RST invariant video watermarking based on 1D DFT and Radon transform

Abstract

In this paper, we propose a new video watermarking algorithm robust against geometric attacks, such as, format aspect ratio change, rotation, cropping, and translation. The watermarking algorithm is also robust against other attacks, such as frame swapping, frame dropping, noise addition, filtering, and compression. This algorithm is scene based. The scene detection is applied to segment a video sequence into different scenes before watermark embedding. The histogram difference is used for scene change detection. 1D Discrete Fourier transform (DFT) within temporal domain is applied in a group of picture (GOP). In the temporal frequency domain, we select the high frequency frame as our watermark embedding frame. We embed a random generated watermark pattern into the ID projection of this frame. The blue channel is the only channel we consider for embedding based on the fact that human visual system is least sensitive to this component. The experimental results demonstrate the feasibility of the algorithm.