2011 | OriginalPaper | Chapter
Automatic Template Labeling in Extensible Multiagent Biometric Systems
Authors : Maria De Marsico, Michele Nappi, Daniel Riccio, Genny Tortora
Published in: Image Analysis and Processing – ICIAP 2011
Publisher: Springer Berlin Heidelberg
Activate our intelligent search to find suitable subject content or patents.
Select sections of text to find matching patents with Artificial Intelligence. powered by
Select sections of text to find additional relevant content using AI-assisted search. powered by
Many works in literature have demonstrated the superiority of multibiometric systems compared to single-biometrics ones, in terms of both accuracy and robustness. Most current multibiometric systems implement a static architecture, which does not change in time. However, the ability to progressively add more modules, either to process more biometrics or to exploit additional algorithms, might contribute to further enhance recognition performance. The addition of a new module (agent) to an already fully operational multiagent system usually requires its preliminary setup and training. In particular, it must be provided with a brand-new gallery, whose templates are suitably labeled according to the represented identities; alternatively, an existing database of templates, formerly built according to the suited feature extraction procedure, might be updated to include better quality items. It would be of paramount importance if the new agent can “inherit” the “experience that was already acquired by the other agents, including the creation of its gallery without having to undergo a full enrolling phase in its turn. We present here an algorithm to align a new module to the already existing ones in an automatic and unsupervised way. Experimental results show that our algorithm is effective both when the new database must be created from scratch (sample labeling), as well as when it is pre-existing and must be updated (sample updating). The latter operation can also be iteratively performed in running modules to dynamically update their galleries. In particular, we present here results achieved for face recognition.