A new pooling strategy for image quality metrics: Five number summary

Abstract

Different pooling strategies for image quality metrics have been proposed over the years. Simple spatial averaging is one of the widely used methods. However, given the fact that a distortion may not be uniformly distributed over the image space, we argue that mean is not a good representation for the image quality distribution. We propose an alternative pooling strategy based on the five number summary. The performance of this pooling strategy has been tested on TID2008, TID2013 and LIVE databases and is shown to produce improvements in correlation with subjective quality assessment.