Hidden influences on image quality when comparing interpolation methods

When comparing different interpolation methods, common approach is to use an objective image quality measure that evaluates the numerical errors between the original image and interpolated image. The problem with this approach is impossibility to have original image in two or more different resolutions. Common method is to start with an original image, calculate a lower resolution version of original image (downscaling), and then use different interpolation methods to magnify low resolution image (upscaling). After that original and magnified interpolated images are compared to evaluate difference between them. In our research we found that downscaling process has great influence on results of upscaling. This influence is not properly addressed in references that deal with image interpolation methods. Therefore in this paper we will try to examine how the process of achieving lower resolution image affects picture quality of interpolated images. Our results show that comparison of image interpolation methods depends on downscaling technique, image contents and quality metric.