Ensemble local fractional LDA for face recognition

The classification performance of traditional LDA is often degraded by the fact that its separability criteria are not directly related to the classification accuracy in the output space. The fractional LDA (F-LDA) can solve the problem by more heavily weighting classes that are closer together. This paper proposes a novel face recognition method based on an ensemble of local F-LDA (ELF-LDA) classifiers. Firstly, an effective preprocessing scheme is employed incorporating a Logarithmic transformation and a local normalization procedure. Then the local block Gabor features are extracted by applying Gab or filters to each spatial block of preprocessed facial images. After that, multiple F-LDA classifiers are obtained on each local block of Gabor features. Finally, all the classifiers are fused to an ensemble classifier. The experimental results on CAS-PEAL-R1 face database show that our method significantly outperforms state-of-art face identification techniques. And it is noticeable that EFL-LDA obtains the best performance reported in the literature to the best of our knowledge.