Video quality metrics
Video quality metrics are measurements used to judge how good a video looks or how well it plays, covering both picture quality and playback performance. Picture metrics such as VMAF, PSNR and SSIM compare an encoded video with its source. Playback metrics such as startup time and rebuffer ratio describe what viewers actually experience.
Why it matters
Every cost-saving change in video, from a cheaper encoding setting to a new CDN, can quietly hurt what viewers see. Quality metrics make those trade-offs visible before they cause churn. They also give teams a shared language for setting targets and proving improvements.
How it works
- Full-reference metrics like VMAF, PSNR and SSIM compare each encoded video with its original source.
- Playback metrics, such as startup time and stalls, are collected from the player during real views.
- Scores are tracked per title, rendition, device and region to find weak spots.
- Teams set targets for each metric and check new encodes or delivery changes against them.
What are the most important video quality metrics?
The most important video quality metrics fall into two groups. Picture quality metrics, such as VMAF, SSIM and PSNR, measure how good the encoded video looks compared with its source. Playback metrics, such as startup time, rebuffer ratio, start failures and average bitrate, measure what viewers actually experience. A healthy service tracks both.
How do you measure video quality objectively?
Measure video quality objectively with full-reference metrics that compare the encoded video against the original, and with player data from real viewers. VMAF, SSIM and PSNR are standard for the first. Player analytics covering startup, stalls, bitrate and resolution changes cover the second. For high-stakes choices, confirm results with a human viewing test that produces a mean opinion score.
Where it's used
OTT services put picture quality and playback numbers on one dashboard, so a cheaper encoding setup or new CDN never quietly makes viewing worse for subscribers.
Where it usually goes wrong, and how FastPix handles it
Picture quality is often checked once at encoding time, but viewers may still end up watching low resolutions or upscaled video when their network or device forces the player to drop down.
Average bitrate, upscale and downscale percentages and resolution changes are reported for every view by FastPix Video Data. These feed a render quality score, which is combined with stability and startup scores into one QoE score per view. You can check that a cost-saving encoding or delivery change did not quietly lower the picture viewers actually get.
By the numbers
- ITU-T P.1203, a standard model for estimating the quality of progressive download and adaptive streaming sessions, was first approved on November 29, 2016. (ITU-T Recommendation P.1203 (11/2016))
