An image resampling method using combined directional kernels

Abstract

A new edge-directional image resampling method is proposed. The method uses a weighted sum of two adaptive 4x4 interpolation kernels to construct high-resolution pixels. The weights are chosen according to the local gradient features for each pixel. The interpolation kernels are learned using pairs of low- and high-resolution images taken from LIVE image database. The method has low complexity and low memory consumption. It outperforms existing fast edge-directional image interpolation methods.