Discriminant Improved Local Tangent Space Alignment with adaptively weighted complex wavelet for face recognition

Improved Local Tangent Space Alignment (ILTSA) is a recent nonlinear dimensionality reduction method but there exists the out-of-sample problem. In this paper, based on linearization and discriminant extension of ILTSA, a novel feature extraction method named Discriminant Improved Local Tangent Space Alignment (DILTSA) is proposed. DILTSA can preserve both local within-class and between-class geometry structures. Motivated by the recent development of sub-pattern face recognition, an adaptively weighted complex wavelet feature extraction method is proposed. Experimental results on ORL and PIE face databases demonstrate the effectiveness of DILTSA and its combination with complex wavelet features.