Table 4 Summary of platform provenance analysis approaches. The column “Analysis” indicates whether the detectors operate on the full image (“Global”) or on image patches (“Local”)
Method | Cues | Preprocessing | ML classifier | Analysis | Number of sharing | Dataset |
|---|---|---|---|---|---|---|
[32] | Signal-based | DCT-based feature extraction | Random Forest | Global | Single | MICC social UCID, MICC public UCID |
[39] | Both | DCT-based feature extraction + metadata extraction | LR, SVM, RF | Global | Multiple | ISIMA |
[34] | Signal-based | DCT-based feature extraction | 1D CNN | Local | Single | UNICT-SNIM, MICC social UCID, MICC public UCID |
[35] | Signal-based | DCT-based feature extraction + noise residuals extraction | 1D CNN + 2D CNN | Local | Single | UNICT-SNIM, MICC social UCID, MICC public UCID, VISION |
[36] | Signal-based | Noise residuals extraction | 2D CNN | Local | Single | UNICT-SNIM, MICC social UCID, MICC public UCID, VISION |
[33] | Metadata-based | Metadata extraction | − | Global | Single | UNICT-SNIM |
[38] | Metadata-based | Metadata extraction | Distance-based K-NN + Decision trees | Global | Single | UNICT-SNIM |
[40] | Both | DCT-based feature extraction + metadata extraction | CNN + feature fusion | Local | Multiple | R-SMUD, V-SMUD |
[37] | Metadata-based | JPEG header analysis | Random forest | Global | Single | − |
[42] | Signal-based | Pixel + DCT domain | SVM | Global | Single | − |
[41] | Signal-based | Pixel domain | Siamese 2D CNN | Local | Single | − |