Robust Dual Images Super-resolution

With the developing of 3D mobile phone, there are more and more dual images which seen from slightly different angles caught by two mobile phone cameras. The images caught by mobile phone camera usually have lower resolution (LR) than the images caught by normal camera. To solve this problem, this paper proposes a robust Bayesian SR method to recover high-frequency information from dual LR images, which solves the registration, interpolation, kernel estimation, restoration, and post-processing problems simultaneously. It's noticed that the blur kernel (Gaussian blur kernel) of the proposed SR processing is recalculated in iterations, other than a constant kernel as usual. Our experimental results show great visual improvement and a suppressing of high-frequency artifacts.