Biometric Fusion and Multimodal Systems Flashcards

Which level of biometric fusion combines raw data or feature vectors from multiple sensors before matching occurs?

Answer: Feature-level fusion

Feature-level fusion operates on the feature vectors extracted from biometric samples prior to matching, merging them into a single combined feature set.

What is the primary advantage of using a multimodal biometric system over a unimodal system?

Answer: Improved accuracy and resistance to spoofing

Multimodal systems combine multiple biometric sources, improving overall accuracy, reducing false accept/reject rates, and making spoofing significantly more difficult.

In decision-level fusion, the AND rule requires that:

Answer: All matchers accept the identity claim

The AND rule in decision-level fusion grants access only when every individual matcher independently accepts the claim, maximizing security at the cost of higher false rejection.

Which fusion strategy is typically most difficult to implement due to incompatible feature spaces across different biometric modalities?

Answer: Feature-level fusion

Feature-level fusion is the most challenging because feature vectors from different modalities (e.g., fingerprint minutiae and iris codes) often have different dimensionalities and statistical properties.

Score normalization is most commonly applied before which type of fusion?

Answer: Score-level fusion

Score-level fusion requires normalization to map match scores from different matchers onto a common scale before combining them, since raw scores from different systems are not directly comparable.

A biometric system uses fingerprint and face recognition independently and combines their binary accept/reject outputs. This is an example of:

Answer: Decision-level fusion

Combining binary accept/reject decisions from independent matchers is the definition of decision-level fusion, the simplest but least flexible fusion approach.

What is a 'chimeric' database in the context of multimodal biometric research?

Answer: A database created by virtually pairing samples from different unimodal databases

A chimeric database is an artificially constructed multimodal dataset where multiple unimodal databases are combined by randomly pairing individuals, enabling fusion research without requiring real simultaneous multi-trait capture.