WVU Biometric Fusion Tool
Biometrics is the science of establishing human identity based on the physical or behavioral attributes of an individual  . Multibiometric systems overcome many practical problems that occur in single modality biometric systems, such as noisy sensor data, non-universality and/or lack of distinctiveness of a biometric trait, unacceptable error rates and spoof attacks, by consolidating multiple biometric information pertaining to the same identity  . Biometric fusion can be implemented at various levels, such as raw data level, image level, feature level, rank level, score level and decision level. Fusion at the score level is the most popular approach discussed in the literature  and  .
The initial goal of this work was to build a GUI tool via Matlab, which can implement multiple biometric fusion rules on the match score level. There are three different families of fusion rules included in this tool.
- Simple Fusion Rules: We consider fusion rules like Sum Rule, Product Rule, Min Rule and Max Rule as simple fusion rules. Those rules present a shorter running time, however, a normalization procedure is required to transform match scores into a common domain before running them. The following normalization schemes are included in this tool, and the details of these schemes can be found in .
- Likelihood-ratio Based Rule: This family is based on the likelihood ratio test and the main advantage is to directly achieves optimal performance at any desired operating point (FAR), provided the score densities are estimated accurately. In this tool, three density estimation methods are offered to estimate genuine and impostor match score densities:
- Quality-Based Fusion Rules: The quality of biometric samples has a significant impact on the accuracy of a matcher . Therefore, dynamically assigning weights to the outputs of individual matchers based on the quality of the samples presented at the input of the matchers can improve the overall recognition performance of a multibiometric system. Here, we offer 3 different quality-based fusion rules:
- User Designed Fusion: As an extension, we add a "user-design fusion rules" option to the tool. Users can upload their own Matlab functions (.m files) to the tool, do experiments and compare the performance with multiple fusion rules. We also offer a template of this kind of user-design functions see the template.
Back to top
Input and Output
Back to top
- Load Match Scores
- New Fusion: Press the button if the user need to clear the current computation or start a new fusion.
- Load Genuine Scores & Load Imposter Scores: Match scores can be imported into the tool via .txt files. Unique format is used here because it is fair when the comparison of running time is required under some situation. In the .txt file, scores from different modalities should be separated as different columns. Imposter scores and Genuine scores should be separated as different files.
- Conventional Fusion: Choose this option if no qualify scores will be used
- Quality-Based Fusion : Choose this option if quality scores are available to the fusion. A new Window will jump out, where the user can import quality scores for Genuine and Imposter subjects, separately.
- Prune Modalities : If not all the columns fo the .txt file are used in the fusion, then this function can help you choose the right columns.
- Generate Training and Test Set
- Generate Training Set: A training set will be generated by choosing a training rate. The clients in the original data file will be randomly selected based on this training rate.
- Density Estimation
- Density Plot: Plot estimated density plots of the training set, test set or the entire dataset.
- Normalization: Some fusion rules, like the sum rule, require the normalization process before fusion. The GUI tool offers several normalization schemes: “Min-max? “Median and MAD? “Biweight? “Double sigmoid?, “Z-score? “Decimal scaling?and “Tanh? “No normalization?is the final option.
- Customizable Parameter Setting: During the normalization process, the normalization parameters are obtained from the training set directly most of the time. However, some users may assign parameters by themselves. The GUI tool offers an option that allows users to assign parameters manually. The normalized scores can also be plotted by this function.
- Fuse and ROC Plot: The tool offers the following fusion methods:
- Simple Fusion Rules: Simple fusion rules include “Min Rule? “Max Rule? “Sum Rule? and “Product Rule?
- Likelihood-Ratio Based Fusion (LRF): The key point of LRF methods is the density estimation of the training scores. Here, the tool offers several options for density estimation: the Gaussian assumption, the Gaussian Mixture Model and the Parzen Window scheme.
- Quality-Based Fusion: The tool offers 3 simple quality-based fusion methods: "Quality-Weighted Sum","Quality-Based Max","Quality-Based Likelihood Ratio".
- User Designed Fusion: In this option, users can upload their fusion code in the specific format. Then the tool embeds the code and displays the final fusion results.
- Export XML File for GRR-FRR Pair: The pair-wise values of the Genuine Reject Rate (GRR) and False Reject Rate (FRR) is the critical information for the subsequent analysis. For instance, combined with the cost matrix, GRR-FRR pair can be used to generate cost curves. So here, the tool offers two options to generate this pair-wise value: the “Export XML File for GRR-FRR Pair?function will export GRR-FRR pairs under multiple thresholds, and the “Generate Arbitrary GRR-FRR Pair?function allows users to simply switch thresholds, and compare GRR-FRR pairs in real time.
Back to top
Authorship and Use
This Matlab GUI tool was written by Yaohui Ding and Dr. Arun Ross of the iPRoBe laboratory in the Lane Department of Computer Science and Electrical Engineering, West Virginia University. The tool may be distributed free of charge and used by anyone if credit is given. It has been tested fairly well, but it comes with no guarantees and the authors assume no liability for its use or misuse.
Back to top
Development of this software has been supported by the grant from Center for Identification Technology Research (CITeR).
Back to top
References(Please Don't Forget Them)
. A.K.Jain, P. Flynn, A.A. Ross, Handbook of Biometrics, Springer, 2008
. A.K.Jain, K.Nandakumar, A.Ross, Score normalization in multimodal biometric systems, Pattern Recognition 38(12)(2005)2270?285.
. A.A.Ross, K.Nandakumar, A.K.Jain, Hand book of Multibiometrics, Springer, Secaucus, NJ,USA,2006.
. Karthik Nandakumar, Yi Chen, Anil K. Jain Quality-based Score Level Fusion in Multibiometric Systems, ICPR2006
Back to top