Edge-oriented two-step interpolation based on training set

Preserving the sharpness of edge structures is highly challenging to image interpolation. In this paper, we propose an edge-oriented two-step interpolation method that utilizes an edge training set. For edge interpolation, the LR edge map is converted into the HR edge map by using the training set. Then, an image is classified into smooth and edge regions using the HR edge map, and both regions are interpolated separately. For edge regions, adaptive edge-oriented interpolation is performed by using the detailed edge structures learned from training. The proposed method is extensively evaluated, and its performance is compared with the conventional edge-based methods. Experimental results show that the proposed method can not only reconstruct the missed edge information by the training set, but also significantly reduce blurring and jagging artifacts around edges by separately interpolating smooth and edge regions.