Advancing Deepfake and Face Mask Detection Through Deep Learning in the AI Era

Detection of deepfakes is important because the technology can seriously undermine the authenticity of digital media. In this paper, a novel deepfake detection method is presented by employing a Recurrent Neural Network (RNN) with the Inception V3 model, and further optimized with MobileNet to facilitate effective feature extraction. The RNN model improves detection accuracy by learning temporal dependencies between video frames, allowing it to identify subtle differences that are typically inherent in deepfake content. The integration of MobileNet’s lean architecture with Inception V3’s high-fidelity feature descriptions facilitates strong extraction capabilities with minimal computational needs. The RNN-based method established has an 82% accuracy. In parallel, face mask detection is explored using a Convolutional Neural Network (CNN) with pre-trained ImageNet weights. This configuration enables precise localization and classification of masked and unmasked faces to mitigate the spread of infectious diseases in public areas. With the use of pre-learned features of ImageNet, the model exhibits excellent generalization across diverse mask types and facial variations. The proposed methods are highly accurate in the tasks of deepfake and face mask detection, based on experimental results. The CNN model scores a remarkable 96% in detecting masks.Unlike prior models, we integrate RNN with InceptionV3 and MobileNet for enhanced deepfake detection, and a CNN-ImageNet pipeline for face mask detection, achieving improved accuracy with reduced computational cost.