Comparison of visibility metrics - detailed results

This is a comparison of several metrics for predicting visibility of differences between two images (discrimination). The table below lists prediction errors for a set of metrics (rows) and test sets (columns). The prediction error is given as:

The Speak value is the peak sensitivity parameter of the metric used to fit the data. Note that this parameter is specific to each metric and only the differences between the test-sets fitted with the same metric are relevant.

Click on the name of the data set to see more information and the gallery of all stimuli

ModelFest CSF lum. range Blackwell's t.v.i. Glare at scotopic lum. Foley's masking data CSF flattening Complex images
VDP'93 R=4.3 dB
chi2red=1.4
Speak=5.6
see details
R=15 dB
chi2red=9.4
Speak=4.7
see details
R=28 dB
Speak=3.6
see details
R=15 dB
chi2red=2.2
Speak=4.5
see details
R=7.7 dB
Speak=5.3
see details
R=3.9 dB
chi2red=5.8
Speak=4.8
see details
R=7.1 dB
chi2red=3.7
Speak=5
see details
HDR-VDP CPP 1.7 R=3.3 dB
chi2red=0.75
Speak=2.4
see details
R=7.5 dB
chi2red=2.4
Speak=2.6
see details
R=41 dB
Speak=5.4
see details
R=2.7 dB
chi2red=0.071
Speak=2.1
see details
R=7.1 dB
Speak=2
see details
R=6 dB
chi2red=13
Speak=2
see details
R=7 dB
chi2red=3.6
Speak=1.9
see details
HDR-VDP-2 R=2.8 dB
chi2red=0.6
Speak=2.5
see details
R=1.9 dB
chi2red=0.15
Speak=2.4
see details
R=3.5 dB
Speak=2.5
see details
R=1.1 dB
chi2red=0.011
Speak=2.7
see details
R=2 dB
Speak=2.1
see details
R=2.6 dB
chi2red=2.5
Speak=2.7
see details
R=5.3 dB
chi2red=2.1
Speak=2.4
see details