Dft - 2 D - Based Fast Superresolution

Recently we have successfully exploited and applied the new and powerful non-parametric Integrated Nested Laplace Approximation (INLA) Bayesian inference method to the problem of Superresolution (SR) image reconstruction, generating the INLA SR algorithm. Even though achieving superior image reconstruction results in comparison to other state-of-the-art methods, such algorithm manipulates matrices that are quite large, although sparse. Thus, even using data structures suitable to store and manipulate sparse matrices, the demand for memory usage and computation is high. In this paper, it is shown that the computation cost of the INLA SR algorithm may be reduced (from O(n4) to O(n log(n))), as well as the dimension of the matrices handled (from n2 × n2 to n× n, the size of the HR image), at the cost of a slight reduction of the HR image quality. Keywords— Superresolution, INLA, DFT, Block Toeplitz Circulant Matrices.