Comparison of EWPSNR and MOS on an Eye-tracking Labelled Video Dataset
Saman Zadtootaghaj; Hamed Ahmadi; Sebastian Möller
Perceptual video compression has been increasingly utilized to obtain higher compression gains by removing perceptual redundancies according to the Human Visual System (HVS). Due to non-linear complex mechanisms of the HVS, conventional quality metrics fail to assess the performance of perceptual video compressors. Therefore, subjective assessment is employed as the most reliable method for this purpose. However, it costs both time and money which consequently impedes perceptual compression development. There have been several attempts to develop perceptual objective metrics as an alternative to subjective assessments. Among them, Eye-tracking Weighted PSNR (EWPSNR) is believed to be the fairest where the gaze points are available. In this paper, we measure the correlation between EWPSNR and Mean Opinion Score (MOS) on an eye-tracking dataset to evaluate the performance of EWPSNR to predict the subjective quality.
The paper can be downloaded from the ISCA archive.