Hybrid image interpolation with soft-decision kernel regression

Parametric linear autoregressive (AR) model has been widely used in image processing but is known to induce unstable results. The recently emerged nonparametric kernel regression is an effective structural method for forestalling outliers but often brings over-smoothed output. This paper introduces a hybrid algorithm for image interpolation through combining the strength of parametric and nonparametric modeling techniques. More specifically, it is a soft-decision kernel regression (SKR) method in which the soft-decision AR model is embedded into the adaptive kernel regression framework. Compared with the state-of-the-art interpolation methods, simulation results show that the proposed SKR algorithm achieves comparative or better results in terms of objective and subjective quality.