Towards Sequential Deepfake Detection Using Deep Learning for Privacy Protection

With the development of deep learning, deep generative models have generated hyper-realistic facial images that are virtually indistinguishable from authentic images. Recently, the misuse of deepfake technology in electronic consumption is becoming increasingly prevalent. This poses a significant threat to consumer privacy and property security. In reaction to this issue, an array of methods for deepfake detection has been proposed to evaluate the authenticity of images. Sequential deepfake detection is an extension of the deepfake detection approach. It aims to detect various facial manipulation operations and accurately identify the sequence of facial manipulations. To enhance the accuracy of sequential deepfake detection and protect consumer privacy, we propose a deep learning-based detection method for sequential deepfake detection. It is designed to extract fine-grained features for detecting facial manipulation sequences. Compared to the state-of-the-art methods, the proposed method improves the Fixed-Acc and Adaptive- Acc metrics by 1.43% and 3.89%, respectively.