Abstract
Kalka, Bartlow, and Cukic adopt the work of Kryszczuk and Drygajlo on unimodal and bimodal biometric classification [1], and present a rather straightforward extension to multimodal classification.
In their 2008 paper [1], Kryszczuk and Drygajlo employ a subjective Bayesian methodology to determine the credibility of a decision. They consider a single event probability as a degree of belief in the occurrence of that event. They demonstrate that such credence estimates can help both in predicting and rectifying verification errors. Consequently, the classification performance can be improved. Moreover, they show in a small experiment that their technique can also be of use for fusion of verification decisions, using two modalities.
Kalka, Bartlow, and Cukic’s results show marginal improvement in classification performance. More than anything else, their paper illustrates the complexity of multimodal biometric classification. As it is often shown, sensors’ multimodal biometric information processing cannot keep up with the rapid developments in sensor technology.
1) Kryszczuk, K.; Drygajlo, A. Credence estimation and error prediction in biometric identity verification. Signal Processing 88, 4 (2008), 916–925.
Original language | Undefined |
---|---|
Pages (from-to) | CR137554 |
Number of pages | 1 |
Journal | Computing reviews |
Publication status | Published - 8 Dec 2009 |
Keywords
- EWI-18400
- Bayesian methodology
- Review
- Biometrics
- Multimodal
- HMI-IE: Information Engineering