CBP Biometric Fusion and Multimodal Systems 2 — Questions and Answers
Question 1: The Sum Rule in score-level fusion computes the final score by:
- Taking the maximum score across all matchers
- Multiplying all matcher scores together
- Adding the normalized scores from each matcher (Correct answer)
- Selecting the score from the most reliable matcher
Correct answer: Adding the normalized scores from each matcher
The Sum Rule combines normalized matcher scores by addition, and has been shown to be robust and effective in practice even when score distributions are not ideal.
Question 2: Which score normalization technique uses the mean and standard deviation of match scores to map them to a standard distribution?
- Min-Max normalization
- Tanh estimator normalization
- Z-score normalization (Correct answer)
- Decimal scaling
Correct answer: Z-score normalization
Z-score normalization (also called z-normalization) subtracts the mean and divides by the standard deviation to produce a zero-mean, unit-variance distribution.
Question 3: In a cascaded (sequential) multimodal biometric architecture, what happens when the first modality produces a high-confidence match?
- The second modality is always invoked for verification
- Processing may stop and identity is accepted without querying further modalities (Correct answer)
- The system automatically rejects due to suspicion of spoofing
- Feature-level fusion is triggered
Correct answer: Processing may stop and identity is accepted without querying further modalities
In cascaded fusion, subsequent modalities are only invoked when earlier stages produce uncertain results, saving processing time and resources when initial confidence is high.
Question 4: The OR rule in decision-level fusion is most likely to result in:
- Lower False Accept Rate (FAR) compared to any single matcher
- Higher False Accept Rate (FAR) compared to any single matcher (Correct answer)
- Lower False Reject Rate (FRR) compared to any single matcher only
- Equal error rates across all participating matchers
Correct answer: Higher False Accept Rate (FAR) compared to any single matcher
The OR rule accepts an identity claim if any single matcher accepts, which tends to raise FAR because more pathways exist for an impostor to be accepted.
Question 5: Parallel fusion architecture differs from cascaded (sequential) fusion in that:
- Parallel fusion uses only one biometric modality at a time
- All biometric modalities are processed simultaneously and results are combined (Correct answer)
- Parallel fusion applies only to decision-level combination
- Parallel fusion eliminates the need for score normalization
Correct answer: All biometric modalities are processed simultaneously and results are combined
In parallel fusion, all matchers run concurrently and their outputs are combined afterward, maximizing accuracy at the cost of processing all modalities regardless of intermediate confidence.
Question 6: Which of the following is a key challenge specific to multimodal biometric systems that is less prominent in unimodal systems?
- Preventing feature drift over time
- Ensuring sensors are calibrated correctly
- Managing increased computational complexity and fusion strategy selection (Correct answer)
- Storing biometric templates securely
Correct answer: Managing increased computational complexity and fusion strategy selection
Multimodal systems introduce the additional challenges of selecting an appropriate fusion strategy, normalizing heterogeneous scores, and managing the computational overhead of running multiple matchers.
Question 7: Soft biometrics (e.g., height, gender, eye color) are most commonly incorporated into multimodal systems at which fusion level?
- Feature-level fusion only
- Sensor-level fusion only
- Score-level or decision-level fusion as supplementary evidence (Correct answer)
- They cannot be used in biometric fusion
Correct answer: Score-level or decision-level fusion as supplementary evidence
Soft biometrics lack sufficient discriminability to be used alone but can augment fusion at the score or decision level by boosting or penalizing match scores based on demographic consistency.
CBP Biometric Fusion and Multimodal Systems 2
The Sum Rule in score-level fusion computes the final score by: