Mobile authentication of copy detection patterns

EURASIP Journal on Information Security

Table 3 The OC-SVM classification error in spatial domain (in %)^a

Train on	Originals	Fakes #1	Fakes #1	Fakes #2	Fakes #2
Train on	(\(P_{miss}\))	White (\(P_{fa}\))	Gray (\(P_{fa}\))	White (\(P_{fa}\))	Gray (\(P_{fa}\))
With respect to the digital templates:
- Grayscale \({\textbf {x}}\)	3.1	2.54	3.82	0	0
- RGB \({\textbf {x}}\)	2.82	2.1	1.4	0	0
With respect to the physical references:
- Grayscale \({\textbf {x}}\)	11.44	35.86	40.58	1.72	1.12
- RGB \({\textbf {x}}\)	11.16	31.84	39.54	1.44	0.98

^aThe python OneClassSVM method from the sklearn package is used with the next training parameters: kernel = “rbf”; gamma = 0.1; nu = 0.03 for the digital templates and nu = 0.1 for the physical references