The Pearson correlation coefficient (PCC) measures how linearly a metric's scores track the human Mean Opinion Score, answering the accuracy question: does a VMAF of 80 land where a MOS of 4.0 should. It runs from -1 to 1, where 1 means a perfect straight-line relationship; higher values mean the metric and the panel move together more closely. PCC is the covariance of the two columns divided by the product of their standard deviations. A crucial catch is that you cannot compute it directly between a 0-100 metric and a 1-5 MOS because the scales differ, so a monotonic fitting curve (a five-parameter logistic) first maps the metric onto the MOS scale; PCC is computed on those mapped values. Unlike SROCC, which judges only rank order, PCC penalizes uneven spacing, so a metric can score a perfect SROCC yet a slightly lower PCC. It is one of the four ITU-T P.1401 validation statistics, alongside SROCC, RMSE, and the outlier ratio.
The handbook · 7 volumes
Fora Soft Video Engineering Handbook
Pearson correlation shows how linearly a metric tracks MOS, but it can look strong while the metric still fails on certain content. We wrote seven volumes covering the whole video stack, and for PCC start with Volume 1: objective metrics, subjective scores and metric validation, with Volume 6 on evaluating models. Written by Nikolay Sapunov, CEO at Fora Soft.
Choosing a quality metric based on a published Pearson correlation?
Share your content and any viewer scores: we check PCC and related statistics on your own video and tell you which metric to trust. 250+ video products shipped since 2005.
