Addressing Missing Values in Kernel-Based Multimodal Biometric Fusion Using Neutral Point Substitution

Abstract

In multimodal biometric information fusion, it is common to encounter missing modalities in which matching cannot be performed. As a result, at the match score level, this implies that scores will be missing. We address the multimodal fusion problem involving missing modalities (scores) using support vector machines with the Neutral Point Substitution (NPS) method. The approach starts by processing each modality using a kernel. When a modality is missing, at the kernel level, the missing modality is substituted by one that is unbiased with regards to the classification, called a neutral point. Critically, unlike conventional missing-data substitution methods, explicit calculation of neutral points may be omitted by virtue of their implicit incorporation within the SVM training framework. Experiments based on the publicly available Biosecure DS2 multimodal (scores) data set shows that the SVM-NPS approach achieves very good generalization performance compared to the sum rule fusion, especially with severe missing modalities.

Fusing voice and query data for non-invasive detection of laryngeal disorders

“…It would be also interesting to explore decision-level fusion by meta-learner using other combiners, for example, the isotonic or logistic regression (Brümmer & de Villiers, 2011) as well as a specific solution when some modalities are missing -a support vector machine with the neutral point substitution (Poh et al, 2010). Parameter Value Description qq.pr 0.5 speech probability threshold qq.ts 0.4 * length(s)/fs mean talkspurt length (in seconds) qq.tn 0.1 * length(s)/fs mean silence length (in seconds) qq.ti 0.1 desired output frame increment (in seconds) qq.tj 0.1 internal frame increment (in seconds) qq.gx 50 maximum posterior SNR as a power ratio qq.gz 0.001 minimum posterior SNR as a power ratio qq.ne 1 noise estimation (0 = min statistics, 1 = MMSE) min (or max) value and its respective relative position within a signal, range, arithmetic mean, 2 linear regression coefficients and linear and quadratic error, standard deviation, skewness, kurtosis, quartile 1-3, and 3 inter-quartile ranges …”

A Kernel-Based Framework for Medical Big-Data Analytics

“…However, methods developed by the author (the 'neutral point substitution' method [17][18][19]) render this tractable. (Neutral point substitution involves the unbiased missing value substitution of a placeholder mathematical object in multi-modal SVM combination).…”

Kernel combination via debiased object correspondence analysis

“…In general, the problem domain will determine whether object correspondence information is available. For instance, it is not uncommon in multimodal biometrics to obtain distinct sets of exemplar subjects for each individual biometric measurement (e.g., iris scans, finger prints, photographic images), particularly when employing separate commercial sources [28]. In this case, we would wish to utilize the information collectively contained within each data set for a given test subject, but would lack object correspondences in the collective set of multi-modal data sets.…”