GEOMETRICALLY INVARIANT OBJECT-BASED WATERMARKING USING SIFT FEATURE Viet Quoc PHAM  Takashi MIYAKI ‡ Toshihiko YAMASAKI  Kiyoharu AIZAWA  Dept. of Information and Communication Engineering Dept. of Frontier Informatics, The University of Tokyo E-mail: {pqvietvn, miyaki, yamasaki, aizawa }@hal.k.u-tokyo.ac.jp ABSTRACT In this paper, we have developed a robust object-based watermarking algorithm using the scale-invariant feature transform (SIFT) features in conjunction with a new data embedding method based on Discrete Cosine Transform (DCT). The message is embedded in DCT spaces of randomly generated blocks in the selected object region. To recognize the object region after being distorted, its SIFT features are registered in advance. In the detection scheme, we firstly detect the object region by using feature matching. The transformation parameters are then calculated, and the message can be detected. Experimental results demonstrated that our proposed algorithm is very robust to geometrical distortions such as JPEG compression, scaling, rotation, shearing, aspect ratio change, image filtering, and so on. Index Terms— Digital watermarking, geometrically invariant, scale-invariant feature transform, object matching 1. INTRODUCTION Owing to the development of the Internet, digital imaging has experienced tremendous growth over the last decade. We now can easily find and download a large number of images within a few seconds. In order to protect and preserve the owner’s right, a number of copyright protection methods have been proposed. Digital watermarking is a technology used for copy control and media identification and tracing. In digital watermarking, they embed a short message (a watermark) in an image or video without affecting the quality but that can be detected using dedicated analysis program. Due to the advances in image and video editing software, it has been made possible to copy a certain object in an image or a frame and paste it to the others. In addition, such illegally copied object may be geometrically distorted by lossy compression, affine transforms, and so on. The purpose of this paper is embedding and detecting a watermark in such a situation. In this point of view, the scope of this paper is different from conventional object- based watermarking for MPEG-4, in which object layers are pre-defined. In our case, we do not assume such predefined object layers nor do we have to conduct object segmentation for watermarking. In this paper, we have developed a robust object-based watermarking algorithm using object matching in conjunction with a new data embedding method based on Discrete Cosine Transform (DCT). The general idea of this method is shown in Fig. 1. The watermarked object “akiyo” in Fig. 1 is attacked by being mixed with another object and then geometrically transformed. To detect the hidden information in the object, we first detect the object region by using object matching. And by calculating the affine parameters, we can geometrically recover the object, and can easily read the hidden message. In our method, we employed the SIFT feature [1] for the object matching operation. The experimental results demonstrated that our method can resist to very strong attacks such as 0.4x scaling, all angle rotation, 30° shearing, JPEG compression (Q=20), StirMark random distortions, or the combination of them. 2. RELATED WORKS There are several approaches related to geometrically invariant watermarking. We categorize them into two groups. In the first group, the watermarking systems employ object segmentation. As a representative of this group, Dajun et al. [2] proposed an object-based video authentication system in which a set of angular radial transformation coefficients was selected as the feature to represent the video object and the background. Error correction coding and cryptographic hashing were applied to those selected coefficients to generate the authentication watermark. The watermark embedding and extraction were done by modifying Discrete Fourier Transform (DFT) coefficients. In their detection scheme, the scaling ratio was supposed to be known so that the received video object could be scaled back to its original resolution. Besides, there are some other methods that employ object segmentation, such as [3] [4], etc. One of the most significant disadvantages of such methods is that object segmentation V - 473 1-4244-1437-7/07/$20.00 ©2007 IEEE ICIP 2007