Video watermarking using neural networks

Abstract

Copyright protection for videos is important to prevent revenue loss for video generation companies by using video watermarking methods. Though many methods exists, still certain scope is noticed in robust video watermarking methods. Achieving features like trade-off between robustness and imperceptibility, speed, blind watermarking simultaneously is very challenging. The proposed work achieves the above said features using log-polar, DWT, and SVD techniques to embed watermark in a video and extract it when necessary. The objective is to protect the copyright and make the watermarking system blind, robust against frame drop attacks as well as achieving above features. This work also leverages scrambling, deep learning-based approach to generate secret sharing image from watermark to improve the speed compared to conventional tabular-based approach. We evaluated the method on our own dataset and proved that this method is outperforming compared to state-of-the-art methods in DWT and SVD domain.