Fusing Shearlets and LBP Feature Sets for Face Recognition

To aim at the challenge of face recognition to uncontrolled situations, robust face recognition system requires to take into account different kinds of face appearance feature. However, most existing methods only use features of just one type. We show that integrating two of global and local representations, Shearlets features and local binary pattern (LBP), which gets better performance than either alone. Shearlets features not only utilize scale and position information of different scales of decomposed image, but also use directional information. Shearlets features primary capture facial global attributes while LBP encode small local appearance details. Both feature sets are high dimensional so it is beneficial to apply block-based fisher linear discriminant (BFLD) and PCA to reduce dimensionality prior to normalization and integration. Then low dimensional Shearlets and LBP sets are combined by score level fusion. the proposed method is evaluated on two challenge face databases including MPIE and FERET with promising results.