Reduced reference video quality metrics

Image for reduced reference metrics

Whereas they are very useful to benchmark and monitor encoders, full-reference video quality metrics cannot be used to monitor quality after transmission, from the end-user point of view.

Indeed, the reference video (without distortions) is not available at the end-user side.

To monitor quality from end to end, no-reference metrics or reduced-reference metrics must be used.

A reduced-reference video quality metric measures the quality of the distorted video (received by the end-user) by analyzing it and comparing it to a reduced number of features extracted from the reference video.

The set of these features constitutes a reduced description of the reference video. The size (in bytes) of this description must be limited (hence the word "reduced") so that this description can be transmitted in parallel of the video stream.

Our reduced-reference video quality metrics extract features which depend on the used codec.

Therefore, we propose three different reduced-reference video quality metrics: one for HEVC, one for H.264 and one for MPEG-2.

The features extracted from the reference video and transmitted are compared to the features extracted from the distorted video. This comparison produces a quality score expressed on a DMOS scale (Differential Mean Opinion Score).

A DMOS score indicates the loss of quality of the video under test, compared to the reference video.

Related product: Audio Video Sync & Quality Monitor (AVSQM) (ask for a free evaluation version)