20.01.2013 Views

Master Thesis - Department of Computer Science

Master Thesis - Department of Computer Science

Master Thesis - Department of Computer Science

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Q = �<br />

(q 1 x, q 1 y, β 1 ), ....., (q Q x , q Q y , β Q ) �<br />

. (A.35)<br />

where |P| = P , |Q| = Q, and (p i x , pi y , αi ) are the three features associated with the<br />

i th minutiae in set P. The basic assumption made here is that the second point set<br />

Q is a rotated, scaled, and translated version <strong>of</strong> the first set P, where points may be<br />

shifted by random noise, some points may be added and some points deleted. The<br />

task <strong>of</strong> fingerprint registration is to recover this unknown transformation. Since we<br />

do not know whether the two fingerprints are the same or not (i.e., images <strong>of</strong> the same<br />

finger), we attempt to find the ’best’ transformation in the sense that when applying<br />

the transformation to the minutiae points <strong>of</strong> the set P, as many <strong>of</strong> these points as<br />

possible overlap with the minutiae points from the set Q. Two overlapping points are<br />

considered as a match only if they have the same direction. There may be minutiae<br />

points in either set that do not match with any point in the other set.<br />

The matching score is computed for each transformation after discretizing the set<br />

<strong>of</strong> all allowed transformations. Consider a transformation Fs,θ,∆x,∆y : R 2 → R 2 , given<br />

by,<br />

⎛<br />

⎜<br />

Fs,θ,∆x,∆y ⎝ x<br />

⎞ ⎛<br />

⎞ ⎛<br />

⎟ ⎜<br />

⎠ = s ⎝<br />

y<br />

cosθ<br />

−sinθ<br />

sinθ ⎟ ⎜<br />

⎠ ⎝<br />

cosθ<br />

x<br />

⎞ ⎛<br />

⎟ ⎜<br />

⎠ + ⎝<br />

y<br />

∆x<br />

⎞<br />

⎟<br />

⎠ ,<br />

∆y<br />

(A.36)<br />

where s, θ, and (∆x, ∆y) are the scale, rotation, and shift parameters, respectively.<br />

The space <strong>of</strong> transformations consists <strong>of</strong> quadruples (s, θ, ∆x, ∆y), where each pa-<br />

rameter is discretized into a finite set <strong>of</strong> values:<br />

s ∈ s1, .., sK , θ ∈ θ1, ..., θL,<br />

∆x ∈ ∆x1, ..., ∆xM and ∆y ∈ ∆y1, ..., ∆yN<br />

Matching scores for the transformations are collected in the accumulator array A,<br />

where the entry A(k, l, m, n) counts the evidence for the transformation Fsk,θl,∆xm,∆yn.<br />

The array A is filled as follows: For each pair (p, q), where p = (pi x , piy ) is a point in<br />

the set P and q = (q j x, q j y) is a point in the set Q, find all possible transformations that<br />

map p to q and increment the evidence for these evidence for these transformations in<br />

array A. For every pair <strong>of</strong> values, (sk, θl), there is exactly one shift vector (∆x, ∆y) t<br />

such that Fsk,θl,∆x,∆y(p) = q, and it can be obtained as,<br />

⎛<br />

⎜<br />

⎝ ∆x<br />

⎞ ⎛<br />

⎟ ⎜<br />

⎠ = q − sk ⎝<br />

∆y<br />

cosθl<br />

⎞<br />

−sinθl<br />

sinθl ⎟<br />

⎠ p.<br />

cosθl<br />

(A.37)<br />

140

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!