Face Detection and Modeling for Recognition - Biometrics Research ...
Face Detection and Modeling for Recognition - Biometrics Research ...
Face Detection and Modeling for Recognition - Biometrics Research ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
5.4.1 Component Weights <strong>and</strong> Matching Cost<br />
After the two phases of face alignment, we can automatically derive a weight (called<br />
semantic component weight) <strong>for</strong> each facial component i <strong>for</strong> a subject P with N p<br />
training face images by<br />
scw P (i) =<br />
d(i) = 1<br />
N P<br />
{<br />
1 + e −2σ2 d (i)/d2 (i)<br />
N p > 1,<br />
1 + e −1/d2 (i)<br />
N p = 1,<br />
∑N P<br />
k=1<br />
(5.19)<br />
SF D i (G 0 , G Pk ) · MS P k<br />
(i), (5.20)<br />
σ d (i) =SD k<br />
[<br />
SF Di (G 0 , G Pk ) · MS P k<br />
(i) ] , (5.21)<br />
where SF D means semantic facial distance, MS is the matching score, SD st<strong>and</strong>s<br />
<strong>for</strong> st<strong>and</strong>ard deviation, G 0 <strong>and</strong> G Pk<br />
semantic face graphs, respectively.<br />
are the coarsely aligned <strong>and</strong> finely de<strong>for</strong>med<br />
The semantic component weights take values<br />
between 1 <strong>and</strong> 2. The semantic facial distance of facial component i between two<br />
graphs is defined as follows<br />
SF D i (G 0 , G Pk ) =Dist(SGD G 0<br />
i<br />
[<br />
1 ∑L i<br />
∣<br />
= ∣a G 0<br />
L i<br />
k=0<br />
i<br />
, SGD G P k<br />
i )<br />
(k) − a G P k<br />
i<br />
] 0.5 (k) ∣ 2 , (5.22)<br />
where SGD st<strong>and</strong>s <strong>for</strong> semantic graph descriptors. The distinctiveness of a facial<br />
component is evaluated by the semantic facial distance SF D between the generic<br />
semantic face graph <strong>and</strong> the aligned/matched semantic graph.<br />
The visibility of a<br />
facial component (due to head pose, illumination, <strong>and</strong> facial shadow) is estimated<br />
132