22-06-2012, 01:08 PM
NABS: Novel Approaches for Biometric Systems
INTRODUCTION
IDENTITY checking through fast and effective automatic
systems relates both to longer investigated applications, such
as controlling limited access areas or locating dangerous individuals,
and to the growing diffusion of electronic transactions.
In biometric systems, recognition relies on physical, e.g., fingerprints
or behavioral characteristics. The latter are gaining increasing
attention due to their better reaction to spoofing, though
suffering from low repeatability [37], [45]. At present, even the
measurement of personality traits is being investigated [6]. Two
operation modes exist verification (one-to-one matching to decide
if the query subject actually corresponds to the claimed
identity) and identification (one-to-many matching, to decide
which is the identity corresponding to the query) [19].
PROPOSED ARCHITECTURES
One of the characterizing aspects of a multibiometric architecture
is the stage at which fusion is implemented. We do not
give here a complete survey about existing methods. However,
it is worth introducing some observations to support the implemented
design choices. Fusion at feature level is quite complex
and also rigid, since it usually implies to exactly knowin advance
both the involved biometries and their peculiarities. Updating
such kind of system is quite hard.Anumber ofworks in literature
aim to compare match score level (or simply score level) fusion
techniques with rank level ones. It can be deduced from them
that it is not possible to decide which of these two approaches is
the very optimal one: the choice is tightly bound to the working
context where fusion is performed.
SYSTEM RESPONSE RELIABILITY
The fusion of the results from different classifiers can be
considered as one of the most significant advances in pattern
classification in recent years [21]. On the other hand, subsystems
might not be equally reliable, due to a possibly different
accuracy of their procedures, and not all responses by a single
subsystem might be equally reliable, due to a possibly different
quality of input from time to time. An unreliable response,
rather than producing a rejection, may represent a valid reason
for a further check. A measure for response reliability is crucial
for fusing the single results.
CONCLUSION
Recent research in the biometric field has been very intense,
and a high number of physical and behavioral features have been
studied. Some of them are more reliable than others, yet none
is free from limitations. For this reason, the present trend is to
combine more biometries in one system.We described twomultimodal
architectures: the hierarchical protocol, already known
in literature, and the NCTP, introduced here.We also addressed
the problem of data normalization and proposed the new QLS
function, which overcomes the limits of the present ones.
INTRODUCTION
IDENTITY checking through fast and effective automatic
systems relates both to longer investigated applications, such
as controlling limited access areas or locating dangerous individuals,
and to the growing diffusion of electronic transactions.
In biometric systems, recognition relies on physical, e.g., fingerprints
or behavioral characteristics. The latter are gaining increasing
attention due to their better reaction to spoofing, though
suffering from low repeatability [37], [45]. At present, even the
measurement of personality traits is being investigated [6]. Two
operation modes exist verification (one-to-one matching to decide
if the query subject actually corresponds to the claimed
identity) and identification (one-to-many matching, to decide
which is the identity corresponding to the query) [19].
PROPOSED ARCHITECTURES
One of the characterizing aspects of a multibiometric architecture
is the stage at which fusion is implemented. We do not
give here a complete survey about existing methods. However,
it is worth introducing some observations to support the implemented
design choices. Fusion at feature level is quite complex
and also rigid, since it usually implies to exactly knowin advance
both the involved biometries and their peculiarities. Updating
such kind of system is quite hard.Anumber ofworks in literature
aim to compare match score level (or simply score level) fusion
techniques with rank level ones. It can be deduced from them
that it is not possible to decide which of these two approaches is
the very optimal one: the choice is tightly bound to the working
context where fusion is performed.
SYSTEM RESPONSE RELIABILITY
The fusion of the results from different classifiers can be
considered as one of the most significant advances in pattern
classification in recent years [21]. On the other hand, subsystems
might not be equally reliable, due to a possibly different
accuracy of their procedures, and not all responses by a single
subsystem might be equally reliable, due to a possibly different
quality of input from time to time. An unreliable response,
rather than producing a rejection, may represent a valid reason
for a further check. A measure for response reliability is crucial
for fusing the single results.
CONCLUSION
Recent research in the biometric field has been very intense,
and a high number of physical and behavioral features have been
studied. Some of them are more reliable than others, yet none
is free from limitations. For this reason, the present trend is to
combine more biometries in one system.We described twomultimodal
architectures: the hierarchical protocol, already known
in literature, and the NCTP, introduced here.We also addressed
the problem of data normalization and proposed the new QLS
function, which overcomes the limits of the present ones.