Master Thesis - Department of Computer Science
Master Thesis - Department of Computer Science
Master Thesis - Department of Computer Science
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
[105] in combining face and hand-geometry features have met with only limited<br />
success.<br />
Since the features contain richer information about the input biometric data than the<br />
matching scores obtained at decision level, integration at feature level is believed to<br />
be more effective than decision level fusion. But feature fusion is <strong>of</strong>ten not possible<br />
for the unknown or incompatible relationship between different feature spaces, and<br />
<strong>of</strong>ten simple concatenation <strong>of</strong> features may cause “curse <strong>of</strong> dimensionality problem”.<br />
2.3.3.2 Fusion After Matching<br />
Information fusion after matching can be divided into two main categories: dynamic<br />
classifier selection and decision fusion.<br />
A dynamic classifier selection technique chooses the result <strong>of</strong> that classifier which<br />
is most likely to give correct decision for the specific input pattern [132]. This is<br />
well-known as winner-take-all approach and the device that performs the selection is<br />
known as associative switch [24].<br />
Information fusion at decision level can be at abstract, rank and measurement<br />
level based on the type <strong>of</strong> matcher’s output.<br />
• Abstract level: The biometric matcher individually decides on the best match<br />
based on the input presented. Methods like majority voting [71], behavior<br />
knowledge space [70], weighted voting based on the Dempster-Shafer theory <strong>of</strong><br />
evidence [135], AND and OR rules [31], etc. are use to obtain the final decision.<br />
• Rank level: The output <strong>of</strong> each biometric matcher is a subset <strong>of</strong> possible<br />
matches sorted in decreasing order <strong>of</strong> confidence. Ho et al. [46] described three<br />
methods to combine the ranks assigned by different matchers. In the highest<br />
rank method, each possible match is assigned the highest (minimum) rank as<br />
computed by different matchers. Final decision is made based on the combined<br />
ranks and ties are broken randomly. The Borda count method uses the sum <strong>of</strong><br />
the ranks assigned by the individual matchers to calculate the combined ranks.<br />
The logistic regression method is a generalization <strong>of</strong> the Borda count method<br />
where weighted sum <strong>of</strong> the individual ranks is calculated and the weights are<br />
38