A blind spatio-temporal data hiding for video ownership verification in frequency domain

Abstract

By widespread usage of video-based applications, digital watermarking has been recommended as a solution to prevent illegal and malicious copying and distribution of digital media. In this paper, a blind watermarking system for video ownership verification is proposed. Inspired by the used code division multiple access (CDMA) techniques in image watermarking with 2D form, and considering that video is inherently three dimensional, we expand CDMA techniques in 3D form based on discrete wavelet transform for watermarking of video data. First, a 3D-DWT is performed on the video frames. Then, by using CDMA techniques and two different pseudo random number (PRN) sequences, a binary image (as a watermark) is spread into mid and high frequency sub-bands of wavelet. In extraction step, the original video is not needed, namely, blind detection. Eventually, by defining a heuristic algorithm, the watermark is clearly detectable. Unlike other schemes, which are resistant to a limited series of attacks, the proposed method has admirable robustness against a large range of attacks such as frame averaging, frame dropping, frame transposing, median and low-pass filtering, Gaussian noise, salt and pepper noise, speckle noise, resizing and a selected group of lossy compressions like MJPEG, MPEG and H.264/AVC. In addition, evaluation of well-known visual quality assessment (VQA) metrics like PSNR, MSE, SSIM, MS-SSIM and MOVIE shows the high transparency of the proposed system for human visual system (HVS).