Information Technology Reference
In-Depth Information
The results of are shown in the Fig. 6 and Fig. 7 correspond to original transfor-
mation and new transformation. The vertical axis is the error rate which is range [0,
1], and the horizontal axis is the epsilon (
as defined in Section 3.6. Epsilon is the
minimum distance among points in the vault. Using the original transformation, we
can get the acceptable error rates at
0.22
. With this value, the error rates are:
0.3
. Otherwise, using the new transformation, we can get the better
error rates
0.25
at
0.0175
.
0.8
0.7
0.6
0.5
0.4
FRR
FAR
0.3
0.2
0.1
0
Fig. 7.
FAR and FRR of proposed scheme with new transformation
5
Conclusion and Future Works
In this paper, we proposed a hybrid scheme which combines two approaches, namely
fuzzy vault and periodic function-based feature transformation, to protect biometric
templates. In this scheme, we perform a transformation on the biometric template and
then let it as an input to the fuzzy vault. By this way, we can strengthen the fuzzy
vault with the revocability property. Our transformation function is non-invertible
because the transformation function, i.e. sine function, is periodic with the period
2
.
With the knowledge of sin(x), we cannot infer the true value of x.
The results of the evaluation confirm the effective and the practical properties of
our scheme to protect biometric template.
The next works we do with our research are to reduce the error rates (FAR, FRR)
so that it can be used in practice and to find a proper way to add chaff points to the
vault. For the first task, we need more researches on the range of each component of
the feature vectors so that we can adjust our parameters, namely the minimum dis-
tance among points in the vault and the maximum range of
x
in the vault. If we can
decide these parameters more precisely we can archive a lower error rates in our
scheme. The other approach is that we can apply multimodel biometrics schema to
Search WWH ::
Custom Search