Unmasking Illusion of Daily-used Deepfake Applications through Landmark Focused Image
Minh-Khoi Nguyen-Nhat, Trung-Truc Huynh-Ngo, Minh–Triet Tran, Trong-Le Do · 2023
The wide spread of deepfake technology has raised significant concerns regarding the authenticity and trustworthiness of visual media. In this paper, we present a novel approach for deepfake detection by incorporating the Facial Landmark Focusing Image (FAFI) technique. To evaluate the effectiveness of our method, we curated a comprehensive dataset specifically designed to capture deepfakes prevalent in daily-used applications. The dataset comprises diverse video clips obtained from different famous platforms, ensuring a realistic representation of deepfake scenarios encountered in everyday life. Through extensive experiments and evaluation, we compare the performance of our proposed feature with a multi-scale retinex feature in baseline models using widely adopted evaluation metrics. The results showcase the superiority of the FAFI technique, highlighting its ability to enhance deepfake detection accuracy. Code and dataset will be made publicly available upon acceptance.