Advancing Deepfake Detection: An Optimized Neural Network Model with Keyframe Analysis

Artificial intelligence (AI) can rapidly progress and foster AI algorithms to create a new machine-created video called deepfakes. These misleading videos are serious threats to our society and politics, or they can be used as weapons against the viewers. Deepfakes refer to synthetic media created with AI; therefore, such techniques can produce real-looking visual and auditory content. The first element of mitigating the malice of such videos is detecting them on social media platforms. This paper proposes a novel neural network-based method for deepfake video detection, focusing on critical video frame extraction to lessen computational complexity. The proposed method utilizes the architecture of a CNN coupled with a classifier network and a unique algorithm to accurately detect deepfake videos while keeping very low computational needs. Processing key video frames achieves impressive results based on single-keyframe processing for deepfake video detection. The technique is simple and efficient in authenticating video content concerning the social and economic problems of false videos. After evaluation over an entire dataset of 600 videos downloaded from various websites, the model detects deepfakes across different databases with an overall accuracy of 97.3%. Its performance surpasses that of earlier methods with reduced computation time, thus providing an efficient solution for deepfake detection.