Impact of embedded bit length on neural network performance
| Database | Original weights | Bits embedded (15,200 bits) | Bits embedded (total weights = 22,829,811) | |
|---|---|---|---|---|
| MICC-F220 | Reconstruction SSIM | 0.8135 | 0.8053 | 0.8042 |
| Retrieved information BER | – | 0 | 0 | |
| Realistic tampering | Reconstruction SSIM | 0.7398 | 0.7365 | 0.7353 |
| Retrieved information BER | – | 0 | 0 | |
| Coverage | Reconstruction SSIM | 0.8018 | 0.8003 | 0.7998 |
| Retrieved information BER | – | 0 | 0 | |
| High-resolution | Reconstruction SSIM | 0.7234 | 0.7194 | 0.7205 |
| Retrieved information BER | – | 0 | 0 |
| Database | Original weights | Bits embedded (15,200 bits) | Bits embedded (total weights = 22,829,811) | |
|---|---|---|---|---|
| MICC-F220 | 0.8135 | 0.8053 | 0.8042 | |
| – | 0 | 0 | ||
| Realistic tampering | 0.7398 | 0.7365 | 0.7353 | |
| – | 0 | 0 | ||
| Coverage | 0.8018 | 0.8003 | 0.7998 | |
| – | 0 | 0 | ||
| High-resolution | 0.7234 | 0.7194 | 0.7205 | |
| – | 0 | 0 |