A Fast L-curvature Estimation for Super-Resolution Image Restoration

Abstract

L-curvature can be used to evaluate regularization parameters in super-resolution image restoration. L-curvature is computed expensively through Arnoldi process. The paper proposes an efficient approximate method based on the incomplete orthogonalization Arnoldi Process. The method uses a few previous orthogonal vectors for Arnoldi process. The others are not needed in the process and may be discarded. Hessenberg matrix obtained from the incomplete orthogonalization process has a band structure different from the previous similar upper triangular matrix. The computational complexity of the L-curvature may be dropped. Experimental results with synthetic and real image sequences are presented that demonstrate the effectiveness and robustness of the method.