Survey on Deepfake Detection for Preventing Audio and Video Frauds Using Advanced Deep Learning Techniques

In order to effectively address the growing threat of deepfakes, we have developed sophisticated preprocessing techniques and utilized state-of-the-art model architectures. To accurately extract facial regions from our dataset, we refined it using Multi-Task Cascaded Convolutional Neural Network (MTCNN). We then combined ResNeXt, ResNet, and Long Short-Term Memory (LSTM) architectures for their ability to handle complex visual data and temporal connections. The resulting ensemble demonstrated exceptional detection accuracy against evolving deepfake technology on three important datasets – Facebook Deepfake Detection Challenge (DFDC), FaceForensics++ (FF-DF), and Celeb DF-datasets. Our method's generalizability was confirmed through comprehensive assessments across these diverse scenarios.