Adaptive biometric systems

Template update and selection algorithms are aimed to make the biometric system “adaptive” to the intra-class variations of the input data.

Recently, novel solutions have been introduced in the form of template updating where the classifiers adapts itself to the intra-class variations based on learning methodologies like supervised or semi-supervised learning. Although these methods are promising, the state-of-art related to them is still in its infancy.

Our contributions to the state-of-the-art are as follows:

• Introduction of semi-supervised learning techniques to template update problems.
• A critical review of template update methods to biometrics. The aim is to highlight current state-of-the-art with respect to current approaches, in particular, the learning methodology adopted and experimental evaluation followed.
• Proposal of novel approaches to template update, namely, self update, graph-mincut and co-update ones.
• Replacement algorithms
• Investigation of pros and cons of several template update methods as function of the given user population.

People working on this topic:

• Luca Didaci
• Biagio Freni
• Gian Luca Marcialis
• Ajita Rattani
• Fabio Roli