Fast multi-scale local phase quantization histogram for face recognition

Multi-scale local phase quantization (MLPQ) is an effective face descriptor for face recognition. In previous work, MLPQ is computed by using Short-term Fourier Transformation (SFT) in local regions and the high-dimension histogram based features are extracted for face representation. This paper tries to improve MLPQ based face recognition in terms of accuracy and efficiency. It has two main contributions. First, a fast MLPQ extraction algorithm is proposed which produces the same results with original MLPQ method but is about three times faster than the original one in practice. Second, a novel feature selection method combining Adaboost and regression is proposed to select the most discriminative and suitable features for the subsequent subspace learning. Experiments on FERET and FRGC ver 2.0 databases validate the effectiveness and efficiency of the proposed method.

► A fast multi-scale LPQ (MLPQ) extraction method is proposed.
► A suitable feature selection method for subspace learning by using Adaboost and linear regression is presented.
► Extensive experiments validate the effectiveness and efficiency of the proposed method.