Genetic Programming for Object Detection: Improving
Fitness Functions and Optimising Training Data
M. Zhang, and M. Lett. The IEEE Intelligent Informatics Bulletin, 7 (1):
12--21(December 2006)
Abstract
This paper describes an approach to the improvement of
a fitness function and the optimisation of training
data in genetic programming (GP) for object detection
particularly object localisation problems. The fitness
function uses the weighted F-measure of a genetic
program and considers the localisation fitness values
of the detected object locations. To investigate the
training data with this fitness function, we categorise
the training data into four types: exact centre, close
to centre, include centre, and background. The approach
is examined and compared with an existing fitness
function on three object detection problems of
increasing difficulty. The results suggest that the new
fitness function outperforms the old one by producing
far fewer false alarms and spending much less training
time and that the first two types of the training
examples contain most of the useful information for
object detection. The results also suggest that the
complete background type of data can be removed from
the training set.
%0 Journal Article
%1 Zhang:2006:IIIB
%A Zhang, Mengjie
%A Lett, Malcolm
%D 2006
%J The IEEE Intelligent Informatics Bulletin
%K algorithms, classification, computing, data detection, evolutionary fitness function, genetic localisation, object programming, recognition, training
%N 1
%P 12--21
%T Genetic Programming for Object Detection: Improving
Fitness Functions and Optimising Training Data
%U http://www.comp.hkbu.edu.hk/~cib/2006/Dec/iib_vol7no1_article2.pdf
%V 7
%X This paper describes an approach to the improvement of
a fitness function and the optimisation of training
data in genetic programming (GP) for object detection
particularly object localisation problems. The fitness
function uses the weighted F-measure of a genetic
program and considers the localisation fitness values
of the detected object locations. To investigate the
training data with this fitness function, we categorise
the training data into four types: exact centre, close
to centre, include centre, and background. The approach
is examined and compared with an existing fitness
function on three object detection problems of
increasing difficulty. The results suggest that the new
fitness function outperforms the old one by producing
far fewer false alarms and spending much less training
time and that the first two types of the training
examples contain most of the useful information for
object detection. The results also suggest that the
complete background type of data can be removed from
the training set.
@article{Zhang:2006:IIIB,
abstract = {This paper describes an approach to the improvement of
a fitness function and the optimisation of training
data in genetic programming (GP) for object detection
particularly object localisation problems. The fitness
function uses the weighted F-measure of a genetic
program and considers the localisation fitness values
of the detected object locations. To investigate the
training data with this fitness function, we categorise
the training data into four types: exact centre, close
to centre, include centre, and background. The approach
is examined and compared with an existing fitness
function on three object detection problems of
increasing difficulty. The results suggest that the new
fitness function outperforms the old one by producing
far fewer false alarms and spending much less training
time and that the first two types of the training
examples contain most of the useful information for
object detection. The results also suggest that the
complete background type of data can be removed from
the training set.},
added-at = {2008-06-19T17:46:40.000+0200},
author = {Zhang, Mengjie and Lett, Malcolm},
biburl = {https://www.bibsonomy.org/bibtex/267b9903b95c75243fef6baea2f2e2e8f/brazovayeye},
interhash = {ea5dc4e1ed9d86e2c8a19356f5ca5edd},
intrahash = {67b9903b95c75243fef6baea2f2e2e8f},
issn = {1727-5997},
journal = {The IEEE Intelligent Informatics Bulletin},
keywords = {algorithms, classification, computing, data detection, evolutionary fitness function, genetic localisation, object programming, recognition, training},
month = {December},
notes = {formerly IEEE Computational Intelligence Bulletin)},
number = 1,
pages = {12--21},
size = {10 pages},
timestamp = {2008-06-19T17:55:44.000+0200},
title = {Genetic Programming for Object Detection: Improving
Fitness Functions and Optimising Training Data},
url = {http://www.comp.hkbu.edu.hk/~cib/2006/Dec/iib_vol7no1_article2.pdf},
volume = 7,
year = 2006
}