Video Compression

VideoNerd

Taken from the paper “Study of Subjective and Objective Quality Assessment of Video”, Kalpana Seshadrinathan et al.

Notes:

  • The paper was published at 2009, prior to VMAF developed

 

  • Good HVS-correlated  metrics (like MOVIE) are unable to provide real-time measurement of video quality, since they are too complex.

 

  • DMOS = MOS_pristine – MOS_compressed

 

  • It’s worth mentioning that almost all visual quality algorithm (incl. VMAF) doesn’t model temporal masking effects (e.g. the scene-cut masking). The score of each frame is measured independently from the score of previous and next frames. However, the human visual system does not work in that way.

 

  • In subjective quality assessment there is one problem (in my opinion, it’s rarely mentioned in technical literature) – the same stimuli typically do not receive the same rating by all the observers, the diversity in scores inevitably exists in any subjective rating process. Therefore averaging of scores is applied to get MOS (or DMOS). However, if inter-observer differences are high then the average or median of scores get less informative. Indeed, if MOS is high, we usually do not know from this single value how many observers are actually dissatisfied with the video quality.

 

 

 

6 Responses

    1. i don’t understand. if you wish you are welcome to use any my content in your site.
      if you wish me to put a link to your site, i can do it, provided that your site is non-profitable (communicate with me via slavah264@gmail.com).

Leave a Reply

Your email address will not be published. Required fields are marked *