Due to some problems linked
with the unimodal biometric data, such as small difference over the population,
large intra-variability over time, absence of biometric sample in fraction of a
population etc., the use of multimodal biometrics is a first choice explanation.
The main objective of a multimodal biometric system is to improve the acknowledgment
performance of the system and to make the system robust over the boundaries
associated with unimodal biometric systems. Over the years, several approaches
have been planned and developed for multimodal biometric verification system
with different biometric personality and with different fusion mechanisms.
Multimodal biometric systems use multiple sources of biometric
information, whereas information synthesis is indispensable for analysis,
indexing and repossession of such information. There are numbers of fusion
techniques for any meticulous information. Choosing apposite fusion techniques
for any specific information depends on the necessity of the application and
the performance of the fusion techniques proven by previous research. There is
a compromise in biometric literature that all various levels of multimodal
biometric information fall into two broad categories: before matching and after
matching fusion. Fusion before matching category contains sensor
level fusion and feature
level fusion, while fusion after matching contains match score
level fusion, rank level fusion and decision level fusion. A novel fusion mechanism
has been conventional recently in BT Lab is based on fuzzy logic fusion, and hence named a fuzzy
biometric fusion. Fuzzy biometric fusion can be employed either in
the initial stage, i.e. before matching occurred or in the latter stage, i. e.
after matching occurred.
Fusion
in this category integrates evidences before matching or assessment of data
samples against the user sample occurs. “By combining low level features it is promising
to achieve a more abstract or a more precise demonstration of the world”. Thus,
biometric sources at the earlier stage contain much more information than after
dispensation.
However,
the extra costs of storing raw data and additional difficulty in developing corresponding
methods do not make this approach quite practical.
Fusion
after-matching methods strengthen information obtained after individual
biometric matching or assessment is done. Most multimodal biometric systems
have been using these fusion methods as the information needed for fusion is
easily obtainable compared to fusion before matching methods. The matching
scores, the ranking list (sorted order) based on matching scores or the human
being biometric decision (Yes/No) can be used for fusion in this category.
In
this chapter, information fusion techniques applied in multimodal biometrics
area are discussed. Usually, the information originated from different sources
in a multimodal biometric system can be collective in senor level, feature pulling
out level, match score level, rank level, and conclusion level. Among all of
the fusion methods, senor fusion and feature pulling out level fusion
considered as the stage for combining raw data or the actual biometric data.
Match score, rank and decision level fusion methods combine processed data or
data obtained through some experimentations. There is also another novel fusion
method which is becoming highly popular: the fuzzy combination.
There are a number of challenges in this area, requiring further examination.
The first one is rooted in the choice of a fusion method, most appropriate for
the application domain. The decision is often made ad-hoc, or based on
non-essential constraints such as availability of the fusion module, low cost,
etc, instead of being made based on actual fit of the application area and the
method.
No comments:
Post a Comment