Graphics Reference
In-Depth Information
Table 5.5
Comparison of verification rates (i.e., true acceptance rate,
TAR
)at
FAR
=
0
.
1% on the FRGC v2.0 data set for the state-of-the-art methods
TAR@FAR
=
0.1%
Method
ROC III
All vs. All
Kakadiaris et al. (2007b)
97%
-
Faltemier et al. (2008a)
94.8%
93.2%
Berretti et al. (2010b)
-
81.2%
Queirolo et al. (2010a)
96.6%
96.5%
Spreeuwers (2011)
94.6%
94.6%
Wang et al. (2010)
98.4%
98.13%
Drira et al., (2012)
97.14%
93.96%
The highest published results on this data set are those reported in the works of Queirolo
et al. (2010a), Spreeuwers (2011), Wang et al. (2010), with a rank-1 recognition rate greater
than 98% for the neutral versus neutral experiment. The overall best recognition score on
FRGC v2.0 is reported by Spreeuwers (2011), which uses an intrinsic coordinate system on
the basis of the vertical symmetry plane through the nose.
To evaluate the performance of the state-of-the-art approaches in a verification scenario,
receiver operating characteristic (ROC) curves for the ROC III mask of the FRGC v2.0 and
the “All vs. All” experiment are reported in Table 5.5 using the verification results at FAR of
0.1 percent as performance indicator.
Comparative Evaluation on GAVAB Data Set
The GAVAB DB (see Section 5.2) has many noisy 3D face scans with large expressions
and also scans with missing parts because of acquisitions where subjects exhibit large pose
variations with respect to the frontal one. The GAVAB experimental protocol assumes that
one of the two frontal scans with the neutral expression for each person is taken as a gallery
model, and the remaining are used as probes.
Table 5.6 compares published results of methods using the same evaluation protocol on
the GAVAB. The approaches in Drira et al. (2010) and Huang et al. (2012) attain the highest
recognition rate for faces with non-neutral expressions and side scans.
Table 5.6
Recognition results comparison between different methods on the GAVAB data set
neutral
expressive
looking-down
looking-up
right side
left side
Li et al. (2009)
96.67%
93.33%
-
-
-
-
Moreno et al. (2005)
90.16%
77.9%
-
-
-
-
Mahoor et al. (2009)
-
72%
85.3%
88.6%
-
-
Huang et al. (2012)
100%
93.99%
96.72%
96.72%
93.44%
78.69%
Drira et al. (2010)
100%
94.54%
100%
98.36%
70.49%
86.89%
