Artifact-Free Image Upscaling : An Evaluation of an Artifact-Free Color Interpolation Algorithm with Respect to Visual Quality

Abstract: The vast number of different sizes and resolutions for digital displays today means that images regularly have to be upscaled or downscaled to fit the current screen. This scaling is almost always done using interpolation. This thesis evaluates an interpolation method presented by Jens Ogniewski in terms of how it compares to classic interpolation methods concerning visual quality and how it behaves in different color spaces. Both an objective and a subjective evaluation were conducted. The objective evaluation applied three different image quality measures to the upscaled images. The subjective evaluation consisted of an online test where a forced-choice pairwise-comparison methodology was employed. Additionally, the method was implemented as a plugin for the GIMP. The objective and subjective results show a relatively good correlation and suggest that the performance of Ogniewski’s algorithm depends on the image it is applied to. It works well for graphical images, like pixel art and game assets, but it can sometimes have a blocky appearance when applied to photo images. Furthermore, it is concluded that upscaling in the sRGB color space produces the visually best result of the compared color spaces.