Image resolution improvement: A signal processing approach

High-resolution (HR) images are in demand because they not only give the viewer a pleasing view of the scene but also offer additional details that are important for analysis. Signal processing approach can be used to improve the resolution from several sub pixel shifted low resolution images. The image to be super-resolved is estimated from a set of low resolution (LR) observations, using Least squares (LS) technique. Ill-posedness is addressed by Regularized Least Squares (RLS) technique which utilizes smoothness prior. In this paper, we first demonstrate the improved noise robustness of RLS approach over the LS. Also, to avoid over smoothness, we model the image to be super-resolved as Markov random field and estimate it from the observations by a Graduated non convexity optimization procedure. A discontinuity adaptive regularizer is used to preserve the edges in the reconstructed image. Several experimental results are provided to demonstrate the robustness of the proposed algorithm.