Effective coverage as a new metric for image quality assessment databases comparison

Abstract

In the paper a new metric for comparing image quality assessment databases is presented. Its purpose is to show the real coverage of the database in Spatial information vs. Colorfulness space. Currently used metrics are vulnerable to databases having large uncovered areas within convex hull. Proposed method takes into account existing metrics but provides additional fill rate parameter, which define how well the area is utilized. As an added value an overview of the databases with color images for quality assessment is done. It covers 17 publicly available databases having over 5000 pristine images in total. The overview contains general information about databases as well as calculation of different metrics showing the variety of images included in different sources.