Face super-resolution based on singular value decomposition

In this paper, a novel face image super-resolution approach based on singular value decomposition (SVD) is proposed. We prove that the singular values of an image at one resolution have approximately linear relationships with their counterparts at other resolutions. This makes the estimation of the singular values of the corresponding HR face images more reliable. From the signal-processing point of view, this can effectively preserve and reconstruct the dominant information in the HR face image. Interpolating the two other matrices obtained from the SVD of a LR face image does not change either the primary facial structure or the pattern of the face image. Furthermore, the mapping scheme for interpolating the matrices can be viewed as a “coarse-to-fine” estimation of HR face images, which uses the mapping matrices learned from the corresponding reference image pairs. Experimental results show that the proposed super-resolution scheme is effective and efficient.