Abstract:Combining match scores from different biometric systems via fusion is a well-established approach to improving recognition accuracy. However, missing scores can degrade performance as well as limit the possible fusion techniques that can be applied. Imputation is a promising technique in multibiometric systems for replacing missing data. In this paper, we evaluate various score imputation approaches on three multimodal biometric score datasets, viz. NIST BSSR1, BIOCOP2008, and MIT LL Trimodal, and investigate the factors which might influence the effectiveness of imputation. Our studies reveal three key observations: (1) Imputation is preferable over not imputing missing scores, even when the fusion rule does not require complete score data. (2) Balancing the classes in the training data is crucial to mitigate negative biases in the imputation technique towards the under-represented class, even if it involves dropping a substantial number of score vectors. (3) Multivariate imputation approaches seem to be beneficial when scores between modalities are correlated, while univariate approaches seem to benefit scenarios where scores between modalities are less correlated.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in a multimodal biometric system, how to handle missing matching scores to improve the effectiveness of fusion techniques and the overall performance of the system. Specifically, the researchers explored the impact of different imputation methods on multimodal biometric fusion and evaluated the performance of these methods on different datasets. ### Specific background of the problem 1. **Fusion in multimodal biometric systems** - Biometric systems can improve recognition accuracy and security by combining multiple biometric features (such as face, fingerprint, iris, etc.). - Score - level fusion is a common fusion method, that is, combining matching scores from different modalities or matchers. 2. **Impact of missing scores** - Missing scores can be caused by various reasons, such as sample collection failure or insufficient quality, or the introduction of a new modality resulting in a mismatch between the input probe data and the identities in the existing database. - Missing scores will affect the choice and effectiveness of fusion techniques. Some fusion techniques cannot handle missing data, and simply ignoring these data may reduce system performance. 3. **Choice of imputation methods** - The researchers evaluated different imputation methods, including univariate imputation (such as mean, median imputation) and multivariate imputation (such as multivariate imputation by chained equations, MICE). - They also explored the impact of training data balance on the imputation effect and the impact of the correlation between different modalities on the choice of imputation method. ### Main contributions of the paper 1. **Imputation is better than non - imputation** - The study found that, regardless of whether complete score data is required or not, imputing missing scores is generally better than not imputing, and can improve system performance. 2. **Importance of training data balance** - Balancing the class distribution in the training data can reduce the bias of the imputation method against minority classes, even if this means discarding a large number of data points. 3. **Advantages of multivariate imputation** - When the score correlation between modalities is high, multivariate imputation methods perform better; while in the case of low correlation, univariate imputation methods may be more effective. ### Conclusions and recommendations - **Recommendation 1**: Integrate imputation techniques in the design of multimodal biometric systems and choose the imputation method that is most suitable for the data. - **Recommendation 2**: Balance the genuine and impostor score vectors in the training set, although this may require discarding a large number of over - represented data points. - **Recommendation 3**: When designing imputation methods, consider the nature of missing scores and the inherent correlation between modalities, and develop targeted and effective imputation strategies. Through these studies, the paper provides valuable insights into imputation techniques in multimodal biometric systems and lays the foundation for further research.

To Impute or Not: Recommendations for Multibiometric Fusion

On Missing Scores in Evolving Multibiometric Systems

An Approach for Full Reinforcement-Based Biometric Score Fusion

Multimodal biometric system using rank-level fusion approach

Enhancing Recognition in Multimodal Biometric Systems: Score Normalization and Fusion of Online Signatures and Fingerprints

Optimal weight of the score level multiplebiometric feature fusion

Benchmarking Quality-Dependent and Cost-Sensitive Score-Level Multimodal Biometric Fusion Algorithms

A comprehensive overview of biometric fusion

Robust multimodal biometric system based on optimal score level fusion model

Multi-Biometrics Fusion for Identity Verification

Combining Multiple Biometric Traits with an Order-Preserving Score Fusion Algorithm

Serial fusion of multi-modal biometric systems

Quasi-convex Optimization of Metrics in Biometric Score Fusion

A Decision Reliability Ratio Based Fusion Scheme for Biometric Verification

Score Level Fusion Scheme Based on Adaptive Local Gabor Features for Face-Iris-Fingerprint Multimodal Biometric

Imputation using information fusion technique for sensor generated incomplete data with high missing gap

Score, Rank, and Decision-Level Fusion Strategies of Multicode Electromyogram-Based Verification and Identification Biometrics

A hybrid fusion model of iris, palm vein and finger vein for multi-biometric recognition system

Quality-Based Conditional Processing in Multi-Biometrics: Application to Sensor Interoperability

Artificial intelligence-Enabled deep learning model for multimodal biometric fusion

Score Level and Rank Level Fusion for Kinect-Based Multi-Modal Biometric System