Detecting Deepfakes Through the Classification of Facial Active and Passive Features Using Machine Learning

Recently, there has been a much public concern over fake photos and films that contain facial information created by digital modifications, especially with deepfakes techniques. The widely used term “Deepfakes” describes a deep learning technique that swaps faces to produce fake identities. The distinction between actual and deepfake videos has become crucial due to their detrimental effects on the world. An intriguing vulnerability within live deepfakes has largely gone unnoticed by the security research community thus far. This research may be attributed to real-time deepfakes during video calls and conferences, which have recently become a notable cause for concern. Developing and evaluating existing deepfake detection methods requires large-scale datasets. Still, the deepfake datasets available today have inadequate image quality and do not appear like the deepfake videos that have been making the rounds on the Internet. This research presents a novel way for recognizing synthetic media produced using deepfake: an active and passive facial features based end-to-end deep neural network classifier model (AFPF-DNNC model). The facial landmark detector recognizes and extracts the area from each individual video frame that contains the human face, then extracted facial regions used to extract active features in the faces. Simultaneously, a convolutional neural network (CNN) extracts passive features from these faces. The retrieved features are inputs into the end-to-end deep neural network classifier (DNNC), which is the last recognition step in the detection system. The proposed algorithm achieves 90.32% of an area under the receiver operating characteristic curve (AUC), 90.6 % accuracy, 90 % specificity, 91.3 % recall, 87.5 % precision, and 89.3 % F-score on the Celeb-DF (v2) a new deepfake forensics dataset. Finally, the experimental investigation validates the suggested method’s effectiveness over the most advanced techniques.