Adaptive biometric verification system using quality-based co-training

The performance of a biometric verification system may degrade substantially if the input samples vary significantly compared to existing samples in the gallery. Adaptive biometric systems that can improve with use, have recently gained popularity together with using semi-supervised learning methods for accommodating the continuous change in the subject's data. In this study we investigate the value of using quality measures of biometrics to incorporate it in a semi-supervised learning context for the design of an adaptive biometric system. The novelty of the proposed approach is the use of quality information of input samples as an extra source of information to update the user gallery. Our results show that using quality measures in the fusion process will improve system performance.