The value of relative quality in video delivery

Estimating perceived quality of video is typically done by gauging the user's response on an absolute scale of ratings (excellent, good, fair, poor and bad). However, the internal representation of these adjectives to the stimuli varies significantly in different people. Even though the goal is to make an absolute estimate of the perceived quality, these questions reveal merely relative tendencies due the incorporated bias and variability in the responses. We present results from quality assessment based on estimates of relative quality distances between samples, by asking the question in the form or comparison rather than rating. This, two-alternative forced choice method scales the differences in a form of psychometric function, which presents the utility of the perceived quality on a measurable objective value. We argue that this relativistic mapping with low variance is more useful in video delivery because it offers an accurate way to optimize the resources.