Digital Fingerprinting for g g p g f g g p g f Multimedia Forensics

Digital Fingerprinting for Digital Fingerprinting for g g p g fg g p g fMultimedia ForensicsMultimedia Forensics

K.J. Ray Liu, Wade Trappe*, Z. Jane Wang**, and Min Wu

ECE Dept., University of Maryland, College Park* WINLAB/ECE, Rutgers University

Include research sponsored in part by Air Force Research Lab and NSF; and

** ECE, University of British Columbia, Canada

March 2005

p p y ;contributions from Hong Zhao, Shan He, Hongmei Gou, and Zang Li.

Digital Fingerprinting and Tracing TraitorsDigital Fingerprinting and Tracing Traitors

Leak of information as well as alteration and repackaging poses serious threats to government operations and commercial markets – e.g., pirated content or

classified documentstudio

Th L d f

Alicew1

w2

SellSell

Promising countermeasure:

The Lord ofthe Ring

Bob

Carl

w2

w3

grobustly embed digital fingerprints– Insert ID or “fingerprint” (often through conventional watermarking)

to identify each userto identify each user– Purpose: deter information leakage; digital rights management(DRM)– Challenge: imperceptibility, robustness, tracing capability

Digital Fingerprinting for Multimedia Forensics 2

Case Study: Tracing Movie Screening CopiesCase Study: Tracing Movie Screening Copies

Potential civilian use for digital rights management (DRM) Copyright industry – $500+ Billion business ~ 5% U.S. GDPpy g y $5 5

Alleged Movie Pirate Arrested (23 January 2004)– A real case of a successful deployment of 'traitor-tracing'

h i i h di i l lmechanism in the digital realm– Use invisible fingerprints to protect screener copies of pre-release

movies

Carmine Caridi Russell friends … Internetw1Last Samurai

Hollywood studio traced pirated version


http://www.msnbc.msn.com/id/4037016/

Embedded Fingerprinting for MultimediaEmbedded Fingerprinting for Multimedia

Multimedia Document

Customer’s ID: Alice

Fingerprinted CopyFingerprinted CopyMultimedia Document

Customer’s ID: Alice

Fingerprinted CopyFingerprinted CopyEmbedded Finger-

embedembedDigital Fingerprint101101 …101101 …

Distribute to Alice

g p pyg p py

embedembedDigital Fingerprint101101 …101101 …

Distribute to Alice

g p pyg p pygprinting

AliceAlice

BobBobUnauthorized Unauthorized rere--distributiondistribution

AliceAlice

BobBobUnauthorized Unauthorized rere--distributiondistribution

Multi-user Attacks

Collusion Attack Collusion Attack (to remove fingerprints)(to remove fingerprints)

Colluded CopyColluded CopyFingerprinted docfor different users

Collusion Attack Collusion Attack (to remove fingerprints)(to remove fingerprints)

Colluded CopyColluded CopyFingerprinted docfor different users

ExtractExtract AliceIdentifyIdentifyExtractExtract AliceIdentifyIdentifyExtract Extract FingerprintsFingerprints

Suspicious Suspicious CopyCopy

101110 …101110 …

Codebook

Alice, Bob, …

Identify Identify TraitorsTraitors

Extract Extract FingerprintsFingerprints

Suspicious Suspicious CopyCopy

101110 …101110 …

Codebook

Alice, Bob, …

Identify Identify TraitorsTraitorsTraitor

Tracing


AntiAnti--Collusion FingerprintingCollusion Fingerprinting

Build anti-collusion fingerprinting to trace traitors and colluders Potential impact: gather digital evidence and trace culpritsPotential impact: gather digital evidence and trace culprits

– Deter unauthorized dissemination in the first place let the bad guys know their high risk of being caught

– Support secure information sharing in government, military, and intelligence operations

– Commercial use for digital rights management (DRM)

Traditional colluder tracing is designed for generic data– E g Boneh-Shaw (1995)E.g Boneh-Shaw (1995)

Multimedia poses unique challenges and opportunities– Fingerprint should be embedded seamlessly and robustly


– Multimedia-specific characteristics may be exploited

Outline of the TutorialOutline of the Tutorial

Preliminaries – Review of Robust Embedding– Overview of Collusion strategies

Orthogonal Fingerprinting – Collusion Resistance Analysis Group-oriented Fingerprintingp g p g Coded Fingerprinting

Data Embedding in Curves for Fingerprinting Maps Secure and Efficient Distribution of Fingerprinted Video


Preliminary: Robust Embedding and CollusionPreliminary: Robust Embedding and CollusionPreliminary: Robust Embedding and CollusionPreliminary: Robust Embedding and Collusion


Background: Invisible WatermarkBackground: Invisible Watermark

10011010 …10011010 …

© Copyright …


py g

Spread Spectrum Embedding Procedure

( )iii wJNDxy

Human Visual

Watermark wi ~ N (0,1)

JND

iiiy

DCT

Visual Model

IDCT

x y

Watermarked ImageOriginal Image

Key Points: Human perceptual model for imperceptibility Use spreading and noise-like watermark for robustness and


Use spreading and noise-like watermark for robustness and imperceptibility

Detection of Spread Spectrum WatermarkDetection of Spread Spectrum Watermark

Detection can be formulated as a hypothesis testing problem. Optimal detector can be calculated from assumptions onOptimal detector can be calculated from assumptions on

distortion (host media and noise from attacks). If distortion is i.i.d. N(0, d

2) then optimal detector is a correlator:

presenceabsence

H0: <(x + noise), w >H1: < y + noise, w > = < w + (x + noise), w >

orH0 ( i ) i

b/bli d 22TT

TN

H0: < (x + noise) - x, w > = < noise, w > H1: < y + noise - x, w > = < w + noise, w >

b=1b=-1

./(:blind-non

;/:blind22

22

www)y

wwy

TdN

dT

N

T

T


TN Antipodal Code Modulation

Fingerprinting upon SpreadFingerprinting upon Spread--Spectrum EmbeddingSpectrum Embedding

Spread-spectrum embedding– Provide very good tradeoff on imperceptibility and robustness, y g p p y ,

esp. under non-blind detection– Typical watermark-to-noise (WNR) ratio: -20dB in blind

detection 0dB in non-blind detectiondetection, 0dB in non-blind detection

Choice of modulation schemes# of fingerprints

Orthogonal modulation jj uw

B

# of fingerprints= # of ortho. bases

(Binary) coded modulation

for or

i

iijj b1

uw

1,0ijb 1ijb


# of fingerprints >> # of orthorgonal bases

Collusion Attacks by Multiple UsersCollusion Attacks by Multiple Users

Collusion: A cost-effective attack against multimedia fingerprints– Users with same content but different fingerprints come together to g p g

produce a new copy with diminished or attenuated fingerprints– Fairness: Each colluder contributes equal share through averaging,

interleaving and nonlinear combininginterleaving, and nonlinear combining

Originally fi i d

Alice ChrisBob

Originally

Alice Chris

fingerprintedcopies

fingerprintedcopies

Collusion byaveraging 1/3

Colluded copy

Cut-and-paste attack


Colluded copyColluded copy

Order Statistic Based Nonlinear CollusionOrder Statistic Based Nonlinear CollusionAlice Bob Chris

Originally fingerprintedcopies

172 173 176174 175Multi-usercollusion Minimum MaximumMedian (Min+Max)/2

C ll d d


Colluded copy

Linear vs. Nonlinear CollusionLinear vs. Nonlinear Collusion

Linear collusion by averaging is simple and effective Colluders can output any value between the minimum and maximumColluders can output any value between the minimum and maximum

values, and have high confidence that such spurious value is within the range of JND.

I id li ll i ll Important to consider nonlinear collusion as well. Order statistics based nonlinear collusions

kk wgJNDxygV )()( )()(

– Analyze detection stat.’s distribution

m in m a x

m in m a x m in m a x

; ; ; a v e m e d ia nj j j jV V V V

V V V

CC SkjjjSkjj wgJNDxygV )()( )()(

– Many nonlinear collusions can be modeled as attacks by averagingplus additive noise

m in m a x m in m a x

m o d m in m a x

m in

,

w .p .

j j j

n e g m e dj j j j

jr a n d n e g

V a v e r a g e V V

V V V V

V pV


p=> Focus on averaging collusion

m a x

pw .p . 1

jr a n d n e gj

j

pV

V p

Averaging and Nonlinear Collusions (cont’d)Averaging and Nonlinear Collusions (cont’d)

Thresholding detector is robust to different types of attacks:averaging collusion; order-statistic based (min, max, …)

Rationale from detector’s view pointpDetection statistics of averaging and many nonlinear collusions are (approx.) Gaussian distributions with ( pp )same mean

=> Yield similar performance if the overall distortion is the same.

1( , )

1j cg j S

s d

s d

nonlinear attacks

average attacks


2c

jj SK

s d average attacks

Orthogonal FingerprintingOrthogonal FingerprintingOrthogonal FingerprintingOrthogonal Fingerprinting


Orthogonal FingerprintingOrthogonal Fingerprinting

Straightforward concept and easy to implement– Prior works extended from watermarking by Cox et al., Stone, g y , ,

Killian et al.– Advantage in distinguishing individual fingerprints

Di d i fi i i d i ll i– Disadvantage in fingerprint attenuation during collusion

Tradeoff:Orthogonality helps distinguish individual users– Orthogonality helps distinguish individual users

– Correlated fingerprints help resist collusion=> Orthogonal fingerprinting can be good for a small user groupg g p g g g p

How many colluders out of how many users can we trace accurately using orthogonal fingerprinting?


Z. Wang, M. Wu, H. Zhao, W. Trappe, and K.J.R. Liu: "Collusion Resistance of Multimedia Fingerprinting Using Orthogonal Modulation", accepted by IEEE Trans. on Image Proc., June 2004 (to appear in 2005).

Orthogonal Fingerprinting and CollusionOrthogonal Fingerprinting and Collusion

2( ) ~ (0, ), for 1,..., .j j

dd i N i N

y s x

2 2

,

: || || / || ||j k j k

WNR

s s

s d

1 1the correlator ( ) / || ||, for 1,...,T

N jT j j n y s s1 1

c c

j jj S j SK K

y y z s d 2

2

( ) || ||, , ,

(|| || / , ), if ( ( ) | , )

(0, ), o.w.

N j

d cN K c

d

j j

N K j Sp T j H S

N

y

s


How Many Colluders Can We Trace?How Many Colluders Can We Trace? Two issues limit the anti-collusion capability:

– Orthogonal fingerprints get attenuated with more colludersl d t d d d t ti t ti ti di t ll d leads to reduced detection statistics corresponding to colluders

– Probability of false alarm increases as the total # of users increases the detector correlates with more orthogonal fingerprint signals the detector correlates with more orthogonal fingerprint signals,

each corresponding to a user to identify which user is involved assume the same threshold is used on detection statistics false alarm depends on the maximum detection statistics with non- false alarm depends on the maximum detection statistics with non

involved users, which becomes larger as # users increases

Tracing Capability: How many colluders out of how many users are sufficient to break down a fingerprinting system?sufficient to break down a fingerprinting system?

To meet desired probability of detection (Pd) & false alarm (Pfp)– We can analyze the maximum allowable colluders

> Thi id d i id li t fi i ti t f


=> This provides design guidelines to fingerprinting systems for applications with different protection requirements

Formulations for Max. Number of ColludersFormulations for Max. Number of Colluders

Threshold detector: (indices of colluders) 1,..,

ˆ arg max{ ( ) }Nj nT j h

j

(indices of colluders)

Performance criteria: ˆ{ } 1 (1 ( / ))n Kfp r c dP P S Q h j

S stem req irement:

|| || /ˆ{ } 1 (1 ( ))Kd r c

d

h s KP P S Q

j

System requirement:fp

d

PP

maxK The maximum number of colluders

allowed by a fingerprinting system

– The desired Pfp determines an expression of threshold for detector– The desired Pd determines the maximum # of colluders allowed by

d y g p g y


The desired Pd determines the maximum # of colluders allowed by the fingerprinting system

Bounds for Max. Number of ColludersBounds for Max. Number of Colluders

Lower bound and Upper bound for Kmax

Obtained by analytic approximations on Q functions– Obtained by analytic approximations on Q-functions

max 2 2 2min{ , }, where L LN NK n K K

~ N

max 2 2 2

max 1

{ , },log( /(2 log(2 )))

min{ , }, where (1 1 )

L LH

H H K

h n n

NK n K K

h Q

log( )n

(1 1 )KLh Q

– two auxiliary variables are defined as2

1

log(2 )

(1 1 )

L

nL

h n

NK

h Q


(1 1 )Lh Q

Results on Colluder BoundsResults on Colluder Bounds

28K– Stringent requirement: correctly identify at least one colluders without falsely

accusing any– The colluder tracing capabilities for a thousand-user system is limited to several

max 28K


The colluder tracing capabilities for a thousand user system is limited to several dozens colluders

Collusion Resistance: Kmax vs. WNR; Kmax vs. Pfp

based on numerical calculations


based on numerical calculations

Performance Criteria for Different NeedsPerformance Criteria for Different Needs

Why? different goals and interests; a decision-making system consists of many components.a decision making system consists of many components.

Capture one:– To identify at least one colluder with high confidence without accusing

innocent usersinnocent users.

Capture more:– To catch more colluders, possibly at a cost of falsely accusing more– Tradeoff between the expected percentage of colluders that are

successfully captured and the expected percentage of innocent users that are falsely placed under suspicion.

Capture all:– Tradeoff: measured by an efficiency rate, which describes the amount of

expected innocents accused normalized by the number of colluder and


expected innocents accused normalized by the number of colluder and the probability of capturing all colluders.

Performance Criteria for Different NeedsPerformance Criteria for Different Needs

Catch one– With high accuracyWith high accuracy– Measured by Pd and Pfp discussed earlier

C h Catch morethe expected fraction of innocents falsely suspected: ( / )

|| || /th t d f ti f ll d f ll t d ( )

i dr Q hh KQ

s

Catch all

|| ||the expected fraction of colluders successfully captured: ( )cd

r Q

the expected number of innocents captured the expected number of colluders captured

ˆ

R


( )d r cP P S j

Collusion ResistanceCollusion Resistance

Catch morei ir

1( )i dh Q

NK

System requirement Collusion Resistance

c cr max 1 1( ) ( )i c

KQ Q

Catch allDifferent sets of performance criteria were studied. It seems that an orthogonal fingerprinting system can resist to the collusion attacks based oncollusion attacks based on a few dozen independent copies.


GroupGroup--Oriented FingerprintingOriented FingerprintingGroupGroup--Oriented FingerprintingOriented Fingerprinting

Z. Jane Wang, M. Wu, W. Trappe, and K. J. Ray Liu, “Group-Oriented Fingerprinting for Mul-


g, , pp , y , p g p gtimedia Forensics", EURASIP Journal on Applied Signal Processing (special issue on MultimediaSecurity and Rights Management), 2004:14, pp. 2153-2173, 2004.

GroupGroup--Oriented ForensicsOriented Forensics

Overcome the limitations of orthogonal fingerprinting– Recall: orthogonal FP treats everybody equallyRecall: orthogonal FP treats everybody equally– Orthogonal strategy has to suspect more to accurately find a

colluder

Colluders often come together in some foreseeable groups– Due to their geographic, social, or other connections

Our approach: design users’ FP in a correlated way– Cluster users into groups based on prior knowledgeg p p g

Intra-group collusion is more likely than inter-group

– Revise orthogonal FP and add correlation to the same group to


g g phelp narrow down the suspicion group

Proposed Group FingerprintingProposed Group FingerprintingDesign of collusion-resistant fingerprinting systems:

Design of anti-collusion fingerprints to trace traitors and colluders

2

xsy ijij

Design of detection schemes

||||energy equal ;,,...,1for ),,0(~)( 2

sss liNiNid

lmij

d

Solution: construct intra-group FP in two parts and use threshold detector


Solution: construct intra-group FP in two parts, and use threshold detector (at desired intra-group false alarm) to avoid estimating ki

Group Fingerprint DesignGroup Fingerprint Design

Orthogonal modulation between groups– Design L orthogonal sub-systems to represent independent groupsg g y p p g p– M users per group => Total: n = M x L users

Assumption: users in the same group are equally likely to p g p q y ycollude with each other.

Real-valued code modulation within a groupd l l i i hi 1 – Introduce equal correlation within a group

1, 2[ ,..., ]the correlation matrix of { } is

i i iM

i j s

S s s ss R

11

1

s

R

– Each fingerprint consists of common one and individual one:

,{ }i j s 1

)0(~}{where1 21 Niid Iaeeaes


),0( },,...,{ where,1 1 NuiiMiiijij Niid Iaeeaes Can be viewed as a real-valued fingerprint code

ExampleExample Basic idea: first identify groups containing colluders,

then identify colluders within each possible guilty group @

ROC Curves Pd vs. Pfp under different collusion settingsConstraint: equal energy 22

02 ||||}||{||}||{|| syy EE c


Collusion Resistance of Group FP: Collusion Resistance of Group FP: Kmax vs. n

310fpP

0.8dP Group: upper bound

Orth.

Kmax of the proposed scheme is larger than that of the orthogonal scheme (the solid line), when n is large.

Difference between the lower bound and upper bound is due to the fact that ki=K/| i| in our simulations (symmetric collusion pattern)


ki K/| i| in our simulations (symmetric collusion pattern). The smaller the number of guilty groups, the better chance performance.

Extension: TreeExtension: Tree--based Fingerprint Designbased Fingerprint Design

Use tree structure to construct fingerprints combining shared and distinct components

Unified view of fingerprint construction using code modulation– With hierarchically organized basis vectors

Allow for real valued codes– Allow for real-valued codes

a aa1 a2

a11a12 a21 a22

a111 a112 a113a121 a122 a123 a211 a212 a213 a221 a222 a223

s111 s112 s113s121 s122 s123 s211 s212 s213 s221 s222 s223


i1i2 i3 i4

C d d Fi i tiC d d Fi i tiCoded FingerprintingCoded Fingerprinting


Recall: Modulation Schemes for FingerprintingRecall: Modulation Schemes for Fingerprinting

Spread-spectrum embedding– Provide very good tradeoff on imperceptibility and robustness, y g p p y ,

esp. under non-blind detection– Typical watermark-to-noise (WNR) ratio: -20dB in blind

detection 0dB in non-blind detectiondetection, 0dB in non-blind detection

Choice of modulation schemes# of fingerprints

Orthogonal modulation jj uw

B

# of fingerprints= # of ortho. bases

(Binary) coded modulation

for or

i

iijj b1

uw

1,0ijb 1ijb


# of fingerprints >> # of orthorgonal bases

Assumptions & Goals for BonehAssumptions & Goals for Boneh--Shaw Codes (’95)Shaw Codes (’95)

Detectable vs. Undetectable 111 111 111000 000 111

Collusion Scheme: – For each bit position, colluders can manipulate theFor each bit position, colluders can manipulate the

detectable bit value to any possible state, even make it unreadable

M ki A i Marking Assumption– Users cannot change the state of an undetected bit without

rendering the object useless.g j

Goal:– Given the fingerprint string of a suspicious media, output


g p g p , pone colluder.

Coded Fingerprints by BonehCoded Fingerprints by Boneh--Shaw ConstructionShaw Construction

[I]Primitive

[II]Enlarge via

[IV]Permute within

[III]Compose withPrimitive

Code

1 1 1 A

n-1 Step-I

Enlarge viaRepetition

Permute withinEach Block

Compose withRandom Code

A B B iD CStep-III

1 1 1 A0 1 1 B0 0 1 C0 0 0 D

nA B B iD A C iiB A D iii: : : : :

DDC:

CAB:

C D B A C vii

111 111 111 A000 111 111 B000 000 111 C000 000 000 D

Step-II

111111111 000111111 000111111 000000 111000000 000[ ]A B B D C

111 111 111 A100 010 001 C

000 000 000 D

111 111 111 110 111 001 110 111 001 000 000 001010000 100[ ]Step-IV

Permute

Fingerprint code for User-i to insert into host datafingerprint


g p

Coded Fingerprinting: Prior Work and New IssuesCoded Fingerprinting: Prior Work and New Issues

Collusion-secure codes by Boneh and Shaw ’98– Targeted at generic data with “Marking assumptions”g g g p

~ an abstraction of collusion model

– Codes are too long to be reliably embedded & extracted (Su et al.) ~ millions bits for 1000 users millions bits for 1000 users

– Focus on tracing one of the colluders

New issues with multimedia– “Marking assumptions” may no longer hold …– Some code bits may become erroneously decoded due to strong

noise and/or inappropriate embeddingpp p g– Can choose appropriate embedding to prevent colluders from

arbitrarily changing the embedded fingerprint bits

W ll d ibl


Want to trace as many colluders as possible

Spreading + Combinatorial Coded FingerprintingSpreading + Combinatorial Coded Fingerprinting

Overall idea of embedded combinatorial fingerprinting– Explore unique issues associated with multimedia in fingerprint

encoding, embedding & detection– Use appropriate embedding to prevent arbitrary change on code

1st bit1st bit 2nd bit2nd bitB 1st bit1st bit 2nd bit2nd bit

......

B

iiijj b

1uw 1ijb

Build correlated fingerprints in two steps– Binary Anti-collusion fingerprint codes resist up to K colluders

any subset of up to K users share a unique set of code bits

– Use antipodal coded modulation to embed fingerprint codesi th l d t


via orthogonal spread spectrum sequences shared bits get sustained and used to identify colluders

1616--bit ACC for Detecting bit ACC for Detecting 3 Colluders Out of 203 Colluders Out of 20

User-1 ( -1,-1, -1, -1, 1, 1, 1, 1, …, 1 ) ( -1, 1, 1, 1, 1, 1, …, -1, 1, 1, 1 ) User-4

Collude by AveragingUniquely Identify User 1 & 4

Embed fingerprint via HVS-based spread spectrum embedding in block-DCT domain


Extracted fingerprint code ( -1, 0, 0, 0, 1, …, 0, 0, 0, 1, 1, 1 )

ACC Codes Under Averaging CollusionACC Codes Under Averaging Collusion

User-1 User-4 User-8

Averaging of multimedia domain leads to averaging in code-domain, and corresponds to AND operation after thresholding

Can distinguish colluded bits from sustained bits statistically with appropriate modulation and embedding and the sustained bits are unique with respect to


modulation and embedding, and the sustained bits are unique with respect to colluder set

AntiAnti--Collusion Fingerprint CodesCollusion Fingerprint Codes

Simplified assumption (can be relaxed by using soft detection)– Fingerprint codes follow logic-AND operation after collusiong p g p

K-resilient AND ACC code– A binary code C={c1, c2, …, cn} – The logical AND operation of any combination of up to K codevectors is

distinct from the AND of any other combinations of up to K codevectorsExample: {(1110), (1101), (1011), (0111)}

1111000

ACC code via combinatorial design– Balanced Incomplete Block Design (BIBD)

101010111001101111000

Si l E l

010110101100110011110CSimple Example

ACC code via (7,3,1) BIBD for handling up to 2 colluders among 7 users


1001011

Balanced Incomplete Block Design (BIBD)Balanced Incomplete Block Design (BIBD)

Construction example of (7,3,1) BIBD code– X={1,2,3,4,5,6,7}

1 2 3 4 5 6 7x x x{ , , , , , , }

– A={123, 145, 246, 167, 347, 257, 356}

( k ) BIBD i (k 1) ili t AND ACC

x x xx x x

x x x (v,k,)-BIBD is an (k-1)-resilient AND ACC

– Defined as a pair (X,A) X is a set of v points

x x xx x x

x x xf p

A is a collection of blocks of X, each with k points every pair of distinct points is in exactly blocks

vv2

x x x

– # codewords = # blocks

Code length for n=1000 users: O( n0.5 ) ~ dozens-to-hundreds bits6

kkvvn 2


– Shorter than prior art by Boneh-Shaw O( (log n)6) ~ millions bits

Colluder Detector Design: Two ApproachesColluder Detector Design: Two Approaches

Hard Detection:– Detect the bit values and then estimate colluders from these valuesDetect the bit values and then estimate colluders from these values– Uses the fact that the combination of codevectors uniquely identifies

colluders– Everyone is suspected as guilty and each ‘1’ bit narrows down set

Soft Detection – possible candidates for soft detection:– Sorting: Use the largest detection statistics to optimize likelihood

function to first determine bit values, then estimate colluder set.

– Likelihood-based sequential method: Iteratively update the likelihood function and directly identify the colluder set.


ACC Experiment with Gaussian SignalsACC Experiment with Gaussian Signals

– Higher threshold captures more colluders, but suspects more innocents– Soft decoding gives more accurate colluder identification than hard decoding– Joint decoding and colluder identification gives better performance than

separating the two steps


– Sequential colluder identification (Simp. AM) gives a good tradeoff between performance and computational complexity

Digital Fingerprinting Example for Tracing TraitorDigital Fingerprinting Example for Tracing Traitor

Our software toolbox demonstrates

llcollusion-resistant fingerprinting of an aerial image captured by UAV Predator overby UAV Predator over Pentagon


Recent Work: Exploiting ECC for FingerprintingRecent Work: Exploiting ECC for Fingerprinting

Examines Error Correcting Codes (ECC) based fingerprinting based ona joint coding and embedding framework– Symbol-based fingerprint codes are attractive for frame-based multimedia

(such as video and audio)– Identify the performance advantage and disadvantage of ECC based

fingerprinting compared with Orthogonal fingerprinting

Proposed two new joint coding and embedding techniquesP t d b t b ddi– Permuted subsegment embedding

– Group based joint coding and embedding (GRACE) fingerprinting

Main Results (see ICASSP’05 Monday afternoon session)Main Results (see ICASSP 05 Monday afternoon session)– Substantially improve collusion resistance of ECC based fingerprinting

and provide better trade-off b/w collusion resistance & efficient detectionDemonstrates advantages by joint coding and embedding in fingerprint


– Demonstrates advantages by joint coding and embedding in fingerprint design

Similarity between Collusion and MU Comm.Similarity between Collusion and MU Comm.

The Fingerprint Collusion Problem is similar to Multiuser Communication– The colluded signal is simply the host signal plus a mixture of watermarksg p y g p

Collusion Fingerprint Problem Synchronous CDMA Channel

colluderaisuserjthif1}1,0{1

wy

j

n

jjjc d

K

1,1

)()()(1

b

tntsbAty

k

n

kkkk

F d i ti f CDMA h ld h

tsfingerprinj wj

sequencessignature)( tsk

k

For good communication performance: CDMA sequences should have minimum interference between each other. Low Cross-Correlation is Good!

The similarity between Collusion and MU Comm. suggests that good CDMA ld b d fi i t !


sequences would be good fingerprints!

ACC Built from Interference AvoidanceACC Built from Interference Avoidance

How to assign M fingerprints in N dimensions to facilitate colluder detection?

M ≤ N: assign orthogonal fingerprints because they are uncorrelated

M > N: the fingerprints are correlated. How do we find the least correlated set S of size N×M?– Minimize Total Squared Correlation (TSC):

n

i

n

jj

TiTSC

1 1

2)( ss– Welch Bound: TSC is lower bounded by M2/N

– WBE sequence set:N

MTSCNMT

2

NISS

i j1 1

– WBE sequence set is known to be optimal in terms of user capacity in synchronous code-division multiple access (CDMA)

NN


y p ( )

– One approach to get WBE sequence set: Eigen-algorithm

Detection of WBE FingerprintsDetection of WBE Fingerprints

Vector collusion model– Project the signals, noise, and watermarks into the watermark spacej g , , p

nSK

T T: sufficient statistics for detection

S: code matrix: the coll sion indicator ector

– Goal: estimate from the sufficient statistics T: the collusion indicator vector

Maximum likelihood detection

2||||minargˆ ST

– Minimize the distance between the observed vector T and

||||minarg

SK

T

SK


– Integer least square problem… Sphere decoding may be used!K

WBEWBE--ACC Performance Compared with BIBD ACCACC Performance Compared with BIBD ACC

Spreading factor N=10000 (i.e. length of basis). Gaussian i.i.d noise 20 users, 16 dimensions, 3 colluders20 users, 16 dimensions, 3 colluders WBE fingerprints scheme shows improvement in both Pd and Pfa

P b bilit f t d t ti P b bili f f l i1

0 2

0.25ACC-SortingACC-AM WBE S h

Probability of correct detection Pd Probability of false accusation Pfa

0.8

0.9

p d

0 1

0.15

0.2

p fa

WBE-Sphere

0.7 ACC-SortingACC-AM WBE-Sphere

0.05

0.1


-25 -23 -21 -19 -17 -150.6

WNR-25 -23 -21 -19 -17 -150

WNR

A Unified FrameworkA Unified Framework

Unified view of fingerprint construction using code modulation

O th l fi i ti B1. Orthogonal fingerprinting:

– Equivalent code matrix B = identity matrix

2. Coded fingerprinting: a compact way to represent users

i

iijj ubw1

2. Coded fingerprinting: a compact way to represent users– Boneh-Shaw’s and error correction code based construction – Combinatorial based ACC code: take advantage of shared bits

3. General correlated fingerprints– Allow real-valued codes such as in group-oriented fingerprinting

Key: how to strategically introduce correlation and separationbetween codes to capture individual and multiple colluders

Ref: M Wu W Trappe Z Wang and K J R Liu: "Collusion Resistant Fingerprinting for


Ref: M. Wu, W. Trappe, Z. Wang, and K.J.R. Liu: Collusion Resistant Fingerprinting for Multimedia", IEEE Signal Processing Magazine, Special Issue on Digital Rights Management, pp.15-27, March 2004.

Conclusions from AntiConclusions from Anti--Collusion FP StudiesCollusion FP Studies

Important to design anti-collusion fingerprint for multimediaC ll i i t ff ti tt k i t fi i ti– Collusion is a cost-effective attack against fingerprinting

– Anti-collusion fingerprints can allow us trace traitor and deter unauthorized information leakage

Good news– We can tolerate about a few dozens colluders– We can accommodate more users through use of ACC

ChallengesO fi d h ll d t i t th t– One may find enough colluders to circumvent the system

– More research will lead to better designs and increase the collusion resistance


Data Hiding in Curves forData Hiding in Curves forg fg f

Fingerprinting Map DocumentFingerprinting Map Document


Interested in Curves?Interested in Curves?

Curve in our everyday life– Appear in a number of documents, e.g. maps and art/engr designpp , g p g g– Increasing popular with writings/drawings from pen-based input

devices (such as Tablet PC)

Storage formats of curves– Raster format

Represent as an image with (binary) intensity Represent as an image with (binary) intensity information of a 2D array of pixels

– Vector format A piecewise linear approximation with such

geometric primitives as vertex points and lines

– Conversion between different formats


– Conversion between different formats

Protecting MapsProtecting Maps

Map represents important geospatial information – Ubiquitous in military, intelligence, and commercial operations.q y g p

Traditional way to protecting maps: add intentional errors– Wrongly spell the names; bend roads in a wrong way; add fake lakes/hills

Disadvantages of traditional protection– Substantially alter the geospatial

meanings conveyed by a mapmeanings conveyed by a map– For fingerprinting use, a small amount

of intentional errors can be quite easily identified and removed from severalidentified and removed from several fingerprinted copies of a map

Need a modern protection


– Less intrusive; resist collusion & format conversion

Digital Fingerprinting of MapsDigital Fingerprinting of Maps

Fingerprinting for tracing traitors– Different fingerprints are embedded in different copies beforeDifferent fingerprints are embedded in different copies before

distributing them to a number of recipients– For discouraging illegal distribution and tracing traitors in case of

information leakinginformation leaking

Challenge of fingerprinting mapsThe fingerprint must be embedded in a visually non intrusive way– The fingerprint must be embedded in a visually non-intrusive way

– Embedded fingerprints should not be obliterated by attacks multi-user collusion attacks

i di i i li d l i geometric distortions, e.g., rotation, scaling, and translation format & D/A-A/D conversion, such as printing-scanning

– Good news: the original unmarked version is usually available to


Good news: the original unmarked version is usually available to detector

Existing Work on Watermarking MapsExisting Work on Watermarking Maps

Watermarking map images: text-based geometric normalization

Watermarking vector maps Watermarking vector maps– Fourier descriptors of polygonal lines: have visible distortions

– Spectral analysis of mesh models Complex and cannot be easily applied to maps beyond urban areas

– 3-D watermarking works using mesh and surface modeling Provide enlightening analogy for 2-D work Very limited work on robust approach and validation results

for 2-D, esp. on collusion-resistant fingerprinting & format conversion

Watermarking general binary images Watermarking general binary images– Often rely on minor manipulation of pixels– Fragile => difficult to deal with RST issues and survive D/A-A/D

Of d f h i i d i


– Often used for authentication and annotation

New Approach to Fingerprinting MapsNew Approach to Fingerprinting Maps

Select fingerprinting domain and embedding method to address the imperceptibility and robustness requirement

Feature domain: coordinates of control points in the B-spline representation of curvesrepresentation of curves– These form salient features representing the shape of the curve

Embedding: spread spectrum embedding & correlation detection Embedding: spread spectrum embedding & correlation detection– Good tradeoff between imperceptibility/robustness (esp. nonblind detection)– The general collusion resistance have been studied for independent and

d d fi i ti i d t b ddicoded fingerprinting using spread spectrum embedding orthogonal fingerprinting is good for a moderate size of user group

Resist printing scanning & D/A A/D: b hi h i i i t ti


Resist printing-scanning & D/A-A/D: by high precision registration

BB--splines Representation of Curvessplines Representation of Curves

[ ],

0( ) ( )

nB

i i ki

t B t

p c0i

BB--spline spline Basis FunctionBasis FunctionControl Control

PointsPoints

Curve Curve PointsPoints

B-splines: Piecewise polynomial functions giving local approximation for a curve w/ small# parameters B-spline basis functions are defined recursively

)()()()( 111 tBtttBtt kikikii

otherwisettt

tB iii ,0

,1)( 1

1,

1,)()()()(

)(1

1,1

1

1,,

ktt

tBtttt

tBtttB

iki

kiki

iki

kiiki

Control points


Provide weights on the basis function to locally control the shape of the curve rendered

Control Points of BControl Points of B--spline Representationspline Representation

Least-square approach to estimate control points – Given a set of properly chosen samples on the curve, control p p y p ,

points can be estimated using the least squares technique

Properties of control points Properties of control points– Serve as a compact set of salient

features for curves => not affectedby noisy scanning boundaries

– Provide local control on shape– Invariant to affine transformations– Invariant to affine transformations


Data EmbeddingData Embedding

Spread spectrum additive embedding– Use noise-like sequences as digital fingerprints– Represent different users using (approximately) orthogonal sequences

May generate independent seq. via pseudorandom number generator

– A scaled version of the fingerprint sequence is added to the coordinates ofA scaled version of the fingerprint sequence is added to the coordinates of the set of control points

A fingerprinted curve can be obtained by synthesizing from the fi i t d t l i tfingerprinted control points

Fingerprint SequenceEmbedded

' c c w

Original Curve

MarkedControl Points

MarkedCurve

OriginalControl Points


'x x x

y y y

c c wc c w

Data DetectionData Detection

Register with the original unmarked curve Extract control points of the test curve and compute the changes to Extract control points of the test curve and compute the changes to

arrive at an estimated fingerprint sequence

OriginalControl Points

Test CurveRegistered

TestControl Points

Control Points

SpecificUser

Correlation-based Threshold Testing

Fingerprint SequenceEstimated

( ) /

Correlation-based statistics and threshold testing

( ) /( ) /

x x x

y y y

w c cw c c

23)1(2

11log

n

Z : correlation coefficient

n: number of control points


Follow approx. unit-variance Gaussian distribution (threshold: around 3~6)

Fidelity and RobustnessFidelity and Robustness

Where does the robustness come from?– Control points represent salient features of curves: analogous to p p g

perceptually significant spectral components in images– Spread spectrum embedding

S li f t Scaling factor α– Larger α larger distortion, more robustness

determined by application’s requirement; we used 0.5 in our tests

The number of control points– Determine the length of the fingerprint sequence

Use an appropriate number of control points to fit the original curve well– Use an appropriate number of control points to fit the original curve well Too few: details of the curve lost; Too many: over-fitting The number of control points is about 5~8% of the total number of

curve pixels in our tests


curve pixels in our tests– Place knots and control points according to the curvature

Fingerprinting Simple CurvesFingerprinting Simple Curves

Original Curve

(captured by TabletPC)

Detection Statistics

Typical threshold is 3~6 for f l l f 10 3 10 9

Fingerprinted Curve

(100 control points)


( p y ) false alarm of 10-3 ~ 10-9( p )

Fingerprinting Topographic MapFingerprinting Topographic Map

1100x1100 Original Map Fingerprinted Map

9 curves that are long enough are marked


9 curves that are long enough are markedTotally, there are 1331 control points used to carry the fingerprint

Fidelity DemonstrationFidelity Demonstration

original and fingerprinted curves are overlaidoriginal: blue; marked: red


original: blue; marked: red

Resistant to Collusion AttacksResistant to Collusion Attacks

2-User Interleaving Attack5-User Averaging Attack

. . .


Resistant to PrintingResistant to Printing--andand--scanningscanning

Original Curve

Detection Statistics

( 100 control points;l i t ti )manual registration )


Fingerprinted curve after Printing-and-Scanning

Automated HighAutomated High--Precision RegistrationPrecision Registration

Inherent non-uniqueness of B-spline control points– A curve can be well approximated by different sets of control points

Iterative Alignment-Minimization (IAM) algorithm– Formulate jointly undoing geometric transform

d i i h i i l f l iand retrieving the original set of control points as an optimization problem

Test Control Points

Step 1Step 1 Step 2Step 2

Initial Estimationof Sample Points

Test RasterCurve

EstimatedSample Points

Super Curve Alignment

Step 1Step 1 Step 2Step 2

Original Ali d

Projection


OriginalSample Points Nearest Neighbor

RuleAligned

Raster Curve

Step 3Step 3

An Example Using IAM: Aligning Sample PointsAn Example Using IAM: Aligning Sample Points

Original and test curve Test sample points Test sample points


Original and test curve Test sample points after 1 iteration

Test sample points after 15 iterations

An Example Using IAM: (cont’d)An Example Using IAM: (cont’d)

iterations iterations

Transform Parameters Detection Statistics


Resilience to Collusion plus PrintingResilience to Collusion plus Printing--andand--ScanningScanning

Detection statistic for one-user print-scan is 24.93 on the 9-curve map Four user averaging collusion + printing-scanningg g p g g

Print using a HP laser printer Scan back by a Canon scanner with 360dpi resolution

@


@

Summary and Conclusion on Curve FingerprintingSummary and Conclusion on Curve Fingerprinting

Important to design anti-collusion fingerprint for multimedia– Collusion is a cost-effective attack against fingerprintingCollusion is a cost-effective attack against fingerprinting– Anti-collusion fingerprint can allow us trace traitor and deter

unauthorized information leakage

New data hiding technique for curves – Parameterize curves using the B-spline model and embed data

through spread-spectrum based perturbation on control pointsthrough spread spectrum based perturbation on control points– High fidelity in embedding (w.r.t. the original curve)– Resilient to collusion and printing-and-scanning attack

Potential applications for protecting maps and drawings– Fingerprinting for traitor tracing


Secure and Efficient Multicast Distribution ofSecure and Efficient Multicast Distribution ofSecure and Efficient Multicast Distribution of Secure and Efficient Multicast Distribution of Fingerprinted VideoFingerprinted Video


MotivationMotivation

Streaming applications with traitor tracing requirement Distribution of fingerprinted multimedia over networks Distribution of fingerprinted multimedia over networks

– Unicast: not scalable– Traditional multicast technology: the same data to multiple users

Goal of the streaming service provider: – Reduce the communication cost– Accommodate as many users as possible– Protect the secrecy of the multimedia content as well as the embedded

fingerprintsfingerprints

Design secure fingerprint multicast schemes!Design secure fingerprint multicast schemes!


Previous WorkPrevious Work

Watercast (Brown et al. 99)– Trusted intermediaries embed their unique IDs before forwarding the packets

through the network

WHIM (Judge et al. 00)Trusted routers forward differently fingerprinted packets to different users– Trusted routers forward differently fingerprinted packets to different users

Selective Encryption (Wu et al. 97)– Fingerprints were embedded in DC coeff. of I frames– Selectively encrypt and unicast part of the MPEG stream to each user

Multicast Watermark Protocol (Chu et al. 02)A li d h Sh fi i d d i f b f b i– Applied Boneh-Shaw fingerprint code design on a frame by frame basis

Joint fingerprinting and decryption (Kundur et. al. 04)– Fingerprint design similar to Boneh-Shaw, on a pixel by pixel basis


Fingerprint design similar to Boneh Shaw, on a pixel by pixel basis

New Issues on Fingerprint MulticastNew Issues on Fingerprint Multicast

Prior work: – Only applicable to trace a few colludersOnly applicable to trace a few colluders– Adjust fingerprint design to fit the communication framework

New issues: – Video streaming applications: a large number of users, and

t ti ll l b f ll dpotentially a large number of colluders– Consider applications requiring strong traitor tracing capability – Utilize the multicast technology to fit the fingerprint designUtilize the multicast technology to fit the fingerprint design


A General Fingerprint Multicast SchemeA General Fingerprint Multicast Scheme

Spread spectrum embedding– Resistant to many attacks y– Not all coefficients are embeddable due to perceptual constraints

An example: percentage of embeddable coefficientsMi A i 18% C h 28% Fl 34%Miss America: 18%, Carphone: 28%, Flower: 34%

– A non-embeddable coefficient has the same value in all copies.

General fingerprint multicast scheme:General fingerprint multicast scheme:– Reduce the comm. cost in distributing the non-embeddable coefficients

Shared non-embeddable coefficientsmulticasti l fi i d ffi i i Uniquely fingerprinted coefficients unicast

– Can be used with most spread spectrum embedding based fingerprinting systems


A General Fingerprint Multicast Scheme (cont’d)A General Fingerprint Multicast Scheme (cont’d)

AliceAlice

embedembed

embedembedBob

Bob

The Lord of the Rings

Chris

embedembed

Chris


MPEGMPEG--2 Based General Fingerprint Multicast Scheme2 Based General Fingerprint Multicast Scheme

M ti P iti f b dd bl ffi i t Motion vectors, quan. factors, other side info. N

VLC

Positions of embeddable coefficients Multicast to all users

mK

Encrypt

VLC-1

Coded DCT

Q-1 VLCQEmbeddable?

NY

Unicast to user u(i)

( )iK

mKEncrypt

User u(i)’s fingerprint

coeff.K

for user u(i)

The Fingerprint Embedding and Distribution Process for Video on Demand Applications at


Process for Video on Demand Applications at the Server’s Side

MPEGMPEG--2 Based General Fingerprint Multicast Scheme2 Based General Fingerprint Multicast Scheme

Multicasted mKVLC-1 Q-1

VLC-1 Q-1

Multicasted DCT coeff.

Decrypt

Decrypt

mKReconstruct

the 8*8bl k

IDCT

0C Q

Unicasted DCT coeff.

yp( )iK

DCT block

Positions of b dd bl

Motion compensation

embeddable coefficients Motion vectors

The Decoding Process at the Receiver’s Side


Bandwidth EfficiencyBandwidth Efficiency

Measurement of bandwidth efficiency:– General Internet Routing: hop-based link usageGeneral Internet Routing: hop-based link usage

– Compare the communication cost of two schemes: Pure unicast scheme: General fingerprint multicast scheme:

puCfmC

– Communication cost ratio: The smaller the communication cost ratio, the more efficient

pufmfm CC /

the general fingerprint multicast scheme.


Tested Video SequencesTested Video Sequences

Miss America Carphone Flower

R=1.3 bpp


Communication Cost RatioCommunication Cost Ratio


M users; Rate =1.3 bpp.

Joint Fingerprint Design and DistributionJoint Fingerprint Design and Distribution

General fingerprint multicast scheme:– Utilizes the perceptual constraints on spread spectrum embeddingUtilizes the perceptual constraints on spread spectrum embedding– Multicast the non-embeddable coefficients to all users– Can be used with most spread spectrum embedding based

fingerprinting systems

A joint fingerprint design and distribution scheme– Explores the special structure of the fingerprint designExplores the special structure of the fingerprint design– Also multicast some fingerprinted coefficients that are shared by a

subgroup of users to them


– Further improve the bandwidth efficiency

GroupGroup--Oriented Fingerprint DesignOriented Fingerprint Design

Motivation:– Some users are more likely to collude with each other than theSome users are more likely to collude with each other than the

others due to geographical or social reasons– Tree structure to explore the hierarchical relationship between

users

Group oriented fingerprint design1 2a– Group users who are more

likely to collude witheach other together

1a a

1,1a 1,2a 2,1a 2,2aeach other together

– More robust against collusion attacks

1,1,1a 1,1,2a 1,1,3a 1,2,1a 1,2,2a 1,2,3a 2,1,1a 2,1,2a 2,1,3a 2,2,1a 2,2,2a 2,2,3a(1)u (2)u (3)u (4)u (5)u (6)u (7)u (8)u (9)u (10)u (11)u (12)u


1,1U 1,2U 2,2U2,1UBy Wang et. al. ’04

Joint Fingerprint Design and DistributionJoint Fingerprint Design and DistributionUser u(1) X1

(1) X2 (1) X3

(1)

S S S

=

1a 2a

S1 S2 S3

a11 a2

1 a31

+

1,1a 1,2a 2,1a 2,2a

1,1,1a 1,1,2a 1,1,3a 1,2,1a 1,2,2a 1,2,3a 2,1,1a 2,1,2a 2,1,3a 2,2,1a 2,2,2a 2,2,3a

a21,1 a3

1,1

a1,1,1

a a a a a a a a a a a a(1)u (2)u (3)u (4)u (5)u (6)u (7)u (8)u (9)u (10)u (11)u (12)u

User u(2) X1 (2) X2

(2) X3 (2)

S1 S2 S3

=

Host Signal S

S1 S2 S3

a11 a2

1

a21,1

a31

a31,1

+


1 2 3

a1,1,2

Joint Fingerprint Design and Distribution (cont’d)Joint Fingerprint Design and Distribution (cont’d)

User u(1)

S1+a1 S2+a1+a1,1 S3+a1+a1,1+a1,1,1

Multicast to all users

Multicast to u(1)~u(6)

Unicast to u(1)

Unicast to u(2)Multicast to u(1)~u(3)

1a 2a

1,1a 1,2a 2,1a 2,2a

S1+a1 S2+a1+a1,1 S3+a1+a1,1+a1,1,2

1,1,1a 1,1,2a 1,1,3a 1,2,1a 1,2,2a 1,2,3a 2,1,1a 2,1,2a 2,1,3a 2,2,1a 2,2,2a 2,2,3a(1)u (2)u (3)u (4)u (5)u (6)u (7)u (8)u (9)u (10)u (11)u (12)u

User u(2)

1 2 3u u u u u u u u u u u u


Secure Fingerprint Multicast: PerformanceSecure Fingerprint Multicast: PerformancePure Unicast General Fingerprint

MulticastJoint Fingerprint Design

and Distribution

Fi i d i A M d T b d fi iFingerprint design Any Most spread spectrum embedding based

Tree-based fingerprint design

Comm. cost ratio 1 16% ~ 52% 13% ~ 39%

Computation complexity

MG: 0RL: 1

MG: 1RL: 2

MG: 13 ~ 125RL : 4 ~ 5

Collusion resistance Same under fairness constraintsCollusion resistance Same under fairness constraintsSuggested

ApplicationsSmall

number of users

Large number of users, sequences with fewer

embeddable coeff.

Large number of users, sequences with much

more embeddable coeff. (e.g., miss america) (e.g., flower)

–Total number of users: 1000 ~ 10000.


–MG: total number of multicast groups in the system–RL: number of channels that each receiver has to listen to

Summary and Future Work on Fingerprint MulticastSummary and Future Work on Fingerprint Multicast

Summary– Reduce the bandwidth requirement by 48% ~ 87%Reduce the bandwidth requirement by 48% ~ 87% – Protect the secrecy of the video content as well as that of the

fingerprints– Preserve the collusion resistance of the embedded fingerprints

k Future work– Other important aspects of the general fingerprint multicast:

Quality of service management Quality of service management Fingerprint multicast over heterogeneous networks to users

with different processing capability


Tutorial Summary and Recent PublicationsTutorial Summary and Recent Publications

Understanding the linear and nonlinear collusions:H. Zhao, M. Wu, Z. Wang, and K.J.R. Liu: "Nonlinear Collusion Attacks on Independent Multimedia Fingerprints," IEEE Trans. on Image Proc., to appear in May 2005.Multimedia Fingerprints, IEEE Trans. on Image Proc., to appear in May 2005.

Analyzing the limit on collusion resistance by orthogonal fingerprinting:Z. Wang, M. Wu, H. Zhao, W. Trappe, and K.J.R. Liu: "Collusion Resistance of Multimedia Fingerprinting Using Orthogonal Modulation", IEEE Trans. on Image Proc., to appear in June 2005.

Leveraging prior info. on collusion pattern via group fingerprinting:Z. Wang, M. Wu, W. Trappe, and K.J.R. Liu: "Group-Oriented Fingerprinting for Multimedia Forensics" EURASIP Journal on Applied Signal Processing Special Issue on Multimedia Forensics , EURASIP Journal on Applied Signal Processing, Special Issue on Multimedia Security and Rights Management, Nov. 2004.

Proposing a new framework promoting the joint coding, embedding, and detection, with a demonstration through new combinatorial based ACC, g

W. Trappe, M. Wu, Z. Wang, K.J.R. Liu, “Anti-Collusion Fingerprinting for Multimedia,” IEEE Trans. on Sig. Proc., Special issue on Data Hiding in Digital Media & Secure Content Delivery, 51 (4), pp.1069-1087, April 2003.

Explorations on ECC fingerprinting map/curve FP & FP multicast


Explorations on ECC fingerprinting, map/curve FP, & FP multicastSee papers in ICASSP 2005, & upcoming papers in Trans. Sig/Image Proc.

Coming soon 200Fall 2005

Chapter 1 IntroductionChapter 2 Preliminaries on Data EmbeddingChapter 3 Collusion AttacksChapter 4 Orthogonal Fingerprinting and Collusion ResistanceChapter 4 Orthogonal Fingerprinting and Collusion ResistanceChapter 5 Group-Oriented FingerprintingChapter 6 Anti-Collusion Coded (ACC) FingerprintingChapter 7 Secure Fingerprint Multicast for Video StreamingChapter 8 Fingerprinting Curves

To be published by Hindawi Publishing in Fall 2005


To be published by Hindawi Publishing in Fall 2005EURASIP Book Series on Signal Processing and Communications

Contact Information of Tutorial Instructors

K.J. Ray Liu [email protected]://www.isr.umd.edu/Labs/CSPL/kjrliu/kjrliu.html

Wade Trappe [email protected] http://www.winlab.rutgers.edu/~trappe/

j @ bZ. Jane Wang [email protected]://www.ece.ubc.ca/~zjanew/

Min Wu minwu@eng umd eduMin Wu [email protected]://www.ece.umd.edu/~minwu/


Documents

Digital Fingerprinting for g g p g f g g p g f Multimedia Forensics