A fast noise-robust interpolation method based on second-order directional-derivatives

It is a challenging work to reproduce the high resolution (HR) image from a noisy low resolution input while preserving its edge structures. In this paper, a fast noise-robust interpolation method is proposed. In the proposed method, the edge information of a pixel to be interpolated is first estimated using a local curvature (LC), which is a second-order directional-derivative obtained from its local neighborhoods. Based on the edge information of the pixel, edge-adaptive interpolation with noise reduction is performed using the proposed LC adaptive filter whose kernel is adaptively determined by comparing the LC values along the two orthogonal directions. A refinement procedure is adopted to further enhance the edge information of the HR image by applying a Laplacian subtraction method using the precomputed LC values. Experimental results show that the proposed method can preserve the edge sharpness while suppressing noise, with low computational complexity.