Biometric system adaptation by self-update and graph-based techniques

Self-update is the most commonly adopted biometric template update technique in which the system adapts itself to the confidently classified samples. However, the recent works indicate that self-update has limited capability to capture samples representing significant intra-class variations. As an alternative, a biometric template update technique based on the graph-based representation is proposed. This technique can potentially capture samples with significant variations, resulting in efficient adaptation. Until now, the efficacy of these adaptation techniques has been proven only on the basis of experimental evaluations on small data sets. The contribution of this paper lies in (a) conceptual explanation of the functioning of self-update and graph-based techniques to template adaptation leading to efficacy of the latter and (b) evaluation of the performance of these adaptation techniques in comparison to the baseline system without adaptation. Experiments are conducted on the large DIEE data set, explicitly collected for this aim. Reported results validate the superiority of the graph-based technique over self-update.

► Conceptual representation of the functioning of self-update and graph based min-cut technique to updating is proposed.
► The performance of these update techniques has been evaluated over time.
► Large DIEE face and fingerprint databases have been explicitly collected for this aim.
► Updating over time, graph based min-cut may result in stable performance.
► Self-update follows the error trend of the baseline classifier.