Soft Comput (2009) 13:375–382 DOI 10.1007/s00500-008-0328-6 FOCUS A fast SVD based video watermarking algorithm compatible with MPEG2 Standard Di Wu · Wenhai Kong · Bian Yang · Xiamu Niu Published online: 25 June 2008 © Springer-Verlag 2008 Abstract This paper presents a real time video watermar- king algorithm based on the singular value decomposition (SVD). The embedding procedure combines the DCT and SVD. The SVD is applied on the low frequency AC coef- ficients of block DCT of the frame. Then the relation of neighbor coefficients of those middle frequency bands in the singular values is modified. The hidden information can be extracted without the original video or the order of the ori- ginal singular values, i.e., this is a blind watermarking algo- rithm. Experiments show that the algorithm is very robust against MPEG-2 compression, median filtering, small shift, and rotation, etc. Keywords Video watermarking · SVD · DCT 1 Introduction Digital watermarking is one of the efficient tools applied in digital data protection. It embeds various types of copyright information into the digital multimedia carrier such as image, audio, video, etc. Commonly, watermarks can be embedded either by directly modifying the samples of the original car- rier or modulating the coefficients in the transform domain, and in most cases, watermarks should be imperceptible and robust to potential attacks. D. Wu (B ) · W. Kong · B. Yang · X. Niu School of Computer Science, Harbin Institute of Technology, Harbin 150080, China e-mail: di.wu@ict.hit.edu.cn In the research domain of video watermarking, water- mark can be embedded in the spatial domain (Mobasseri 2000) or transform domain such as DCT (Wu et al. 2004; Tsai and Chang 2004; Ge et al. 2003), DFT (Barni et al. 2000) and DWT (Liu et al. 2002; Niu and Sun 2001). The techniques are well summarized in Pan et al. (2007) and Pan et al. (2004). We consider the singular value decom- position (SVD) (Kahaner et al. 1989) based watermarking algorithm in DCT domain. Traditionally, SVD is employed in image compression, character extraction and other signal processing scenarios (Andrews and Patterson 1976; Garguir 1979; OLeary and Peleg 1983). In some recently published papers (Liu and Tan 2002; Zhou and Chen 2004; Ganic and Eskicioglu 2005, 2004), SVD based image watermarking algorithms are proposed, declaring that their watermarking algorithms are blind. But as claimed in these papers, the algo- rithms need the singular values or the orthogonal matrices as a secret key for watermark retrieval. Because the original sin- gular values are in decreasing order, to modify the singular values may destroy the order. But after the next SVD is perfor- med, the decomposed singular values are ordered again. The- refore, the synchronization of the watermark is lost, which may cause the wrong extraction. So traditional algorithms must get the position information of the modified singular values in advance. Since these algorithms need cover work relevant information for watermark retrieving, they cannot be used in video watermarking applications considering the huge storage of the prior information. Another shortcoming of SVD is the high computing complexity. In our previous work (Kong et al. 2006), the algorithm is rather slow and not very practical. In this paper, a fast SVD based algorithm implement on blocked DCT AC coefficients is proposed. This algorithm is compatible with the MPEG2 I-frame, and the original order of singular values can be maintained. Experiments 123