Protecting Biometric Features by Periodic Function-Based Transformation and Fuzzy Vault - Transactions on Large-Scale Data-and Knowledge-Centered Systems - page 62

3 The Proposed Scheme

3.1 General Architecture

Our proposed scheme can be applied for many kinds of biometric data whose feature
is in vectors. In this paper, we use the face biometric data for demonstration. Our
general architecture includes two main phase: enrollment and authentication. In en-
rollment phase, a set of biometric features is first extracted from the users' face imag-
es. After standardized, these features are then transformed by a sine function. The
randomly generated key is used to construct a polynomial function and its hashing is
stored for matching purpose in authentication phase. The transformed features apply
this polynomial function to generate a set of genuine points in fuzzy vault set. To
complete the fuzzy vault encoding step, a set of chaff points is also added into fuzzy
vault set. After that, all these points are stored in fuzzy vault database.
In authentication phase, sensor will take the image of a user and provide it for the
system. This image is also extracted in order to gain the user's biometric feature. The
sine transformation is performed on the extracted feature. If this transformed feature
has substantial overlap with the enrolled ones, the secret key will be correctly re-
trieved by the fuzzy vault decoding step. Afterwards, the recovered key is hashed in
order to compare with the hashed versions of the keys generated in the enrollment
phase. If the new key is matched, the user is authenticated. The overview of the sys-
tem is illustrated in the Fig. 1.

3.2 Feature Extraction

A feature extraction technique is used to extract the biometric feature. Among many
different feature extraction techniques, PCA (Principal Component Analysis), and
ICA (Independent Component Analysis) are popular ones for face recognition. In this
paper, we choose PCA for its significant outperformance on human face recognition
task [22].
In the Eigenfaces method, the PCA is applied to the training set to find a set of
standardized face ingredients, called eigenfaces. The training set is a large number of
images depicting different human faces, including Γ 1 ,Γ 2 ,Γ 3 ,…,Γ images. We de-
fined the average face of set as:

ψ

M

M ∑ Γ

(1)

The difference between each face and the average is shown by vector:

Γ ψ

(2)

Then, the covariance matrix is calculated by:

C

∑

(3)

where the matrix … .