COMPENSATED SIGNATURE EMBEDDING BASED MULTIMEDIA CONTENT AUTHENTICATION SYSTEM Sufyan Ababneh, Ashfaq Khokhar, Rashid Ansari Dept. of Electrical and Computer Engineering, University of Illinois at Chicago, Illinois, USA ABSTRACT Digital content authentication and preservation is an extremely challenging task in realizing decentralized digital libraries. The concept of compensated signature embedding is proposed to de- velop an effective multimedia content authentication system. The proposed system does not require any third party reference or side information. Towards this end, a content-based fragile sig- nature is derived and embedded into the media using a robust watermarking technique. Since the embedding process intro- duces distortion in the media, it may lead to authentication fail- ure. We propose to adjust the media samples iteratively or using a closed form process to compensate for the embedding distor- tion. Using an example image authentication system, we show that the proposed scheme is highly effective in detecting even minor modifications to the media. Index Terms— authentication, preservation, watermarking, com- pensated signature embedding 1. INTRODUCTION The unprecedented accessibility of online multimedia contents in re- cent times has given rise to their vulnerability to corruption and tam- pering, particularly in an open, peer-to-peer, and possibly malicious environment. Several digital library initiatives have been launched to design and develop technologies for authentication and preser- vation of contents, format, presentation and functionality of online multimedia documents. In this regard, authentication and preser- vation of multimedia contents such as video, images and audio is extremely challenging particularly in an open decentralized environ- ment where a central entity is not always available or may have been compromised. We argue that the amenability of non-textual multi- media data to imperceptible insertions offers feature-rich alternatives for authentication and preservation. Depending on the end user ap- plication, authentication may or may not tolerate modifications to the original data. Both fragile and robust watermarking techniques have been used in conjunction with content-based signatures to authenticate a digital media. A disadvantage of these schemes arises from the fact that the very process of embedding a watermark alters the media, causing the subsequent authentication test to fail if the signatures are very fragile (highly sensitive to modifications). To prevent this, most of the schemes either do not use content-based signatures or tend to divide the media domain into two parts: verifiable part from which the signature is derived, and embedding part where the signature is embedded. The level of robustness or fragility is determined based on the application requirements. Robust watermarking schemes combined with robust content- based signatures have been used for authentication and quality as- sessment of multimedia [1][2]. However, the target applications for these schemes -such as multimedia retrieval- require tolerance to- ward normal and incidental distortions while detecting tampering and significant distortions. On the other hand, fragile watermark- ing based authentication techniques can use high sensitive signatures (fragile) [3] [4]. In such schemes, the watermark is lost and signa- tures are altered as soon as any modification is applied to the media. It has also been used without content-based signatures. However, in such schemes the use of predefined patterns can not guarantee that no one has intentionally tampered with the media. This is because an adversary can forge a fragile watermark if the embedded pattern is not derived from media contents [5]. In this paper, a new multimedia authentication system is pro- posed which allows the combining of robust watermarking and frag- ile signatures to realize a highly effective authentication framework in the absence of a central entity or side information. The system allows for the flexibility to use any robust embedding technique in conjunction with the proposed compensated signature embedding (CSE) concept to provide feature-rich multimedia authentication. This scheme can provide an ideal mechanism for digital multimedia preservation in which verifying the authenticity of the multimedia is an essential step before further attempts for recovery are initiated. The proposed scheme can be applied on any type of digital media in- cluding still images, video, audio, text and any combination of them. 2. CSE AUTHENTICATION SYSTEM The framework of the compensated signature embedding scheme (CSE) consists of two main functionalities: encoding and decoding. As shown in Fig.1, the encoder generates a fragile signature, embeds a robust watermark and compensates for signature embedding. The decoder extracts the embedded signature, generates a new signature and evaluates the results. All these functionalities operate on the me- dia signal itself or its transformed version. For convenience, some notation is introduced. The set of integers is denoted by Z. For any positive integer K, let I K = {k ∈Z :0 ≤ k ≤ K − 1}. The set D denote the domain of the signal. Let M and N be positive integers that are multiples of 2 L , where L is also a positive integer. Define M l = M/2 l and N l = N/2 l , for l ∈Z, 1 ≤ l ≤ L. For a raw M × N image, D = IM × IN . We are interested in the subband/wavelet representation of the media using L-level decomposition. Although D defined for the raw media will work, we will define it differently in order to capture the structure of the wavelet decomposition at different levels and we will use image signals for illustration: D = {n =(l, n l , i, j ): l − 1 ∈ IL,n l ∈ I 3+λ ,i ∈ IM l ,j ∈ IN l } (1) Where λ = δ lL and δ kl is the Kronecker’s delta. The signal w (i.e. the image or its transformed version) is defined as a mapping {w : D −→ R}, with some additional structure such as square I - 393 1-4244-1437-7/07/$20.00 ©2007 IEEE ICIP 2007