L 2, 1-based regression and prediction accumulation across views for robust facial landmark detection

We propose a new methodology for facial landmark detection. Similar to other state-of-the-art methods, we rely on the use of cascaded regression to perform inference, and we use a feature representation that results from concatenating 66 HOG descriptors, one per landmark. However, we propose a novel...

Full description

Bibliographic Details
Main Authors: Martinez, Brais, Valstar, Michel F.
Format: Article
Published: Elsevier 2015
Subjects:
Online Access:https://eprints.nottingham.ac.uk/31304/
_version_ 1848794172438347776
author Martinez, Brais
Valstar, Michel F.
author_facet Martinez, Brais
Valstar, Michel F.
author_sort Martinez, Brais
building Nottingham Research Data Repository
collection Online Access
description We propose a new methodology for facial landmark detection. Similar to other state-of-the-art methods, we rely on the use of cascaded regression to perform inference, and we use a feature representation that results from concatenating 66 HOG descriptors, one per landmark. However, we propose a novel regression method that substitutes the commonly used Least Squares regressor. This new method makes use of the L2,1 norm, and it is designed to increase the robust- ness of the regressor to poor initialisations (e.g., due to large out of plane head poses) or partial occlusions. Furthermore, we propose to use multiple initialisations, consisting of both spatial translation and 4 head poses corresponding to different pan rotations. These estimates are aggregated into a single prediction in a robust manner. Both strategies are designed to improve the convergence behaviour of the algorithm, so that it can cope with the challenges of in-the- wild data. We further detail some important experimental details, and show extensive performance comparisons highlighting the performance improvement attained by the method proposed here.
first_indexed 2025-11-14T19:11:58Z
format Article
id nottingham-31304
institution University of Nottingham Malaysia Campus
institution_category Local University
last_indexed 2025-11-14T19:11:58Z
publishDate 2015
publisher Elsevier
recordtype eprints
repository_type Digital Repository
spelling nottingham-313042020-05-04T17:19:58Z https://eprints.nottingham.ac.uk/31304/ L 2, 1-based regression and prediction accumulation across views for robust facial landmark detection Martinez, Brais Valstar, Michel F. We propose a new methodology for facial landmark detection. Similar to other state-of-the-art methods, we rely on the use of cascaded regression to perform inference, and we use a feature representation that results from concatenating 66 HOG descriptors, one per landmark. However, we propose a novel regression method that substitutes the commonly used Least Squares regressor. This new method makes use of the L2,1 norm, and it is designed to increase the robust- ness of the regressor to poor initialisations (e.g., due to large out of plane head poses) or partial occlusions. Furthermore, we propose to use multiple initialisations, consisting of both spatial translation and 4 head poses corresponding to different pan rotations. These estimates are aggregated into a single prediction in a robust manner. Both strategies are designed to improve the convergence behaviour of the algorithm, so that it can cope with the challenges of in-the- wild data. We further detail some important experimental details, and show extensive performance comparisons highlighting the performance improvement attained by the method proposed here. Elsevier 2015-10-09 Article PeerReviewed Martinez, Brais and Valstar, Michel F. (2015) L 2, 1-based regression and prediction accumulation across views for robust facial landmark detection. Image and Vision Computing . ISSN 0262-8856 (In Press) Facial landmark detection Regression 300 W challenge http://www.sciencedirect.com/science/article/pii/S0262885615001092 doi:10.1016/j.imavis.2015.09.003 doi:10.1016/j.imavis.2015.09.003
spellingShingle Facial landmark detection
Regression
300 W challenge
Martinez, Brais
Valstar, Michel F.
L 2, 1-based regression and prediction accumulation across views for robust facial landmark detection
title L 2, 1-based regression and prediction accumulation across views for robust facial landmark detection
title_full L 2, 1-based regression and prediction accumulation across views for robust facial landmark detection
title_fullStr L 2, 1-based regression and prediction accumulation across views for robust facial landmark detection
title_full_unstemmed L 2, 1-based regression and prediction accumulation across views for robust facial landmark detection
title_short L 2, 1-based regression and prediction accumulation across views for robust facial landmark detection
title_sort l 2, 1-based regression and prediction accumulation across views for robust facial landmark detection
topic Facial landmark detection
Regression
300 W challenge
url https://eprints.nottingham.ac.uk/31304/
https://eprints.nottingham.ac.uk/31304/
https://eprints.nottingham.ac.uk/31304/