|
A successful gender classification approach can boost the
performance of many other applications including face recognition
and smart human-computer interfaces. Despite its importance, it
has received relatively little attention in the literature. Based on the type of features used, existing gender classification
approaches fall into one of two categories: geometry-based
and appearance-based. Geometry-based methods use metric features, e.g., face
width, face length, mouth size, eye size, distances, angles and
areas among salient feature points (eyes, nose, etc.). Appearance-based methods learn the decision boundary between the male and
female classes from training imagery without
extracting any geometrical features.
Most gender classification methods reported in the literature
use all the features extracted for classification purposes.
As a result, gender-irrelevant information might be fed to the
gender classifier which might not allow the classifier to
generalize nicely, especially when the training set is small.
Automatic feature subset selection distinguishes the proposed gender
classification method from all other reported approaches. In particular,
Genetic Algorithms are employed to select features that encode
important gender information and improve classification performance.
GAs belong to the class of
randomized heuristic search techniques, offering an attractive approach to
feature subset selection. Although they have been used in various pattern
recognition applications, their use in the area of computer vision is rather
limited.
|
