Feature Extraction Based on Equalized ULBP for Face Recognition

The exiting local binary pattern (LBP) operators have two disadvantages: (1) They produce rather long histograms, which slow down the recognition speed especially on large-scale face database, (2) Under some circumstances, they miss the local structure as they don´t consider the effect of the central pixel. Aiming at these problems, we propose a novel feature extraction approach based on equalized uniform local binary pattern (EULBP) for face recognition. EULBP operator has two advantages: (1) It reduces significantly the histograms´ dimension by doubling one-dimension pattern, (2) It considers the effect of the central pixel, thus improving the discrimination ability. The proposed feature extraction approach has been evaluated and compared with the conventional LBP and ULBP on two databases, FERET and Yale. Furthermore, in order to test the robustness of the proposed method under the condition that the resolution of the face image is low, we have also carried out experiments on the two databases by reducing the image resolution. The experimental results show that the proposed method gives the highest recognition accuracy in both normal and low-resolution conditions.