Biologically inspired framework for pose invariance in visual face recognition
Invariant recognition is the ability to recognize an object regardless of how it is transformed given that the transformation is allowable. A small change in image will cause a dramatic change in signals. Visual system needs to be able to perform recognition without being influenced by these signal...
Saved in:
Main Author: | |
---|---|
Format: | Thesis |
Published: |
2012
|
Subjects: | |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
id |
my-mmu-ep.5443 |
---|---|
record_format |
uketd_dc |
spelling |
my-mmu-ep.54432014-04-22T06:59:03Z Biologically inspired framework for pose invariance in visual face recognition 2012-02 Tay, Noel Nuo Wi TA Engineering (General). Civil engineering (General) Invariant recognition is the ability to recognize an object regardless of how it is transformed given that the transformation is allowable. A small change in image will cause a dramatic change in signals. Visual system needs to be able to perform recognition without being influenced by these signal variations. There are contradictory but justified different views on how invariance is achieved, yet the problem is still considered unsolved. For this research work, a biological-inspired framework is built upon these views and findings to achieve invariance in facial recognition. Biologically linked framework is preferred since human visual system is far superior compared to existing artificial visual systems, but little is known about it due to its complexity. Thus, it is the intention of the researcher to develop a framework based on biology that acts as a guide for further studies. Due to the broad scope of the field of biological visual system, the framework is built based on problems on 2D translation and scaling invariance and 3D pose invariance. Therefore, the motivation is to build a framework to provide insights into solving translation, scaling and 3D pose invariance problems. The framework is first built through the compilation of related literatures. It can be divided into early/primary and secondary visual stages. Primary visual stage models the visual pathway from retina to the striate cortex (V1), whereas the modelling of secondary visual stage is mainly based on current psychophysical evidences. Early visual stages employ simple and complex cells to provide translation and scaling invariance. Signals handled at this primary stage are information-rich but fragmented. Subsequent stages employ geometrical transformation to achieve 3D pose invariance, where the signals are integrated but in an abstract form. A link between the primary and secondary stage is established through appropriate dimension reduction which is biologically justified. 2012-02 Thesis http://shdl.mmu.edu.my/5443/ http://library.mmu.edu.my/diglib/onlinedb/dig_lib.php masters Multimedia University Faculty of Information Science and Technology |
institution |
Multimedia University |
collection |
MMU Institutional Repository |
topic |
TA Engineering (General) Civil engineering (General) |
spellingShingle |
TA Engineering (General) Civil engineering (General) Tay, Noel Nuo Wi Biologically inspired framework for pose invariance in visual face recognition |
description |
Invariant recognition is the ability to recognize an object regardless of how it is transformed given that the transformation is allowable. A small change in image will cause a dramatic change in signals. Visual system needs to be able to perform recognition without being influenced by these signal variations. There are contradictory but justified different views on how invariance is achieved, yet the problem is still considered unsolved. For this research work, a biological-inspired framework is built upon these views and findings to achieve invariance in facial recognition. Biologically linked framework is preferred since human visual system is far superior compared to existing artificial visual systems, but little is known about it due to its complexity. Thus, it is the intention of the researcher to develop a framework based on biology that acts as a guide for further studies. Due to the broad scope of the field of biological visual system, the framework is built based on problems on 2D translation and scaling invariance and 3D pose invariance. Therefore, the motivation is to build a framework to provide insights into solving translation, scaling and 3D pose invariance problems. The framework is first built through the compilation of related literatures. It can be divided into early/primary and secondary visual stages. Primary visual stage models the visual pathway from retina to the striate cortex (V1), whereas the modelling of secondary visual stage is mainly based on current psychophysical evidences. Early visual stages employ simple and complex cells to provide translation and scaling invariance. Signals handled at this primary stage are information-rich but fragmented. Subsequent stages employ geometrical transformation to achieve 3D pose invariance, where the signals are integrated but in an abstract form. A link between the primary and secondary stage is established through appropriate dimension reduction which is biologically justified. |
format |
Thesis |
qualification_level |
Master's degree |
author |
Tay, Noel Nuo Wi |
author_facet |
Tay, Noel Nuo Wi |
author_sort |
Tay, Noel Nuo Wi |
title |
Biologically inspired framework for pose invariance in visual face recognition |
title_short |
Biologically inspired framework for pose invariance in visual face recognition |
title_full |
Biologically inspired framework for pose invariance in visual face recognition |
title_fullStr |
Biologically inspired framework for pose invariance in visual face recognition |
title_full_unstemmed |
Biologically inspired framework for pose invariance in visual face recognition |
title_sort |
biologically inspired framework for pose invariance in visual face recognition |
granting_institution |
Multimedia University |
granting_department |
Faculty of Information Science and Technology |
publishDate |
2012 |
_version_ |
1747829575805566976 |