Image interpolation based on nonlocal self-similarity

Abstract

This paper presents an efficient method for the resolution enhancement of low resolution (LR) images. Inspired by the great success of nonlocal filtering, the proposed method interpolates a missing pixel by a nonlocal estimation which takes into account the self-similarity of images and the geometric duality between LR images and high resolution (HR) images. This method firstly exploits the SAI method to generate an initial HR image of the given LR image. Then the initial HR image is further improved by a nonlocal estimation that uses the given LR image as a guiding image. Pixel correction is applied to reduce the distortion of the HR estimate. Experimental results demonstrate that the proposed method provides competitive performance.