ROBUST DETECTION OF INTRA-FRAME COPY-MOVE FORGERIES IN DIGITAL VIDEOS

With over 3.7 million videos shared daily on platforms like YouTube and social media, the proliferation of high-quality forged videos is rapidly increasing. Such forgeries compromise the authenticity and integrity of digital evidence, potentially leading to serious consequences. For instance, in judicial proceedings, a tampered video used as evidence could wrongfully implicate an innocent person or help a guilty individual evade justice. This necessitates robust detection mechanisms to counteract forgery attempts. One prevalent method of forgery is copy-move video forgery, which involves duplicating regions within a single video frame or across consecutive frames. Traditional detection approaches rely on manual pattern recognition and block-matching, often yielding detection accuracies below 70%, particularly in high-resolution and compressed videos. In contrast, deep learning-based techniques have shown significantly improved performance, with Convolutional Neural Network (CNN) and Transformer-based models achieving up to 92.6% accuracy on standard datasets like Kaggle and FaceForensics++. This research leverages pre-trained deep learning architectures to automatically learn discriminative features, enhancing the detection of copy-move forgery in complex and dynamic video environments.