University of Twente Student Theses
Deepfake Manipulation Detection
Saha, N.S. (2024) Deepfake Manipulation Detection.
|
PDF
1MB |
| Abstract: | The advent of generative AI have led to the rise of synthetic media, also referred to as deepfakes. Just like any new technology, deepfakes can be a boon and a bane. Hence it is imperative to identify deepfakes that may be used for malicious purposes. This paper presents a lightweight video deepfake manipulation detection method based on the temporal differences of facial mesoscopic properties between frames. Mesoscopic properties refer to the characteristics of images that lie between the microscopic and macroscopic scales, such as textures and edges. Most contemporary deepfake detection methods perform excellently at detecting deepfakes, but they often require high computational power. This makes them unsuitable for real time applications or deployment on resource constraint devices. Taking these observations into account, I propose a lightweight deepfake manipulation detection framework that utilizes the combination of a lightweight CNN network and an LSTM network to take both spatial and temporal dimensions into account. Through experiments on open source datasets, I show that this framework is effective in identifying deepfakes to a certain extent at a low computational cost. |
| Item Type: | Essay (Bachelor) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science BSc (56964) |
| Link to this item: | https://purl.utwente.nl/essays/101248 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page