Image Enlargement by Applying Coordinate Rotation and Kernel Stretching to Interpolation Kernels

Conventional image interpolation methods such as the bilinear and bicubic algorithms determine the weights of the reference pixels based on the special distance between the supplementary pixel and the reference pixels. They suffer from some problems such as blurring and jagging around the edges since the edge structure is not considered in interpolation. In this paper, a coordinate rotation and kernel stretch strategy combined with the bilinear or bicubic algorithm is proposed to tackle these problems. For the edge regions, the spatial coordinate axes are rotated to the edge direction and the edge normal to calculate the distances between the supplementary pixel and the original reference pixels. The kernel function is also stretched along the estimated edge direction so that the weights of the reference pixels along the edge direction would be higher than the others to smooth the edge and prevent the jagging artifacts. An edge enhancement method is also proposed to further reduce the interpolation error and sharpen the edge. The experiment results demonstrate that conventional bilinear and bicubic image interpolation methods combined with the proposed strategy can greatly improve the quantitative and qualitative performance for image enlargement.