A NOVEL EMBEDDING METHOD FOR AN ANTI-COLLUSION FINGERPRINTING BY EMBEDDING BOTH A CODE AND AN ORTHOGONAL FINGERPRINT Dalwon Jang and Chang D. Yoo Dept. of EECS, Div. of EE, KAIST, 373-1, Guseong Dong, Yuseong Gu, Daejeon, 305-701, Korea dal1@kaist.ac.kr and cdyoo@ee.kaist.ac.kr ABSTRACT In this paper, a fingerprint embedding method better-suited for the AND anti-collusion code (AND-ACC) is proposed. The proposed method embeds both a code and an orthogonal fingerprint using different basis vectors depending on the bit. Although the detec- tion for the embedding method is complex, the performance of the fingerprinting system using proposed embedding method with the AND-ACC against average attack is improved compared with the AND-ACC fingerprinting scheme using code modulation embed- ding method. The system using the proposed embedding method is robust against the linear combination collusion attack (LCCA) whereas the system using the code modulation is not. 1. INTRODUCTION With the increase in Internet users,, the unlawful distribution of digi- tal data is becoming more prevalent. Digital fingerprinting is a tech- nique to prohibit the illegal redistribution of digital multimedia data by embedding into the media a unique label, known as a fingerprint. The distributor can be identified by extracting the fingerprint from the illegally distributed media. From the standpoint of the illegal distributor, a cost-effective attack against fingerprinting is the collu- sion attack where a group of users combines their copies to make an illegal version of the copy. A fingerprinting system must be resistant to such an attack. A fingerprinting can be classified as either an orthogonal or coded fingerprinting. In orthogonal fingerprinting, a unique spread spec- trum sequence assigned to each user is embedded into the media[1, 2]. Whereas, in coded fingerprinting, a code which has good anti- collusion property is constructed and embedded in the media. Ear- lier works on coded fingerprinting focused mainly on the coding layer without considering the detection performance of the embed- ded information[3, 4, 5, 6]. To deal with various practical issues, the performance of fingerprinting has to be evaluated taking into ac- count all steps involved in the fingerprinting: the embedding, the attack, the detection, and the coding[7]. Recent research deals with the embedding layer as well as the coding layer[7, 8, 9]. To construct a fingerprinting system, both the embedding and detecting methods should be considered as well as the design of the code. Trappe et al. proposed a code known as the AND anti-collusion code (AND-ACC)[9]. In the paper, the code modulation method is used to embed the code. This method uses an antipodal signals to embed a bit. Though the signal to represent a bit with a single basis vector is efficient in signal space, it is possible to enhance the embedding method by substituting the antipodal signals by various basis vectors. In this paper, this is what is investigated. In the proposed method, bit ‘0’ is represented by a single basis vector, but bit ‘1’ is represented by any one basis vector from a se- lection of basis vectors. By properly selecting a vector from a set of basis vectors for bit ‘1’ depending on the bit position, an orthogonal fingerprint for each user can be constructed. Using the detecting re- sults for both the embedded code and the orthogonal fingerprint, the colluders can be identified. The remainder of this paper is organized as follows. In Section 2, the embedding and detection methods for the AND-ACC in [9] are explained. In Section 3, a basis vector set for embedding a single bit is explained. In Section 4, the proposed embedding and detection methods are explained. In Section 5, experimental results to verify the method are presented. Finally, Section 6 concludes the paper. 2. EMBEDDING AND DETECTION FOR AND-ACC The AND-ACC is constructed based on the AND assumption that the estimated bit stream extracted from a colluded media should be the same as the logical AND of all fingerprint codes that colluded[9]. Thus, the embedding method must satisfy the assumption. In the AND-ACC fingerprinting system, the code modulation embedding method is used. In code modulation, the fingerprint, w i , is modelled as wi = v  j=1 bij uj (1) where {u j }, (j =1, 2, ··· ,v) is the orthonormal basis to represent the j th bit. The coefficient b ij where b ij ∈{0, 1} or b ij ∈ {±1} is determined by the fingerprint code. The embedding method is gen- eral enough to be used with any fingerprint code. The watermarked signal Y i for the ith user is given by Yi = X + αwi (2) where X is the host signal, and α is a constant used for perceptibility constraint. The bit stream of a colluded copy is detected and determined using the correlation value of the extracted watermark of colluded copy and each basis vector[9]. 3. BASIS SETS FOR EMBEDDING A BIT In the code modulation method, a single basis vector is used to rep- resent a bit because the antipodal signals are used. But a bit does not necessarily have to be represented by a single basis vector. It is pos- sible to assign a different basis vector for each bit value. Moreover,