Enhanced Images Deepfake Detection Using YOLOv8 with MTCNN Face Extraction Method for High-Accuracy Classification

Abstract

With the increasing use of fake images and videos on social media platforms, the issue of verifying the authenticity of images has become of great importance for privacy protection. In this study, a deep fake detection system is developed and presented using the YOLOv8 deep learning model. The proposed system mainly relies on extracting the face region from the input images sample using Multi-Task Cascaded Convolutional Neural Networks (MTCNN) method to reduce the processing time of fake and real face detection process. The extracted faces regions are then analysed and classified based on YOLOv8 model to obtain the fake and real information related to each input image in term of binary classification. The main hyperparameters of Yolov8 model were tuned to achieve higher detection accuracy. The system performance was evaluated using multiple fake and real images datasets. The results showed that the proposed system achieved a detection accuracy of 97.5% and an F-score of 97%.