Beyond Image Quality Comparison

Abstract

We solve the problem of quantitative measuring of image quality, beyond the widely used MSE (mean square error) or even visual comparison. We combine eight feature-based and perceptually-oriented image quality metrics. The need for this comes from virtual archaeology, where various image acquisitions, antialiasing, multiresolution, image reconstruction or texturing methods produce very similar images. The proposed evaluation framework is suitable for any image quality decision-making, being not restricted to virtual archaeology. In particular, we compare a representative database of visually indistiguishable image pairs from different cameras, anisotropic texture filters and various antialiasing methods. For image registration we propose a modified video processing step. The results support the selection beyond commonly used visual comparison.