Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On Missing Scores in Evolving Multibiometric Systems

Melissa R Dale, Anil Jain, Arun Ross

TL;DR

The paper tackles missing scores in evolving multibiometric systems, addressing how to preserve recognition accuracy as modalities are added, merged, or retired. It evaluates simple sum fusion in conjunction with several missing-data imputation methods, including Listwise Deletion, mean/median substitution, and multivariate approaches like MICE and iterative KNN. Across three real-world scenarios, the study shows that imputation substantially improves both verification and identification performance, with iterative KNN consistently delivering the strongest results, even when up to 90% of scores are missing; retraining can be preferable when retiring a modality. The findings highlight the practical viability of imputation-based fusion to maintain system performance during evolution, while also noting dataset-dependent gains and the need for larger, more diverse datasets and future work on score quality and other fusion levels.

Abstract

The use of multiple modalities (e.g., face and fingerprint) or multiple algorithms (e.g., three face comparators) has shown to improve the recognition accuracy of an operational biometric system. Over time a biometric system may evolve to add new modalities, retire old modalities, or be merged with other biometric systems. This can lead to scenarios where there are missing scores corresponding to the input probe set. Previous work on this topic has focused on either the verification or identification tasks, but not both. Further, the proportion of missing data considered has been less than 50%. In this work, we study the impact of missing score data for both the verification and identification tasks. We show that the application of various score imputation methods along with simple sum fusion can improve recognition accuracy, even when the proportion of missing scores increases to 90%. Experiments show that fusion after score imputation outperforms fusion with no imputation. Specifically, iterative imputation with K nearest neighbors consistently surpasses other imputation methods in both the verification and identification tasks, regardless of the amount of scores missing, and provides imputed values that are consistent with the ground truth complete dataset.

On Missing Scores in Evolving Multibiometric Systems

TL;DR

The paper tackles missing scores in evolving multibiometric systems, addressing how to preserve recognition accuracy as modalities are added, merged, or retired. It evaluates simple sum fusion in conjunction with several missing-data imputation methods, including Listwise Deletion, mean/median substitution, and multivariate approaches like MICE and iterative KNN. Across three real-world scenarios, the study shows that imputation substantially improves both verification and identification performance, with iterative KNN consistently delivering the strongest results, even when up to 90% of scores are missing; retraining can be preferable when retiring a modality. The findings highlight the practical viability of imputation-based fusion to maintain system performance during evolution, while also noting dataset-dependent gains and the need for larger, more diverse datasets and future work on score quality and other fusion levels.

Abstract

The use of multiple modalities (e.g., face and fingerprint) or multiple algorithms (e.g., three face comparators) has shown to improve the recognition accuracy of an operational biometric system. Over time a biometric system may evolve to add new modalities, retire old modalities, or be merged with other biometric systems. This can lead to scenarios where there are missing scores corresponding to the input probe set. Previous work on this topic has focused on either the verification or identification tasks, but not both. Further, the proportion of missing data considered has been less than 50%. In this work, we study the impact of missing score data for both the verification and identification tasks. We show that the application of various score imputation methods along with simple sum fusion can improve recognition accuracy, even when the proportion of missing scores increases to 90%. Experiments show that fusion after score imputation outperforms fusion with no imputation. Specifically, iterative imputation with K nearest neighbors consistently surpasses other imputation methods in both the verification and identification tasks, regardless of the amount of scores missing, and provides imputed values that are consistent with the ground truth complete dataset.
Paper Structure (14 sections, 10 figures, 5 tables)

This paper contains 14 sections, 10 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Biometric Systems
  4. Fusion with Missing Scores
  5. Experiments
  6. Experiment 1: Adding a New Modality to an Existing Biometric System
  7. Experiment 2: Merging Biometric Systems
  8. Experiment 3: Retiring a Modality
  9. Results
  10. Experiment 1: Adding a New Modality to an Existing Biometric System
  11. Experiment 2: Merging Biometric Systems
  12. Experiment 3: Retiring a Modality
  13. Conclusions
  14. Future Work

Figures (10)

  • Figure 1: Breakdown of provided Ocular image in the BIOCOP2008 dataset
  • Figure 2: Experiment contexts for NIST BSSR1 dataset.
  • Figure 3: ROC Curves for the complete NIST BSSR1 dataset
  • Figure 4: ROC Curves for the complete BIOCOP2008 dataset
  • Figure 5: ROC curves for imputation methods on the 10% missing in the NIST BSSR1 dataset scenario in Experiment 1. The curves corresponding to fusion with imputed values overlap near the top with the 0% missing baseline curve, which is shown in pink.
  • ...and 5 more figures