Class-Specific Feature Mixture (CSFM)
To get around the one class, one feature assumption,
CSFM assumes that each class is composed of sub-classes
represented by an additive mixture PDF. We assume
that class is composed of subclasses
, that have relative probabilities of
occurrence
and individual sub-class PDFs
.
The mixture PDF for is given by:
|
(12.5) |
where
If we assume that each sub-class has a different feature
(approximate sufficient statistic) to distinguish it from
a sub-class dependent reference hypothesis , we
apply (2.2) to get
|
(12.6) |
Note that each class PDF is represented by the same library of models.
The CSFM classifier is
|
(12.7) |
which may be interpreted as a
data-specific feature classifier because
for each data sample , the factor
has a dominant effect, effectively picking one feature
to classfy the sample.