Research Article
Perceptual Image Hashing Based on Multitask Neural Network
Table 1
Structure of the multitask neural network.
| | Input (32 × 32 RGB image) | conv3-64 + ReLU
conv64-64 + ReLU | | MaxPooling | Conv64-128 + ReLU
Conv128-128 + ReLU | | MaxPooling | | Branch I | Branch II | | Conv128-256 + ReLU | Conv128-256 + BN + ReLU
Conv256-256 + BN + ReLU | | Conv256-256 + ReLU | | Conv256-256 + ReLU | | Conv256-256 + ReLU | | MaxPooling | MaxPooling | | Conv256-512 + ReLU | Conv256-512 + BN + ReLU
Conv512-512 + BN + ReLU | | Conv512-512 + ReLU | | Conv512-512 + ReLU | | Conv512-512 + ReLU | | MaxPooling | AvgPooling | | Conv512-512 + ReLU | FC-512 + ReLU | | Conv512-512 + ReLU | | Conv512-512 + ReLU | | Conv512-512 + ReLU | | MaxPooling | FC-50 | | FC-512 + ReLU | | FC-256 + ReLU | | FC-10 | | Softmax |
|
|
