Table 2 Detection results compison in terms of average p, r, and F (%) between the SD-Net and other methods on CoMoFoD30 and CASIA II33 datasets.
From: Image copy-move forgery detection and localization based on super-BPD segmentation and DCNN
Methods | CoMoFoD30 | CASIA II33 | |||||
|---|---|---|---|---|---|---|---|
p | r | F | p | r | F | ||
Conventional | Ryu et al.35 | 45.78 | 34.35 | 37.37 | 22.71 | 13.36 | 16.40 |
Cozzolino et al.36 | 39.92 | 47.61 | 41.83 | 24.92 | 26.81 | 25.43 | |
Wang et al.1 | 49.09 | 57.45 | 46.44 | 30.64 | 31.23 | 31.08 | |
CNN-based | Wu et al.37 | 36.29 | 40.41 | 31.13 | 23.97 | 13.79 | 14.64 |
BusterNet5 | 57.34 | 49.39 | 49.26 | 55.71 | 43.83 | 45.56 | |
AR-Net17 | 54.21 | 46.55 | 50.09 | 58.32 | 37.33 | 45.52 | |
SD-Net | 59.11 | 57.69 | 50.77 | 57.48 | 51.25 | 48.06 | |
