method achieves robust and generalizable detection performance across diverse in-the-wild datasets and evolving generation techniques. The findings of this thesis emphasize the need for more representative, up-to-date datasets and highlights the importance of complementary approaches, such as watermarking, blockchain, and explainable AI.

Description

Supervisor

Korpi-Lagg, Maarit

Thesis advisor

Iannucci, Letizia

Keywords

deepfake, deepfake detection, deepfake generation

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202512309701

Collections

[kand] Perustieteiden korkeakoulu / SCI