Upload
sharlene-anthony
View
247
Download
0
Embed Size (px)
Citation preview
Crypto Final Presentation
B89902026 林敬倫B89902043 李佳蓉B89902091 王姵瑾B89902102 周振平
Watermarking Maps: Hiding Information in Structured Data
Sanjeev Khanna Francis Zane
Accepted by SODA2000
1. Introduction
What’s Watermarking?Where one embeds hidden info into the
data which encodes ownership and copyright.
Applied to:Image, Video, Audio, etc.
Owner: compiles accurate map dataProvider: provide end-user access.
Map Watermarking Problem Schemes:
Which allow the owner to distribute and identify many different copies( marked copy )
New nodes or edges are not allowed. Change small length are not allowed. Each marked copy must involve only
a slight distortion.
Suspect! An owner should be able to
accurately determine the unique provider from that copy using only public accessible info.
The owner access the provider’s data as an end-user.
Goal Maximize the number of copies the
owner can distribute under these constraints.
If the owner encounter a suspect copy.Then he has complete access to the data for this copy which he can determine the guilty party.
2. Preliminaries How to distortion measured? What type of information does the
provider give in response to queries?
Is the provider free to answer as he chooses in order to evade detection?
Mesaures of Distortion Additive distortion of d Multiplicative distortion of d With respect to a graph G Additive and multiplicative
distortion for a path P. G’ is a d-distortion of G if it has
distortion d with respect to G.
Queries and Responses Edge Queries
Gives the owner complete access to the copy of the provider.
Distance Queries Route Queries
Only a path. Might be using by a cheating provider.
Adversarial vs. Nonadversarial A provider add additional distortion
to evade detection No-adversarial:
If provider answers all queries correctly
Otherwise it’s adversarial No effective scheme against large
distortion or has fairly accurate knowledge.
Marking Schemes M is marking algorithm
Map each r to a copy of the map Gr
Such that Gr is a d-distortion of G.
D is detection algorithm ( detector ) Answers to its queries to recover the
provider r
3. Overview of Results
Nonadversarial Edge and Distance
THEOREM 3.1.
For the nonadversarial edge model, there are marking schemes that encode Ω(m 1/2-ε) bits with additive distortion and Ω(m) bits with multiplicative distortion. For the nonadversarial distance model,there are marking schemes that encode Ω(n 1/2-ε) bits with additive distortion and Ω(n) bits with multiplicative distortion. In each case, the additive distortion is O(1/ε) while the multiplicative distortion is (1 + o(1)).
Adversarial Edge and Distance Models
THEOREM 3.2.
For the adversarial edge model, there is a marking scheme that encodes Ω(m 1/2-ε) bits while for the adversarial distance model, there is a marking scheme that encodes Ω(n 1/2-ε) bits.
Nonadversarial Route Models In route models, problem becomes
significantly more complex. Ask the graph to satisfy some
“good property” Tradeoff between requirement of
the graph and the bits we can encode
Nonadversarial Route Models THEOREM 3.3.
For the nonadversarial route model, there is a marking scheme that encodes Ω(m 1/2-ε) bits with small multiplicative distortion when the underlying graph is 2-edge connected and its length function is nearly-uniform.
4. Nonadversarial Case
Edge Model: Additive Distortion
Goal: generate many distinct graphs while introducing only small distortion
Let P P(G) be a shortest path in G with the largest number of edges, and let L denote the number of edges on P. Let u0, ul, ..., uL be the nodes along the path P.
Edge Model: Additive Distortion
We discuss two cases: L L0
L L0 ( L0 will be defined later)
With L L0:With our marking scheme,
For any pair x , y of nodes,
| dG’(x, y) - dG’(x, y) | < 2.
Edge Model: Additive Distortion
With L L0
By results from the two cases, THEOREM 4.1.
There is a marking scheme that encodes Ω(m1/2-ε) bits of information with only an additive distortion of 1/εfor any 0 < ε < 1/2.
Edge Model: Multiplicative Distortion
THEOREM 4.2. There is a marking scheme that gives
0 (m) bits of information with only a (1 + o(1)) multiplicative distortion.
Distance Model: Additive Distortion
Again, discuss two cases L L0, L L0
With L L0,
For any pair x,y of nodes,| dG’(x,y) – dG’(x,y) | 3
With L L0,Let V' V be obtained by randomly picking each node with probability p = 1/(Lon2ε). Then V' is an ε-good node marking set of size Ω(pn) with probability at least 1/3
Distance Model: Additive Distortion
THEOREM 4.3. There is a marking scheme that gives O(n1/2-ε) bits of information with only an additive distortion of 1/ε for any 0 < ε < 1/2.
5. Adversarial Case
Adversarial Case
Model and Assumptions Assumption 1 (Bounded Distortion
Assumption) : For all (u, v) E V × V, |A(u, v) - dG(u, v)| <= d', where d' is an absolute constant.
DEFINITION 5.1. W V × V is low-bias with respect to S {0, +1, - 1} E if for all (u, v) Ε W,
| Δ(u,v)| <= 1 & for all z E {0, + l , - 1} , Pr[Δ(u, v) = z] <= ½
δ E S
Adversarial Case
Adversarial Case DEFINITION 5.2. S {0 , + 1 , - 1} E is
(γ, ρ)-unpredictable if for any W V X V, such that W is low-bias with respect to S and |W| = ω(1), any strategy AGδ available to the adversary satisfies
Pr [ Σ [AGδ (u,v)= Δ(u,v) > (1/2 + γ )|W| ] < p δ E S (u,v) E W
Adversarial Case
Assumption 2 (Limited Knowledge Assumption) : For any S {0, + 1 , - 1}E such that |S| = ω(1), S is (γ, ρ)-unpredictable.
Framework Marking Algorithm: For each provider r,
we chose a random vector Br E {+1,-1} L. From this vector, we obtain a vector D such that D(i) = a2 if B'(i) = +1 and D(i) = al , otherwise. Now use D to construct δr as guaranteed by the framework conditions and output the graph Gδ, with length function lδr, = l + δr.
Framework Detection Algorithm: Given access to a suspect map, we compute an implied L-dimensional vector Z,
defined by Z(i) = A(ui, vi). Let X(i) = dG(ui, vi) and define amid = (a1+a2)/2, adiff = (a2 – al)/2
For each provider r, we then compute a similarity measure
sim (Br, Z) = 1/ adiff Br • (Z - ( X + amid •1)) Choosing a threshold parameter t = 0.1, if sim(Br, Z)
>= tL, then we say that the provider r is responsible for the suspect copy.
Analysis False Positives
物枉 Show that the probability that an individual
suspect provider generates a false positive is small, even for an adversary with access to the original map.
False Negatives 物縱 Show that the probability of a false
negative(a guilty party evading detection) is also low.
False Positives Y(i) = X(i) + amid + adiff B(i) Assume for all i, |Z(i) - X(i)| <= d'. Using Chernoff bounds, we can show that
PROPOSITION 5.1. Given any valid Z, Pr[sim (Br, Z) > tL] <= e-q^2t^2L/2 when B is generated randomly independent of Z and q = adiff/(d' + amid).
COROLLARY 5.1 If L = Ω(logK), then the probability of a false positive error by the detector is o(1).
False Negatives PROPOSITION 5.2. If sim (B, Z) < tL, then
B • ( Z - Y) < - adiff (1 - t)L. PROPOSITION 5.3. Let 0 < c < d' + 1 be a
constant. Given the vector Y and a vector Z such that 1. B • ( Z - Y ) < - cL, and 2. For all i, |Z Ω(i) - Y(i)| <= d' + 1, there is an algorithm which produces a vector C such that [C(i) = B(i)] >= L/2 + cL/4(d' + 1) with probability at that 1 – e –Ω(L).
False Negatives LEMMA 5.1 If γ < 1 / 9(d'+l) and p >=
e -o(L), then the probability of a false negative by the detector is at most 2p.
Proof by contradiction!
Analysis THEOREM 5.1. Given a scheme
consistent with the framework, γ < 1 / 9(d'+l) , and p >= e -o(L), O(L) bits can be encoded such that the probability of error by the detector is at most max{2p, o(1)}.
Distance Model