Mobile DeepFake detection using EfficientNet and facial landmarks

• Main Author: Toh, Dion Siyong
• Other Authors: Deepu Rajan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Neural networks; Transfer learning; Deepfake detection; Mobile application
• Online Access: https://hdl.handle.net/10356/175096
• Institution: Nanyang Technological University

DeepFakes are a significant concern in today’s digital age. The advancement of DeepFake generation techniques has led to incredible growth in the quality of the manipulated content, raising concerns regarding misinformation and other forms of fraud. Current DeepFake detection models are designed for high accuracy and often tend to be complex and large. Coupled with the absence of a user-interface, these renders the detection models are inaccessible for general utilization. Therefore there is a need to develop a detection technique is accessible and simple for general usage — A mobile application that is able to run an accurate DeepFake detection model locally. Experimentation was done on the impacts of using facial landmarks to augment the training data available in the FaceForesics++ dataset. Although the usage of facial landmarks did not yield better results, the models were still able to obtain 96% validation accuracy. This level of accuracy is comparable to other larger detection models. A mobile application, FakeGuard, was also designed and developed using Flutter to offer general users a simple user interface to access the DeepFake detection models.