FACE RECOGNITION HYBRID FISHER LINEAR DISCRIMINANT ANALYSIS

摘要

Face recognition algorithms using the Fisher linear discriminant (FLD) approach currently suffers low generalization issues when tested against images that differ significantly from those contained in the training set or gallery. In this paper, a Hybrid-FLD (H-FLD) approach is proposed to mitigate this disadvantage. This technique divides the face into smaller sub-sections comprising of major facial components such as eyes, mouth and nose, consequently exploiting the global and local approaches of face recognition algorithms. The proposed method was tested on the ORL face database. Our experiment shows that H-FLD was able to achieve 99.2% success rate for correctly authenticating a probe from the gallery, an achievement 16.8% higher than the conventional FLD approach.