Description of Invention
Deepfake technology has advanced at a very fast rate, and there are severe concerns about its misuse for the dissemination of misinformation, fraud, and privacy violations. Its capability to edit video and audio content with high accuracy has made it hard to identify these manipulations. This paper introduces a new approach to deepfake detection using a multimodal pipeline that combines spatial, temporal, and frequency-domain analysis to improve detection accuracy and reliability. The employed dataset is the UADFV[1], preprocessed and optimized for effective training and testing. A tailored Convolutional Neural Network (CNN) model was trained to detect deepfake frames with high precision, recall, and accuracy. The results affirm the efficacy of preprocessed frames and a robust detection algorithm in detecting manipulated content. The study highlights the importance of developing advanced detection systems to counter the growing threat of deepfake technology, offering trust and authenticity in digital media.