Graph-embedding discriminant analysis on Riemannian manifolds for visual recognition

Shirazi, Sareh, Alavi, Azadeh, Harandi, Mehrtash T. and Lovell, Brian C. (2013). Graph-embedding discriminant analysis on Riemannian manifolds for visual recognition. In Yun Fu and Yunqian Ma (Ed.), Graph Embedding for Pattern Analysis (pp. 157-176) New York, NY, USA: Springer. doi:10.1007/978-1-4614-4457-2

Author Shirazi, Sareh
Alavi, Azadeh
Harandi, Mehrtash T.
Lovell, Brian C.
Title of chapter Graph-embedding discriminant analysis on Riemannian manifolds for visual recognition
Title of book Graph Embedding for Pattern Analysis
Place of Publication New York, NY, USA
Publisher Springer
Publication Year 2013
Sub-type Research book chapter (original research)
DOI 10.1007/978-1-4614-4457-2
Open Access Status
ISBN 9781461444565
Editor Yun Fu
Yunqian Ma
Chapter number 7
Start page 157
End page 176
Total pages 20
Total chapters 10
Collection year 2014
Language eng
Formatted Abstract/Summary
Recently, several studies have utilised non-Euclidean geometry to address several computer vision problems including object tracking, characterising the diffusion of water molecules as in diffusion tensor imaging, face recognition, human re-identification, texture classification, pedestrian detection and action recognition...

Instead of using tangent spaces to do inference on manifolds, we propose to embed Riemannian manifolds into Reproducing Kernel Hilbert Spaces (RKHS). This in turn opens the door for employing many kernel-based machine learning algorithms. As such, we tackle the problem of Discriminant Analysis (DA) on Riemannian manifolds through RKHS space and propose a graph-based local DA that utilises both within-class and between-class similarity graphs to characterise intra-class compactness and inter-class separability, respectively. Our graph-based DA is inspired by findings in the Euclidean space that explain why the conventional formalism of DA is not optimal when data comprises outliers and multi-modal classes and contains outliers. Our experiments for several recognition problems show that considerable gains in discrimination accuracy can be obtained by exploiting the geometrical structure and local information on Riemannian manifolds.
Q-Index Code B1
Q-Index Status Confirmed Code
Institutional Status UQ

Version Filter Type
Citation counts: Scopus Citation Count Cited 0 times in Scopus Article
Google Scholar Search Google Scholar
Created: Thu, 09 Jan 2014, 14:48:42 EST by Azadeh Alavi on behalf of School of Information Technol and Elec Engineering