Predictor-corrector image interpolation

Abstract

A novel image interpolation methodology is proposed in this paper, called the predictor-corrector interpolation (PCI). Given a low-resolution (LR) image, our PCI scheme begins with the prediction stage, aiming to interpolate the LR-sized input image to a high-resolution (HR) image which is of the same size as the final interpolated image. In the subsequent correction stage, those salient pixels (e.g., edge pixels) of the predicted image are identified and then necessary corrections are made to them for further improving the image quality. To demonstrate the effectiveness of this PCI methodology, the sparse mixing estimator (SME) interpolation is selected as the predictor, and a modified version of the contrast-guided interpolation (CGI) is developed and exploited as the corrector. Hence, the proposed PCI algorithm is denoted as PCI(SME,HR-CGI), which shows a superior performance over a number of comparable state-of-the-art image interpolation algorithms in both objective and subjective image quality assessment.