You are on page 1of 1

d 2d 3 V.


θˆP ' = θ 1"
d 1d 2 + d 2 d 3 + d 3 d 1 To compare two fingerprints’ orientation field, the first
d 3d 1 step is alignment of these two fingerprints. The Hough
+ θ 2" transform based approach has been chosen to finish the
d 1d 2 + d 2 d 3 + d 3 d 1 alignment due to its simplicity. In the matching step, the
d 1d 2 correlation between two aligned orientation fields, A and
+ θ 3"
d 1d 2 + d 2 d 3 + d 3 d 1 B, is computed as below. Let Ω denotes the intersection
(3) of the two effective regions after alignment, and N is the
θˆP is calculated as, total number of points in Ω . The matching score between
two orientation fields is defined as
π 1
π + θˆP ' , − < θˆP ' < 0 s(A, B ) = δ (i, j ) (9)
2 N ( i , j )∈ Ω
θˆP = θˆP ' , 0 ≤ θˆP ' < π

2 π − θˆP ' , π ≤ θˆP ' < In (9), δ (i, j ) is the difference between the
2 (4) orientation values at the point, (i, j) in image A and B,
which is formulated as follows:
The computation is illustrated in Fig. 4. After the π
interpolation step, the minutiae distribute “uniformly” as δo (i , j ), if δo (i , j ) ≤ (10)
shown in Fig. 5(c).
δ (i , j ) = 2
π − δo (i , j ), otherwise
C. Polynomial Model
The orientation field is firstly mapped to a continuous and δo (i, j ) is defined as
complex function. Denoting (x, y) and U(x, y) as the
orientation field and the transformed function, δ o ( i , j ) = θ A ( i , j ) − θ B ( i. j ) (11)
respectively, the mapping can be defined as
here θA(i , j ) and θB (i. j ) are the direction of point, (i, j),
U(x, y) = RE (x, y) + i. IM (x, y)
= cos2 (x, y) + i. sin2 (x, y) (5) in image A and B. If the matching score s (A, B) is higher
than a certain threshold, we say the two orientation fields
where RE(x, y) and IM(x, y) denote respectively the real are “matched.”
part and imaginary part of the complex function, U(x, y).
Obviously, RE(x, y) and IM(x, y) are continuous with x, y,
in those regions. The above mapping is a one-to-one Two experiments are carried out to test the
transformation and (x, y) can be easily reconstructed performance of the system. First, the proposed scheme is
from the values of RE(x, y) and IM(x, y). To globally applied to fingerprint images and the matching accuracy is
represent these, two bivariate polynomial models are evaluated. The second experiment is the comparison of the
established, which are denoted by PR(x, y) and PI(x, y), proposed method with existing methodology.
respectively. These two polynomials can be formulated as A. Performance Evaluation
PR(x, y) = X . P1. Y (6)
PI(x, y) = XT. P2. Y (7)

where X = (1, x, x2……..xN) T and Y = (1, y, y2……..yN)

T. In these two formulas, N is the order of the polynomial
model. There are (N+1) x (N+1) parameters of P1 and P2
which need to be calculated. Computing the parameters is
a fitting process. Using square sum error for evaluation,
the formula becomes

( P1*, P 2*) = arg min [(PR ( x, y , P1) − cos 2θo ( x, y )) 2

P1, P 2
( x , y )∈Ω

+ ( PI ( x, y , P 2) − sin 2θo ( x, y )) 2 ] (8) Figure 6. Results of the proposed algorithm: (a) virtual minutiae by
interpolation (the bigger red minutiae are “real”, while the smaller
where Ω is the set of the effective region, and θo ( x , y ) purple ones are “virtual”); (b) the reconstructed orientation field.

is the original orientation field.