Visual saliency aided High Dynamic Range (HDR) video quality metrics

Abstract

As High Dynamic Range (HDR) video is emerging as the next revolution in digital entertainment, finding effective ways for measuring its visual quality is of paramount importance. In this paper, we utilize the saliency information derived from an HDR Visual Attention Model (VAM), called LBVS-HDR, for assessing the quality of HDR video content. To this end, this saliency information is incorporated into existing state-of-the-art HDR quality metrics such as the HDR-VDP-2, deltaE2000, mPSNR, and tPSNR as well as the Standard Dynamic Range (SDR) benchmark quality metric, PSNR. The Visual Information Fidelity (VIF) index is also included in our comparisons, as it is reported to perform well for HDR content. Comparing the results of the VAM-aided quality metrics with those of the original ones, we verified that, in general, using saliency prediction for HDR quality assessment improves the performance of all the existing quality metrics. We also observed that the VIF index achieves the highest correlation between the objective and subjective test results.