Deepfake Detection Using Deep Learning: A Unified Forensic Approach to Detect AI-Generated Images and Videos with Fusion of Eye, Nose, and Mouth Landmarks

The growing prevalence of AI-generated multimedia content, particularly deepfakes, has facilitated the spread of fake news, propaganda, identity theft, and undermined the importance of privacy and authenticity in digital media. Existing techniques often lack robustness in handling complex deepfake artifacts, especially those involving fused facial regions such as the eyes, nose, and mouth. This study addresses the challenges posed by deepfakes across domains such as social media, politics, and entertainment by proposing a hybrid deep learning framework for forensic analysis. The framework integrates convolutional neural networks (CNNs), recurrent neural networks (RNNs), generative adversarial networks (GANs), and temporal convolutional networks (TCNs) to effectively capture both spatial and temporal features. Experimental results demonstrate a detection accuracy of 99% across diverse datasets, highlighting the model’s effectiveness in detecting deepfakes, even with intricate manipulations. This research advances the field of digital forensics by enhancing methodologies for multimedia verification, promoting trust, and safeguarding authenticity in the digital age.