Skip to main content

Table 9 Fusion under laboratory conditions: tenfold stratified cross-validation with 90% training/10% test split; genuine samples from the Alabama dataset [53]; morphs from LondonDB and UtrechtDB (best result per morph type and application scenario marked in bold)

From: Potential advantages and limitations of using information fusion in media forensics—a discussion on the example of detecting face morphing attacks

  Combined Complete Splicing
SC1 SC2 SC1 SC2 SC1 SC2
EER HTER EER HTER EER HTER EER HTER EER HTER EER HTER
Fixed
DArXivMC 3.8% 4.3% 8.2% 11.2% 3.9% 4.2% 7.2% 9.8% 4.8% 5.0% 9.2% 14.3%
DArXivNaive 1.5% 1.9% 3.9% 7.1% 1.3% 1.8% 3.4% 5.8% 1.8% 2.7% 4.4% 8.5%
DBIOSIGMC 9.3% 13.6% 15.7% 16.6% 9.3% 13.6% 14.7% 15.7% 12.8% 14.7% 20.2% 20.2%
DBIOSIGNaive 7.1% 8.4% 13.4% 13.7% 7.0% 7.9% 11.8% 11.9% 8.4% 9.4% 14.2% 15.4%
Dkeypoints 12.3% 43.9% 18.6% 48.8% 12.2% 43.9% 19.3% 48.8% 8.9% 43.9% 18.3% 48.8%
FM   6.0%   8.2%   6.2%   5.9%   6.4%   9.7%
FWLC   9.6%   10.9%   9.2%   10.6%   9.2%   10.6%
FDST   2.6%   5.9%   3.0%   6.7%   2.9%   7.3%
FLR   2.6%   5.9%   3.0%   6.7%   2.9%   7.3%
Adaptive
DArXivMC 3.8% 3.9% 8.2% 8.3% 3.9% 4.0% 7.2% 7.9% 4.8% 4.9% 9.2% 9.4%
DArXivNaive 1.5% 2.2% 3.9% 5.0% 1.3% 2.1% 3.4% 4.4% 1.8% 3.0% 4.4% 5.6%
DBIOSIGMC 9.3% 9.8% 15.7% 16.4% 9.3% 9.8% 14.7% 15.0% 12.8% 12.7% 20.2% 20.7%
DBIOSIGNaive 7.1% 7.9% 13.4% 13.6% 7.0% 7.6% 11.8% 12.1% 8.4% 9.1% 14.2% 15.0%
Dkeypoints 12.3% 12.7% 18.6% 19.0% 12.2% 12.5% 19.3% 19.3% 8.9% 11.4% 18.3% 19.2%
FM   2.8%   6.0%   2.2%   4.5%   3.3%   6.8%
FWLC   15.2%   39.2%   14.3%   35.5%   17.7%   45.8%
FDST   2.8%   5.8%   3.3%   6.6%   3.1%   7.3%
FLR   2.8%   5.8%   3.3%   6.6%   3.1%   7.3%