Statistical evaluation of no-reference image visual quality metrics
Nikolay N. Ponomarenko, Олег Игоревич Еремеев, Владимир Васильевич Лукин, Karen Egiazarian · 2010
A task of no-reference visual quality metric verification is considered. A test set that contains 500 JPEG format images having different distortions is created. The results of experiments carried out by 316 volunteer observers to evaluate visual quality of images are presented. The experiments allowed obtaining mean opinion scores (MOS) based on averaging the evaluations. Several non-reference image visual quality metrics have been analyzed. Spearman and Kendall correlations between MOS and metrics values are calculated. It is shown that all analyzed metrics have not enough correspondence to human perception.