Precise eye localization using HOG descriptors

In this paper, we present a novel algorithm for precise eye detection. First, a couple of AdaBoost classifiers trained with Haar-like features are used to preselect possible eye locations. Then, a Support Vector Machine machine that uses Histograms of Oriented Gradients descriptors is used to obtain...

ver descrição completa

Detalhes bibliográficos
Autores: Monzó Ferrer, David, Albiol Colomer, Alberto|||0000-0002-1970-3289, Sastre, Jorge|||0000-0002-8612-6717, Albiol Colomer, Antonio José|||0000-0002-0679-912X
Formato: artículo
Fecha de publicación:2011
País:España
Recursos:Universitat Politècnica de València (UPV)
Repositorio:RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Idioma:inglés
OAI Identifier:oai:riunet.upv.es:10251/58789
Acesso em linha:https://riunet.upv.es/handle/10251/58789
Access Level:acceso abierto
Palavra-chave:AdaBoost
Eye detection
HOG
Local feature descriptors
Commercial software
Descriptors
Eye localization
Eye location
Haar-like features
Local feature
Novel algorithm
Pre-selected
Algorithms
Eye protection
Adaptive boosting
TEORIA DE LA SEÑAL Y COMUNICACIONES
Descrição
Resumo:In this paper, we present a novel algorithm for precise eye detection. First, a couple of AdaBoost classifiers trained with Haar-like features are used to preselect possible eye locations. Then, a Support Vector Machine machine that uses Histograms of Oriented Gradients descriptors is used to obtain the best pair of eyes among all possible combinations of preselected eyes. Finally, we compare the eye detection results with three state-of-the-art works and a commercial software. The results show that our algorithm achieves the highest accuracy on the FERET and FRGCv1 databases, which is the most complete comparative presented so far. © Springer-Verlag 2010.