The images below plot top-match identification performance against the size of the gallery. Each point on the graphs was obtained by generating a large number of trials for each specified gallery size, and, for each, making a corresponding probe set of all the matching images of the people in that gallery.
For each algorithm the same random galleries were used (by initiating the random number generator with the same seed).
The graphs below are from four studies labelled at left: "duplicate I and II" address time delay between images (subject has aged) "fafb" measures recognition from a different facial expression; and "fafc" estimates the effect of changing illumination.
The graphs at left have a logarithmic x axis. The right column uses a linear x axis.
| Click on each image for a high resolution Postscript file. | ||
| dupI |
|
|
| dupII |
|
|
| fafb |
|
|
| fafc |
|
|