Published Paper Details:

Generative artificial intelligence technology has allowed for the creation of highly realistic synthetic videos called "deepfakes." Deepfakes offer a way to enhance many different uses of media; however, they are also responsible for major problems like misinformation, identity theft, and threats to digital security [1],[2]. Consequently, it is challenging for computer vision methods to differentiate between real and deepfake videos. This paper presents a deep learning framework based on a spatio-temporal approach to identify AI-generated deepfake videos. The spatio-temporal approach includes video preprocessing to retrieve video frames, detect and normalize faces in those frames, and then use convolutional neural networks (CNN) to retrieve spatial features from those normal faces. After retrieving the spatial features from the video frames, the video frames are analyzed for temporal inconsistencies between frame relationships to increase detection accuracy [3],[4]. The detection system is created using a web application developed using Flask, allowing real-time video analysis. The experimental results of this research show that the proposed system exhibits constant improvement (as indicated by reduced loss and increased accuracy) until a steady state is achieved as training progresses. Therefore, the proposed deepfake detection system will show accurate results and provide the computational efficiency required for deployment in a real-world context.