Secure Multicast of Anti-Collusion Hong Zhao and K. J. Ray Liu Fingerprinted Video Communication and Signal Processing Lab 1. Motivation 3. A Joint Fingerprint Design and Distribution Scheme • Multimedia Fingerprinting: A huge volume of data and a large number of users Unicasting each fingerprinted copy is inefficient Traditional multicast cannot be directly applied Robust against collusion attacks and other attacks Preserve the secrecy of the embedded fingerprint Achieves bandwidth efficiency (1) Robust against many attacks and widely used in multimedia fingerprinting Not all coefficients are embeddable due to perceptual constraints, and a non-embeddable coefficients has the same value in all copies • A general fingerprint multicast scheme: Non-embeddable coefficients Æ multicast to all users Uniquely fingerprinted coefficients Æ unicast • Can be used with most spread spectrum embedding based fingerprint design: (2) u u u U 1,1 (3) u (4) (5) u u U1,2 (6) u (7) (8) N VLC-1 Positions of embeddable coeff. Encrypt VLC Y Embeddable? Q-1 Q VLC User u(i)’s fingerprint Multicast to all users Encrypt VLC-1 Q-1 Multicasted coded DCT coeff. Positions of embeddable coeff. Reconstruct the 8*8 DCT block Motion vector u (10) Improve the perceptual quality (1~1.5dB) without extra communication overhead 2,2,3 (11) (12) u u U 2,2 Motion vectors, Positions of embeddable coeff. quantization factors and side info VLC N VLC-1 If TDMA based fingerprint modulation is used, then some fingerprinted coefficients are also shared by the users in the same subgroup. Achieve both the robustness and bandwidth efficiency Explore the special structure of the fingerprint design to further improve the bandwidth efficiency ¾ Shared fingerprinted coefficients Æ multicast to the users in the subgroup Increase the complexity of the underlying network and that of the sender and the receivers VLC-1 Embeddable? Y Q-1 W1(i1 ) ∈ S1? Y Unicast to user i for user u(i) Q VLC N Encrypt # # WL(i1 ,",iL ) Q K (i ) VLC Encrypt 0 Motion compensation (b) Decoding at the receiver’s side A MPEG-2 based general fingerprint multicast scheme for video-on-demand applications Coded DCT coeff. unicasted to user u(i) # Coded DCT coeff. multicasted to subgroup U(i1) VLC-1 Decrypt VLC-1 Q-1 Decrypt Coded DCT coeff. multicasted to all users K m VLC-1 Q-1 Drift compensation signal - DCT motion Motion vectors compensation + - 5. Bandwidth Efficiency M=1000, L=3 D=[2, 5, 100] ρ=[1/4,1/4,1/2] M=5000, L=4 D=[2, 5, 5, 100] ρ=[1/6,1/6, 1/6,1/2] M=10000, L=4 D=[4, 5, 5, 100] ρ=[1/6,1/6, 1/6,1/2] Unicast to user u(i) Positions of embeddable coeff. Unicast to user i FP drift compensation in general fingerprint multicast scheme for user u(i) Reconstruct the 8*8 DCT block K (i ) L’ MG(L’) RB(L’) Miss am Carphone Flower 0 1 2 0 1 2 3 0 1 2 3 1 3 13 1 3 13 65 1 5 25 125 2 3 4 2 3 4 5 2 3 4 5 0.23 0.22 0.20 0.18 0.16 0.15 0.14 0.16 0.14 0.13 0.13 0.41 0.34 0.31 0.35 0.30 0.27 0.25 0.34 0.28 0.26 0.23 0.52 0.43 0.39 0.46 0.39 0.35 0.32 0.43 0.37 0.33 0.30 • Bandwidth efficiency of the two schemes: Q-1 # K (i1 ) Encrypt for embeddable coeff. ? 6. Conclusions for subgroup U(i1) Multicast to subgroup U(i1) K (i) Decrypt Reconstruct the 8*8 DCT block VLC + Y User u(i)’s fingerprint (a) Fingerprint embedding and distribution at the server’s side IDCT Q-1 Q Q-1 Km K ( i1 ) Y Q-1 Multicast to all users Encrypt VLC N K Multicast to all users m for user u(i) Q-1 • Joint fingerprint design and distribution scheme: Encrypt Embeddable? Coded DCT coeff. • In the tree based fingerprint design, some fingerprints are shared by a subgroup of users ∈ SL? Y Q-1 a2,2,1a2,2,2a Some users are more likely to collude with each other than others due to geographical or social reasons Group users who are more likely to collude with other together to be more robust against collusion attacks (a) Fingerprint embedding and distribution at the server’s side Unicasted coded DCT coeff. VLC-1 (9) • Tree based fingerprint design: Km K (i ) • Fingerprint drift compensation: a2,2 u u U 2,1 Motion vectors, Positions of embeddable coeff. quantization factors, etc. Motion vectors, quantization factors and side info a2,1 a1,1,1 a1,1,2a1,1,3 a1,2,1 a1,2,2 a1,2,3 a2,1,1a2,1,2a2,1,3 a1,2 a1,1 2. General Fingerprint Multicast • Spread Spectrum Fingerprint Embedding: a2 a1 • Desired distribution schemes: The encoder and decoder use reconstructed unfingerprinted and fingerprinted copies for motion compensation. Æ The difference: the embedded FP. FP from different frames may accumulate Protect multimedia content after decryption Embed identification info. in each distributed copy • Distribution of Fingerprinted Copies: 4. Fingerprint Drift Compensation • Fingerprint drift: IDCT 0 Motion compensation (b) Decoding at the receiver’s side A MPEG-2 based joint fingerprint design and distribution scheme for video-on-demand applications General fingerprint multicast: reduces the communication cost by 48% ~ 84%. Joint fingerprint design and distribution: reduces the communication cost by 57% ~ 87%. Depend on: the number of users, the characteristics of sequences & network and computation constraints • Comparison of the two schemes: General fingerprint multicast: recommended for sequences with fewer embeddable coefficients. Joint fingerprint design and distribution: preferred on sequences with much more embeddable coefficients.