Iterative edge-directed image upscaling with high directionality
Eri Hosogai, Yuichi Tanaka · 2014
Variety of image upscaling methods has been proposed and used in computer tools due to the increasing number of practical applications. The essential objectives of image upscaling are: it should be fast, natural to human perception and does not yield annoying artifacts. This paper proposes an edge-directed iterative image interpolation based on the second-order derivatives along multiple directions. Our method uses the assumption that the second-order derivatives of the image intensities are continuous along the interpolation directions. We first initialize the pixel values based on a grid filling and iteratively modify the interpolated pixels by minimizing an objective function depending on the second-order directional derivatives. Especially our method provides better results in edge regions while the computation time is reduced compared to other existing upscaling methods. Image coding with the proposed interpolation method is also performed as a possible application.