Skip to main content

Table 5 Realistic performance of the individual detectors and fusion approaches with the fixed decision thresholds and equal fusion weights with the combined LondonDB/UtrechtDB/Alabama datasets (best result per morph type 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

  

SC1

SC2

Detector

Morph type

BPCER

APCER

HTER

BPCER

APCER

HTER

DArXivMC

Combined

7.45%

1.00%

4.22%

3.87%

18.56%

11.22%

DArXivNaive

 

2.01%

1.82%

1.91%

0.97%

13.32%

7.14%

DBIOSIGMC

 

25.76%

1.35%

13.56%

23.10%

10.12%

16.61%

DBIOSIGNaive

 

11.47%

5.25%

8.36%

9.54%

18.05%

13.80%

Dkeypoints

 

87.86%

0.00%

43.93%

96.94%

0.00%

48.47%

FM

 

11.39%

0.56%

5.97%

7.15%

9.30%

8.23%

FWLC

 

18.09%

0.02%

9.05%

19.90%

0.84%

10.37%

FDST

 

25.47%

0.02%

12.74%

35.02%

0.01%

17.52%

FLR

 

23.68%

0.02%

11.85%

27.05%

0.35%

13.70%

DArXivMC

Complete

7.45%

1.00%

4.22%

3.87%

15.73%

9.80%

DArXivNaive

 

2.01%

1.60%

1.81%

0.97%

10.57%

5.77%

DBIOSIGMC

 

25.76%

1.38%

13.57%

23.10%

8.38%

15.74%

DBIOSIGNaive

 

11.47%

4.47%

7.97%

9.54%

14.31%

11.92%

Dkeypoints

 

87.86%

0.00%

43.93%

96.94%

0.00%

48.47%

FM

 

11.39%

0.31%

5.85%

7.15%

3.76%

5.46%

FWLC

 

18.09%

0.02%

9.05%

19.90%

0.60%

10.25%

FDST

 

25.47%

0.02%

12.74%

35.02%

0.02%

17.52%

FLR

 

23.68%

0.02%

11.85%

27.05%

0.27%

13.66%

DArXivMC

Splicing

7.45%

2.57%

5.01%

3.87%

24.81%

14.34%

DArXivNaive

 

2.01%

3.54%

2.77%

0.97%

16.04%

8.50%

DBIOSIGMC

 

25.76%

3.54%

14.65%

23.10%

17.39%

20.25%

DBIOSIGNaive

 

11.47%

7.39%

9.43%

9.54%

21.22%

15.38%

Dkeypoints

 

87.86%

0.02%

43.94%

96.94%

0.00%

48.47%

FM

 

11.39%

1.45%

6.42%

7.15%

12.35%

9.75%

FWLC

 

18.09%

0.07%

9.08%

19.90%

1.42%

10.66%

FDST

 

25.47%

0.03%

12.75%

35.02%

0.03%

17.52%

FLR

 

23.68%

0.03%

11.85%

27.05%

0.55%

13.80%