Fast resolution aware model fitting for noisy low resoluton image

Abstract

Increasing the resolution of images by synthesis may cause errorneous results due to the assymetry problem. As stated in assymetry problem, models could not fit the lower resolution inputs due to increased interpolation and noise. In order to avoid this problem RAF (Resolution Aware Fitting) algorithm has been proposed. In this approach no processing has been performed on low resolution (LR) input, and the intermediate high resolution (HR) synthesis is warped and deformed. Though convergence to the real synthesis parameters is increased, ignorance of the observation noise and warping intermediate HR syntheses undermine the robustness and effectiveness of the algorithm. In this work a new synthesis-based approach is proposed in which again the assymetry problem is avoided. In addition to that the observation noise is incorporated into the model, and the HR warping operation is approximated in reduced dimesions. Thus increase in robustness and serious saving in computational load is obtained.