Sparse representations of image gradient orientations for visual recognition and tracking

Recent results [18] have shown that sparse linear representations of a query object with respect to an overcomplete basis formed by the entire gallery of objects of interest can result in powerful image-based object recognition schemes. In this paper, we propose a framework for visual recognition an...

Full description

Bibliographic Details
Main Authors:	Tzimiropoulos, Georgios, Zafeiriou, Stefanos, Pantic, Maja
Format:	Conference or Workshop Item
Language:	English
Published:	2011
Online Access:	http://eprints.nottingham.ac.uk/31413/ http://eprints.nottingham.ac.uk/31413/ http://eprints.nottingham.ac.uk/31413/1/tzimiroCVPRW11B.pdf

id	nottingham-31413
recordtype	eprints
spelling	nottingham-314132017-10-18T17:11:08Z http://eprints.nottingham.ac.uk/31413/ Sparse representations of image gradient orientations for visual recognition and tracking Tzimiropoulos, Georgios Zafeiriou, Stefanos Pantic, Maja Recent results [18] have shown that sparse linear representations of a query object with respect to an overcomplete basis formed by the entire gallery of objects of interest can result in powerful image-based object recognition schemes. In this paper, we propose a framework for visual recognition and tracking based on sparse representations of image gradient orientations. We show that minimal `1 solutions to problems formulated with gradient orientations can be used for fast and robust object recognition even for probe objects corrupted by outliers. These solutions are obtained without the need for solving the extended problem considered in [18]. We further show that low-dimensional embeddings generated from gradient orientations perform equally well even when probe objects are corrupted by outliers, which, in turn, results in huge computational savings. We demonstrate experimentally that, compared to the baseline method in [18], our formulation results in better recognition rates without the need for block processing and even with smaller number of training samples. Finally, based on our results, we also propose a robust and efficient `1-based “tracking by detection” algorithm. We show experimentally that our tracker outperforms a recently proposed `1-based tracking algorithm in terms of robustness, accuracy and speed. 2011 Conference or Workshop Item PeerReviewed application/pdf en http://eprints.nottingham.ac.uk/31413/1/tzimiroCVPRW11B.pdf Tzimiropoulos, Georgios and Zafeiriou, Stefanos and Pantic, Maja (2011) Sparse representations of image gradient orientations for visual recognition and tracking. In: 2011 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 20-25 June 2011, Colorado Springs, USA. http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=5981809
repository_type	Digital Repository
institution_category	Local University
institution	University of Nottingham Malaysia Campus
building	Nottingham Research Data Repository
collection	Online Access
language	English
description	Recent results [18] have shown that sparse linear representations of a query object with respect to an overcomplete basis formed by the entire gallery of objects of interest can result in powerful image-based object recognition schemes. In this paper, we propose a framework for visual recognition and tracking based on sparse representations of image gradient orientations. We show that minimal `1 solutions to problems formulated with gradient orientations can be used for fast and robust object recognition even for probe objects corrupted by outliers. These solutions are obtained without the need for solving the extended problem considered in [18]. We further show that low-dimensional embeddings generated from gradient orientations perform equally well even when probe objects are corrupted by outliers, which, in turn, results in huge computational savings. We demonstrate experimentally that, compared to the baseline method in [18], our formulation results in better recognition rates without the need for block processing and even with smaller number of training samples. Finally, based on our results, we also propose a robust and efficient `1-based “tracking by detection” algorithm. We show experimentally that our tracker outperforms a recently proposed `1-based tracking algorithm in terms of robustness, accuracy and speed.
format	Conference or Workshop Item
author	Tzimiropoulos, Georgios Zafeiriou, Stefanos Pantic, Maja
spellingShingle	Tzimiropoulos, Georgios Zafeiriou, Stefanos Pantic, Maja Sparse representations of image gradient orientations for visual recognition and tracking
author_facet	Tzimiropoulos, Georgios Zafeiriou, Stefanos Pantic, Maja
author_sort	Tzimiropoulos, Georgios
title	Sparse representations of image gradient orientations for visual recognition and tracking
title_short	Sparse representations of image gradient orientations for visual recognition and tracking
title_full	Sparse representations of image gradient orientations for visual recognition and tracking
title_fullStr	Sparse representations of image gradient orientations for visual recognition and tracking
title_full_unstemmed	Sparse representations of image gradient orientations for visual recognition and tracking
title_sort	sparse representations of image gradient orientations for visual recognition and tracking
publishDate	2011
url	http://eprints.nottingham.ac.uk/31413/ http://eprints.nottingham.ac.uk/31413/ http://eprints.nottingham.ac.uk/31413/1/tzimiroCVPRW11B.pdf
first_indexed	2018-09-06T12:08:35Z
last_indexed	2018-09-06T12:08:35Z
_version_	1610859901858349056

Sparse representations of image gradient orientations for visual recognition and tracking

Similar Items