is caused by definition of a template and a distance on
strings.
Assumption 2. There exist g : Σ
∗
×Σ
∗
→ N and c ∈ N
such that {b| g(t
i
, b) ≤ c, b ∈ Σ
∗
} = B
i
for any 1 ≤ i ≤
|U|.
2.2 Problems
The problem we consider is a spoofing which is
caused by leaked biological information. In fact, in
some practical systems, it is possible to make a fake
or artificial living-body from a piece of biological in-
formation (Matsumoto et al., 2002). Therefore, we as-
sume the following in terms of the model introduced
in the previous subsection.
Assumption 3. For any 1 ≤ i ≤ |U|, a single b ∈ B
i
enable to make u such that f(u, r) ∈ B
i
for r ∈ R.
Some cases of leakage of biological information
caused by man-made factor (such as, carelessness of
a verifier or a cheating verifier) are out of the scope
of cryptographic technologies. On the assumption of
the secure path by suitable cryptographic technolo-
gies, we focus on the following cases of the leakage:
• leakage of a template at the verifier,
• leakage of a piece of biological information at the
scanner.
In usual systems, a template is obtained by a rea-
sonable feature-extraction based on biology from
scanned biological information or is exactly the infor-
mation. In this situation, biological information can
be estimated from a leaked template and it enables a
spoofing as the user of the template.
On our model, the naive method to decide a tem-
plate is expressed by the condition that any element
in B
i
can be t
i
. Moreover, a straightforward feature-
extraction enables an estimation of the definition of
the distance, that is, we should assume g to be open.
Therefore, the essential point of the former case of the
leakage is that an element of B
i
can be estimated from
t
i
by Assumption 2 even if B
i
cannot be obtained ex-
actly. The latter case is exactly the leakage of b ∈ B
i
.
Thus, by Assumption 3 these cases enable the spoof-
ing.
3 SOLUTIONS
To solve the problems in the previous section, we pro-
pose solutions by modifying biological information.
The modification is expressed on the proposed model
as a function from a string to a string with some prop-
erties. Moreover,we consider the entity which should
apply the function to biological information.
3.1 Leakage of Template at Verifier
The problem of a spoofing by a leakage of a tem-
plate from the verifier is expressed on the proposed
model as that an element of B
i
can be estimated from
t
i
. In conclusion, this problem is solved by applying
a generalized idea of “cancelable biometrics (Ratha
et al., 2001)”, although the original idea is proposed
to enable changing a template rather than to prevent
a spoofing by a template. In fact, the results of this
subsection are obtained by interpreting the argument
in (Ratha et al., 2001) into our model.
We consider to prevent a spoofing using t
i
by ap-
plying a function φ : Σ
∗
→ Σ
∗
to biological informa-
tion. Let t
i
= φ(b) for a b ∈ B
i
. The prover (who has
a living body) u
i
submits φ(b
′
) for b
′
∈ B
i
as w to the
verifier. Then, on Assumption 1 and 2, the condition
for realizing identification is described as the follow-
ing property of φ.
Condition 1. There exists g
′
such that, for any p, q ∈
Σ
∗
, g
′
(φ(p), φ(q)) ≤ c if and only if g(p, q) ≤ c.
If we consider to add a step for applying φ into
Protocol 1, the possibility is only between the step (2)
and (3). Therefore, we assume that the scanner has a
suitable function for it, that is, the scanner is redefined
to be another function f ◦ φ and whether f (u, r) ∈ B
i
is examined by φ(f(u, r)) on the previous condition.
Protocol 2. (1) The prover puts u ∈ U on the scanner;
(2) the scanner computes φ( f(u, r)) for an r ∈ R;
(3) the scanner sends φ( f(u, r)) as w to the verifier;
(4) the verifier regards the prover as u
i
∈ U if and only
if φ
−1
(w) ∈ B
i
.
Now, we do not assume any confidentiality of g
′
for preventing the spoofing. Then, a spoofing using t
i
can be prevented if φ has the following property.
Condition 2. For any p ∈ Σ
∗
, it is difficult to find q
such that p = φ(q) for p.
On the previous condition, the verifier does not al-
ways have the result of φ
−1
(w) at the step (4) in Pro-
tocol 2. Formally, we have to refer the idea of “com-
putational indistinguishability (Goldreich, 2001)” for
the definition of the word “difficult”. However, in
some practical systems the properties of Condition 1
and 2 are not required strictly. The former guaran-
tees the property of a kind of “collision-free” and the
latter is the property of “one-way”. On Condition 1,
if a simple idea of distance is used as g
′
, then an at-
tack based on “hill-climbing” successes. Namely, in a
search of q such that p = φ(q) for a given p ∈ Σ
∗
, it is
possible to have an r such that g(q, r) < g(q, r
′
) by
considering whether g
′
(φ(q), φ(r)) < g
′
(φ(q), φ(r
′
))
recursively. This situation contradicts to Condition 2
in a strict sense. One of the solutions for this problem
A NOTE ON BIOMETRICS-BASED AUTHENTICATION WITH PORTABLE DEVICE
101