A wide variety of video quality (VQ) metrics have been introduced over the past decades. VQ metrics are used in a range of applications, including video quality monitoring, encoder optimization, bitrate selection, and adaptive bitrate ladder configuration. While quality metrics have been developed that correlate well with human visual perception, they typically require high computational complexity. Other popular metrics are less complex, but are not accurate enough to base encoder decisions on.
In this paper, we discuss relevant quality metrics and their computational complexity. We address the difficulties around real-time VQ measurement, and determine which metrics can be used inside real-time encoders, to not only measure, but to actively control video quality. We include recent Machine Learning (ML) driven research, with a focus on the complexity introduced by ML techniques for video quality assessment. While these techniques provide benefits to improve VQ measurement and monitoring applications, we illustrate the fine line between accuracy of ML networks and the number of operations performed in these networks and their associated cost.