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.
can address the problem <strong>of</strong> noisy sensor data, but all other potential problems<br />
associated with unimodal biometric systems remain.<br />
• Single Biometry, Multiple Instances: The same biometric unit is acquired<br />
several times by a same sensor and combined to complete and improve the<br />
recognition process (e.g. multiple face images <strong>of</strong> a person obtained under dif-<br />
ferent pose/lighting conditions).<br />
• Single Biometry, Multiple Units: The same biometric, but different units<br />
are acquired and combined to complete and improve the recognition process<br />
(e.g. left and right iris images). This is a recognition system that works on<br />
multiple units <strong>of</strong> the same biometric measurements (e.g. left middle finger<br />
followed by a right thumb).<br />
• Single Biometry, Multiple Representations: The same biometric is ac-<br />
quired once by a single sensor and different approaches <strong>of</strong> feature extraction<br />
and matching are combined to complete and improve the recognition process<br />
(e.g. multiple face matcher like PCA and LDA).<br />
• Multiple Biometrics: Different biometrics <strong>of</strong> the same person are acquired<br />
and combined to complete and improve the recognition process (e.g. face,<br />
fingerprint and iris). This approach is the only well-used multimodal biometric<br />
fusion scenario.<br />
Although the first four methods improve the recognition performance <strong>of</strong> a system,<br />
they still suffer from some <strong>of</strong> the problems faced by unimodal systems. A multimodal<br />
systems based on different traits seems to be more robust to noise, address the prob-<br />
lem <strong>of</strong> non-universality, provide reasonable protection against spo<strong>of</strong> attacks and also<br />
improve matching accuracy.<br />
2.3.3 Fusion Levels and Methods <strong>of</strong> Integration<br />
Fusion in multimodal biometric systems can take place at three major levels, namely,<br />
sensor level, feature level and decision level. These three levels can be broadly clas-<br />
sified into fusion prior to matching and fusion after matching [108]. Decision level<br />
fusion can be divided into three classes based on the type <strong>of</strong> the output <strong>of</strong> biometric<br />
35