Description

We propose a novel technique to predetermine large number of neighbors for SR reconstruction. We extend robust SLO algorithm to simultaneously find neighbor and to solve reconstruction weights. We implement K-NN classifier trained by K-Means clusters. The LR (Low Resolution) patches are classified using K-NN and it is simultaneously converted to SR (Super Resolution) patch through robust SLO algorithm. Finally we synthesize and fuse SR patches and subject it to de-blurring algorithm to obtain a Super-Resolution Image.