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.
• Spo<strong>of</strong> Attacks: An impostor may attempt to spo<strong>of</strong> the biometric trait <strong>of</strong> a<br />
legitimate enrolled user in order to circumvent the system. This type <strong>of</strong> attacks<br />
are well-known in case <strong>of</strong> behavioral traits like signature and voice. Even it is<br />
possible to construct dummy fingers using lifted fingerprint impressions.<br />
Multimodal biometry solves the above defined problems by combining the evidences<br />
obtained from different modalities with the help <strong>of</strong> an effective fusion scheme. An<br />
alternate use <strong>of</strong> multimodal biometry is to perform a search in an efficient and faster<br />
way by using a relatively simple and less accurate modality to prune the database be-<br />
fore using the more complex and accurate modality on the remaining data to generate<br />
the final decision.<br />
There are, however, a few disadvantages <strong>of</strong> using a multimodal biometric sys-<br />
tem. First, a multimodal biometric system is more expensive and requires more<br />
computational and storage resources than a unimodal system. Second, multimodal<br />
systems generally require more time for enrollment and verification, causing some<br />
inconveniences to the user. Furthermore, if the multiple modalities are not properly<br />
combined, the combination may actually degrade a system accuracy.<br />
1.3 Brief Description <strong>of</strong> the Work Done<br />
The contribution <strong>of</strong> the work presented in this thesis comes in terms <strong>of</strong> three novel face<br />
recognition approaches and one decision fusion technique for combining information<br />
from face and fingerprint classifiers for multimodal biometry.<br />
• Among three face recognition approaches, the first approach presents an effi-<br />
cient method for frontal face recognition, using subject-specific subband face<br />
representation. The human face has certain visual features that are common<br />
among everybody and some others that exhibit the unique characteristics <strong>of</strong><br />
an individual. Using the discrete wavelet transform (DWT), we extract these<br />
unique features from the face image for discriminating it from others. The<br />
face image is decomposed into several subbands to separate the common (ap-<br />
proximation) and discriminatory (detail) parts. Subband face is reconstructed<br />
from selected wavelet subbands, in which a suitable approximation subband is<br />
5