A robust video watermarking based on image mosacing and multi-frequential embedding Asma Kerbiche Lab. RIADI - Team of Research SIIVA Higher Institute of Computer Science University Tunis El Manar Abou Raihane Bayrouni, 2080, Ariana, Tunisia Email: asma.kerbiche@gmail.com Saoussen Ben Jabra Lab. RIADI - Team of Research SIIVA Higher Institute of Computer Science University Tunis El Manar Abou Raihane Bayrouni, 2080, Ariana, Tunisia Email: saoussen.bj@laposte.net Ezzeddine Zagrouba Lab. RIADI - Team of Research SIIVA Higher Institute of Computer Science University Tunis El Manar Abou Raihane Bayrouni, 2080, Ariana, Tunisia Email: ezzeddine.zagrouba@fsm.rnu.tn Abstract—Video watermarking is very important in many areas of activity and especially in multimedia applications. There- fore, this domain has attracted more and more attention in both the research and industrial domains. Many video watermarking approaches are proposed but, based on our knowledge, there is no method which verified the compromise between invisibility and robustness against all usual attacks. This paper proposes a new video watermarking approach which is robust against many attacks such as collusion, MPEG4 compression and noise. This approach inserts the signature in the mosaic generated from original video and exactly in the region where the objects move. This region is selected after many tests applied on different feature regions. First, the wavelet transform is applied on the selected region. Then, obtained frequencies are marked using two different approaches to verify the best compromise invisibil- ity/robustness. Experimental results showed that this approach allows obtaining a good invisibility and robustness against the maximum of usual attacks. The performance of the proposed approach was proved by a comparative study with others existing approaches. Keywords–video, watermarking, robustness, invisibility, wavelet, SVD, DCT, mosaic image. I. I NTRODUCTION The recent decade has seen the emergence of video-based application technologies, such as Internet and wireless video, video conferencing, and videophones. Therefore, the need of their protecting from piracy and illegal use rises more and more. The best technique to protect digital video from this manipulation is watermarking. It consists to embed a signature into data and to try to detect it after any manipulation done on marked data. Usually, signature must be imperceptible and should resist to malicious attacks which try to destroy or to remove it. Robustness is often measured in terms of the number of watermarking attacks the watermark is able to resist. Most commons and important attacks for video data are collusion, MPEG4 compression, noise, and frame suppression. The remainder of this paper is organized as follows: in Section 2, an overview of video watermarking is presented while section 3 details the proposed approach based on multi-frequentiel embedding. In fact, feature regions selection is first presented. Then, embedding and detection stages of the proposed watermarking scheme are provided. In Section 4, experiments results are presented to report the performances of the proposed scheme against the most important video attacks. and a comparative study between the proposed approach and other existing video watermarking are shown. Finally, conclusion and perspectives are drawn in Section 5. II. VIDEO WATERMARKING OVERVIEW Existing video watermarking schemes can be classified in different categories according to two criterions. The first one is the format of the original video which can be compressed or uncompressed. The second criterion is embedding domain where signature can be inserted directly on video or by modifying some video transformation like DCT, wavelet and SVD transform. A. Classification according to Video format Based on the video format, watermarking techniques can be classified in two main classes. The first one contains tech- niques which can be applied to uncompressed and compressed video while the second class represents techniques which can only be applied to compressed stream [1, 2, 3, 4]. In this paper, we are only interested by the first class where embedding is done before video encoding and it didn’t depend to video coding algorithm. Many algorithms are proposed for this class and they can be classified in four main categories: 1) Schemes derived from frame by frame watermarking techniques: In this context, video is considered as a succession of images which can be marked using still images watermark- ing techniques. Among this class of watermarking schemes, Chae & al. [5] proposed a video embedding scheme where the signature is generated without knowing the original host video. In this scheme, embedding is based on texture masking and utilizes a multi-dimensional lattice structure for encoding sig- nature information. Signature data is embedded in individual video frames using DCT block. Merchant & al. [6] proposed a video watermarking technique using integer-to-integer discrete wavelet transform. The watermark is embedded in the lowest frequency components of each frame.