Face Detection and Modeling for Recognition - Biometrics Research ...
Face Detection and Modeling for Recognition - Biometrics Research ...
Face Detection and Modeling for Recognition - Biometrics Research ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
minimum energy.<br />
The attraction <strong>for</strong>ce field consists of two kinds of fields in Eq.<br />
(a) (b) (c)<br />
Figure 5.9. Repulsion <strong>for</strong>ce: (a) interacting snakes with index numbers marked; (b)<br />
the repulsion <strong>for</strong>ce computed <strong>for</strong> the hair outline; (c) the repulsion <strong>for</strong>ce computed<br />
<strong>for</strong> the face outline.<br />
(5.11): one is obtained from edge strength, called gradient vector field (GVF) [127],<br />
<strong>and</strong> the other from a region pressure field (RPF) [133].<br />
−∇E image (v i (s)) = GV F + RP F<br />
)<br />
= GV F + ρ · ⃗N(v i (s)) ·<br />
(1 − |Ecomp i (v i (s)) − µ|<br />
,<br />
kσ<br />
(5.11)<br />
where ⃗ N(v i (s)) is the normal vector on the i th contour v i (s); E comp<br />
i<br />
is the component<br />
energy of the i th component; µ, σ are the mean <strong>and</strong> the st<strong>and</strong>ard deviation of region<br />
energy over a seed region of the i th component; k is a constant that constrains the<br />
energy variation of a component. The advantage of using GVF <strong>for</strong> snake de<strong>for</strong>mation<br />
is that its range of influence is larger than that obtained from gradients, <strong>and</strong> can<br />
attract snakes to a concave shape. A GVF is constructed from an edge map by an<br />
iterative process. However, the construction of GVF is very sensitive to noise in the<br />
124