Research Article
Perceptual Image Hashing Based on Multitask Neural Network
Table 5
F1 scores of different schemes in different content retention operations.
| Operations | Proposed | DCP [7] | RP-IVD [4] | RP-NMF [3] | DAE NN-based [24] |
| Speckle noise | 0.9996 | 0.9340 | 0.9409 | 0.9299 | 0.9995 | Scaling | 0.9996 | 0.8487 | 0.8909 | 0.9665 | 0.9993 | Median filtering | 0.9996 | 0.9631 | 0.9782 | 0.9891 | 0.9998 | Circle average filtering | 0.9988 | 0.9063 | 0.7121 | 0.7889 | 0.9895 | JPEG compression | 0.9993 | 0.7957 | 0.9383 | 0.9925 | 0.9998 | Rotation and cropping | 0.9990 | 0.6339 | 0.9936 | 0.9983 | 0.9961 | Gamma correction | 0.9987 | 0.9518 | 0.9903 | 0.8418 | 0.9999 | Gaussian filtering | 0.9991 | 0.9195 | 0.8148 | 0.8656 | 0.9992 |
| Mean | 0.9992 | 0.8708 | 0.9074 | 0.9216 | 0.9979 |
|
|