Super Resolution on The Centroid Reference Grid

Image super resolution (SR) techniques produce a high resolution (HR) image from a sequence of low resolution (LR) images. In all aforegoing SR methods, all of the LR images are registered relative to one of them (called reference image) and their samples are mapped on an up-scaled grid of this reference image (called reference grid (RG)). However, the use of this RG may cause the samples appears with a highly nonuniform distribution. Additionally, final precision and quality may be dependent on the selection of reference image. In this paper we propose a new RG which reduces the non-uniformity of sample distribution. This RG is produced by incorporating all ofthe images; therefore the final result is not dependent on the selection of reference image. We also propose a method to find this RG (will be referred as centroid RG). Moreover, this RG is unique

Keywords:

Super resolution , Reference grid , Image interpolation , Homography , Sample density function