Table 4 The OC-SVM classification error in deep processing domain (in %)a
Model | Originals | Fakes #1 | Fakes #1 | Fakes #2 | Fakes #2 |
|---|---|---|---|---|---|
(\(P_{miss}\)) | White (\(P_{fa}\)) | Gray (\(P_{fa}\)) | White (\(P_{fa}\)) | Gray (\(P_{fa}\)) | |
Based on the Eq. (19) | |||||
\(\mathcal {L}_{\text {One-class}}^1\;\): | 0 | 6.38 | 8.23 | 0 | 0 |
\(- \mathcal {D}_{{\textrm{t}} \hat{\textrm{t}}}\) | |||||
\({\mathcal L}_{\text {One-class}}^2\;\): | 0 | 6.81 | 7.09 | 0 | 0 |
\(- \mathcal {D}_{{\textrm{t}} \hat{\textrm{t}}} + \mathcal {D}_{\textrm{t}}\) | |||||
\({\mathcal L}_{\text {One-class}}^3\;\): | 0 | 1.56 | 0.99 | 0 | 0 |
\(- \mathcal {D}_{{\textrm{t}} \hat{\textrm{t}}} - \beta \mathcal {D}_{{\textrm{x}} \hat{\textrm{x}}}\) | |||||
\({\mathcal L}_{\text {One-class}}^4\;\): | 0 | 2.41 | 2.13 | 0 | 0 |
\(- \mathcal {D}_{{\textrm{t}} \hat{\textrm{t}}} + \mathcal {D}_{\textrm{t}} - \beta \mathcal {D}_{{\textrm{x}} \hat{\textrm{x}}} + \beta \mathcal {D}_{\textrm{x}}\) | |||||
Based on the Eq. (20) | |||||
\({\mathcal L}_{\text {One-class}}^3\;\): | 0 | 0.28 | 0 | 0 | 0 |
\(- \mathcal {D}_{{\textrm{t}} \hat{\textrm{t}}} - \beta \mathcal {D}_{{\textrm{x}} \hat{\textrm{x}}}\) | |||||
\({\mathcal L}_{\text {One-class}}^4\;\): | 0.57 | 0 | 0.14 | 0 | 0 |
\(- \mathcal {D}_{\textrm{t} \mathrm{\hat{t}}} + \mathcal {D}_{\textrm{t}} - \beta \mathcal {D}_{{\textrm{x}} \hat{\textrm{x}}} + \beta \mathcal {D}_{\textrm{x}}\) | |||||
Based on the OC-SVM | |||||
\({\mathcal L}_{\text {One-class}}^3\;\): | 0.28 | 0 | 0 | 0 | 0 |
\(- \mathcal {D}_{{\textrm{t}} \hat{\textrm{t}}} - \beta \mathcal {D}_{{\textrm{x}} \hat{\textrm{x}}}\) | |||||
\({\mathcal L}_{\text {One-class}}^4\;\): | 0.14 | 0 | 0 | 0 | 0 |
\(- \mathcal {D}_{{\textrm{t}} \hat{\textrm{t}}} + \mathcal {D}_{\textrm{t}} - \beta \mathcal {D}_{{\textrm{x}} \hat{\textrm{x}}} + \beta \mathcal {D}_{\textrm{x}}\) | |||||