Objective video quality assessment methods for Video assistant refereeing (VAR) Systems

This report describes the work and conclusions drawn after phase 4 in the project “Assessment methods for Video assistant refereeing (VAR) System”. The performance of six different video quality models have been evaluated, that were identified during phase 1, against the subjective video quality database that was created during phase 3. The results are slightly different for 1080p compared to 1080i. For 1080p the models VQM_VFD, SSIM and VMAF performs the best with Pearson Correlation Coefficients (PCC) above 0.9. For 1080i the PCC drops a bit overall and then VMAF and VQM_VFD are close in performance and performing the best. The overall performance for both formats VMAF an VQM_VFD stands out as the best models. In this comparison VQM_VFD has the added advantage to also be able to perform its own registration i.e. to fix any misalignment between the reference video and the distorted one