Abstract:
Deepfakes can be used to manipulate digital media, creating security concerns on the internet.
As the capabilities of deepfake creation techniques continue to improve, deepfake detection
is becoming increasingly challenging. Several methods have been developed to detect
deepfake videos using convolutional neural networks (CNN), and recurrent neural networks
(RNN). Instead, preventing deepfakes from being created in the first place is becoming more
important. To address this issue, a new deep learning approach was proposed, which combines
watermark encryption with deep learning steganography techniques. This approach embeds
an invisible watermark in video frames by following new technique, making it difficult for
attackers to remove or modify the watermark without our network. The watermark extractor
algorithm used to validate the video requires the same "attention model" to decode the
watermark, making it difficult for attackers to apply deepfake to the frame. The proposed
approach was able to completely prevent deepfake attacks with a 100% success rate.
Moreover, we developed a mobile application using flutter on which we will implement the
research work and train our own model to make it available for end user.
