Watermarking Motion Data

teaser teaser

This research investigates a robust motion data watermarking related to the human motion data. Our method can deal with the motion represented by joint angles or marker positions of an articulated link model. Motion data are usually supposed to undergo through considerable signal processing, which makes it difficult for the embedded watermark to survive adversary attacks. Moreover, the existing watermarking techniques lack the consideration of a major problem like its disappearance in the computer graphics (CG) animation. Aiming to make the watermarking robust, we insert a watermark into the original motion data in the frequency domain, distributing it redundantly across different components of the motion data structure. We also propose a method of finding the watermark embedded in a set of motion data in order the copyright violation can be detected not only in their original format but also in a CG animation sequence which is generated using the watermarked motion data. Certain experimental results are presented for various motion capture data to show that the proposed watermarking method is imperceptible, while it is robust with respect to various types of possible attacks.

Publications

PDF

Shuntaro Yamazaki,
Watermarking Motion Data”,
In Proc. Pacific Rim Workshop on Digital Steganography 2004, pp.177-185,
Nov. 2004

PDF

Shuntaro Yamazaki, Masaaki Mochimaru and Takeo Kanade,
Watermarking Motion Clips”,
In Proc. International Conference on Computer Animation and Social Agents 2005, pp. 171-176,
Oct. 2005

PDF

News article, Nikkei Sangyo Shinbun,
Apr 5, 2005
(Japanese)

Supplementary Material

Demo video

thumbnail

This video is a compilation of the main results of this project. (Watch this at YouTube.com)

Figures in Paper

thumbnail

Overview of Watermarking Process: This picture illustrates the overview of our watermarking process.

thumbnail

Trajectory of a joint: The row of bars means the list of points, and yellow bars is rest points after prunning.

thumbnail

MoCap data used in our experiments. The original and watermarked data are shown in (a) and (b). We simulated various attacks to (b) as shown in (c) - (h). The figures (j) - (l) are the results of our motion registration.

Miscellaneous