Adaptive image rational upscaling with local structure as constraints

In this paper, we develop a new interpolation fusion model, Adaptive Image Rational Upscaling (AIRU), based on classical rational interpolation. This model can synthetically consider the influence of the surrounding 12 pixels within the current interpolation cell. Considering the limitation of edge direction estimation of conventional edge detection methods, we introduce a new method to quantify the edge direction based on the Principal Component Edge (PCE). Adaptive weights for each triangular patch can be generated based on three coefficients: angle coefficient which can be estimated by PCE, variation coefficient and gray similarity coefficient. PCE can also be used to divide the image into non-smooth and smooth area. AIRU and conventional interpolation are used in these two areas respectively. Furthermore, the model parameter optimization can further improve the interpolation performance. Experimental results demonstrate that the proposed fusion model achieves competitive performance when compared with the state-of-the-arts.