Lighting variation is a major challenge for an automatic face recognition system. In order to overcome this problem, many methods have been proposed. Most of them try to extract features invariant to illumination changes or to reduce illumination changes in a pre-processing step and to extract features for recognition. In this paper, we present a procedure similar to the latter where the two steps are complementary. In the pre-processing step we deal with the illumination changes and in the features extraction step we use the BSIF (Binarized Statistical Image Features), a recently proposed textural algorithm. In our opinion, a method capable of reducing the lighting variations is ideal for an algorithm like the BSIF. The performance of our system has been tested on the FRGC dataset and the presented results show the validity of our approach.
On combining edge detection methods for improving BSIF based facial recognition performances / Tuveri, P.; Ghiani, L.; Abukmeil, M.; Marcialis, G. L.. - 9756:(2016), pp. 108-116. (Intervento presentato al convegno 9th International Conference on Articulated Motion and Deformable Objects, AMDO 2016 tenutosi a esp nel 2016) [10.1007/978-3-319-41778-3_11].
On combining edge detection methods for improving BSIF based facial recognition performances
Ghiani L.;
2016-01-01
Abstract
Lighting variation is a major challenge for an automatic face recognition system. In order to overcome this problem, many methods have been proposed. Most of them try to extract features invariant to illumination changes or to reduce illumination changes in a pre-processing step and to extract features for recognition. In this paper, we present a procedure similar to the latter where the two steps are complementary. In the pre-processing step we deal with the illumination changes and in the features extraction step we use the BSIF (Binarized Statistical Image Features), a recently proposed textural algorithm. In our opinion, a method capable of reducing the lighting variations is ideal for an algorithm like the BSIF. The performance of our system has been tested on the FRGC dataset and the presented results show the validity of our approach.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
