Blind Video Watermarking for the Detection of Illegal Downloading of Video Content

Abstract

Illegal distribution of a digital movie is a significant threat to the film industry. Currently, camcorder theft is the single largest source of video piracy and is one of the most significant problems facing movie studios and producers. Approximately ninety percent of the first versions available of illegally distributed new release films have been pirated using a digital camcording device in a movie theatre. With the advent of high-speed broadband Internet access, a pirated copy of a digital video can be easily distributed to a global audience for viewing online and downloading within just days of its release in theatres. Even though there are strict laws in many countries against video piracy, they have proven to be ineffective and it has not been possible to prevent this practice. Therefore, protecting the copyright of digital video content is becoming very important. As the popularity of 3D videos is increasing daily, concern over protecting them from unauthorised distribution is also growing. Digital video watermarking is a possible means of limiting this type of digital distribution. This is the process of embedding extra information in a host video signal so that the watermark is imperceptible and, in a blind detection system, it is robust and difficult to remove or alter. In existing watermarking methods, the watermark is usually embedded in the luminance (Y) channel in a YUV representation of a video frame which affects imperceptibility. In addition, none of the existing techniques are robust to the combination of commonly used attacks, including lossy compression, upscaling, rotation, cropping, downscaling in resolution, aspect ratio change, frame rate conversion and camcording. The key focus of this work is to design a digital video watermarking algorithm for both 2D and 3D content that can mitigate their associated challenges. In the first part of this dissertation, a blind digital video watermarking algorithm is proposed. The watermark is embedded in one level (low-frequency coefficients) of the dual-tree complex wavelet transform (DT CWT) of the chrominance (U) channel, and extracted using the same key and level used for embedding. This embedding in the U channel provides a high-quality watermarked video and the DT CWT enhances its robustness to geometric attacks, such as scaling, rotation and cropping, due to its approximate shift invariance characteristic. To evaluate the overall performance of the proposed method for H.264/AVC compression and geometric attacks, an experimental comparison of the proposed U channel embedding approach, Y channel embedding and two existing methods is undertaken. The experimental results verify the efficacy of the proposed algorithm. The downscaling in resolution of a frame in the spatial domain removes the high-frequency information. As a result, the watermark and low-frequency coefficients in the DT CWT domain spread into the new high-frequency coefficients. In such a situation, watermark detection results in a false negative error if it extracts from only the level used for embedding. Therefore, in the second part of this work, an extension of the first method extracts the watermark from any level(s) of the DT CWT decomposition depending on the resolution of the downscaled version of the watermarked frame. However, this method still suffers in terms of temporal synchronisation attacks, such as frame rate conversion, as the embedding key is required for detection. To overcome this limitation, the watermark of a frame is extracted from the information of that frame without using the key used during its embedding. The experimental results show that the proposed method is not only robust to temporal synchronisation attacks and downscaling to an arbitrary resolution but also upscaling, rotation, cropping, lossy compression, camcording, watermark estimation remodulation, temporal frame averaging and multiple watermark embedding. In the last part of this research, the emphasis is on protecting a depth-image based rendered (DIBR) 3D video. The proposed scheme embeds the watermark in both the chrominance channels (U and V) of the centre view using the DT~CWT. Then, the synthesised watermarked stereo pairs (left and right views) are generated at the decoder from both the depth map and watermarked centre view using the DIBR process. The watermark is extracted independently from the centre, left and right views which can be distributed individually as 2D content. The quality of the watermarked 3D video is evaluated by a subjective test and compared with that of a DIBR 3D image watermarking algorithm. The experimental results demonstrate that the proposed scheme provides superior performance to those of state-of-the-art approaches in terms of Gaussian noise addition, baseline distance adjustment and 3D camcording, as well as the attacks considered in the previous part of this study for 2D video.