Information Technology Reference
In-Depth Information
Chapter 13
Improving the Recognition Performance
of Moment Features by Selection
George A. Papakostas
Abstract
This chapter deals with the selection of the most appropriate moment
features used to recognize known patterns. This chapter aims to highlight the need
for selection of moment features subject to their descriptive capabilities. For this
purpose, some popular moment families are presented and their properties, making
them suitable for pattern recognition tasks, are discussed. Two different types of
feature selection algorithms, a simple Genetic Algorithm (GA) and the Relief algo-
rithm are applied to select the moment features that better discriminate human faces
and facial expressions, under several pose and illumination conditions. Appropriate
experiments using four benchmark datasets have been conducted in order to inves-
tigate the theoretical assertions. An extensive experimental analysis has shown that
the recognition performance of the moment features can be significantly improved
by selecting them from a predefined pool, relative to a specific application.
·
·
·
Keywords
Moment descriptors
Pattern recognition
Feature selection
Genetic
·
algorithms
Relief algorithm
13.1 Introduction
Nowadays, many advanced intelligent systems take part into humans' daily life
helping them to satisfy possible professional or entertainment needs. Thus, advanced
human computer/machine interaction [
3
], human identity authentication [
10
], bio-
metric authentication [
30
] and surveillance systems [
28
], have been developed and
proposed. Such systems mainly consist of a pattern recognition procedure, which
enables the system to interact with the surrounding environment.
Search WWH ::
Custom Search