Detecting and Watermarking Fake Videos using AI Model

Abstract

Detecting deepfake videos has become crucial for ensuring media authenticity in the digital age. This work presented an AI-driven system that utilized a custom Convolutional Neural Networks (CNNs) for spatial analysis to detect manipulated media. The system also integrated a novel watermarking approach to mark identified deepfakes, thereby enhancing traceability and accountability. To improve detection accuracy, the custom CNN model was trained on diverse datasets to generalize across various deepfake techniques, including FaceSwap. The proposed method addressed key challenges such as computational complexity, adversarial robustness, and real-time processing efficiency. Additionally, the watermarking solution embedded digital signatures within detected deepfake media, enabling their identification even after post-processing modifications. By combining deep learning-based detection with watermarking, the system provided a comprehensive approach to mitigate misinformation and preserve media integrity. This research contributes to the ongoing fight against digital deception and offers a scalable, effective solution for deepfake detection.