Filter-Bank Based Super-Resolution for Rotated and Blurry Undersampled Images

The ill-posed problem of image super-resolution (SR) is considered in the frequency domain using a filter bank approach. A low resolution (LR) imaging system is modeled in this paper using different fractional translation, rotation and blurs for each LR image. Synthesis filters are then found which provide a minimal time averaged mean squared error (TAMSE) reconstruction under a wide sense stationary (WSS) signal assumption. Presented simulations for many kinds of images demonstrate the effectiveness of the algorithm.