Face Recognition Technique in Transform Domains

摘要

In a recently published article, a new discriminative sparse representation method for robust face identification via ℓ2 regularization (SRFI) was presented. In our previous works, coefficients from Two-Dimensional Discrete Cosine Transform (2D DCT), 2D Discrete Wavelet Transform (2D DWT), were employed individually or combined to implement face identification systems. In this paper, a mixed SRFI (MSRFI) system is proposed by utilizing weight-based selected coefficients from the two non-orthogonal domains, i.e., 2D DCT and 2D DWT. The use of such mix as an input to the MSRFI maintains the high recognition accuracy of the SRFI while remarkably reducing the storage requirements, and the computational complexity. By referring to our previous works results and to prove the improved features of the MSRFI, extensive simulations were implemented on two face datasets, namely, ORL, and YALE.