Research Article
Adaptive Steganalysis Based on Selection Region and Combined Convolutional Neural Networks
Table 2
Performance comparison of multiple methods based on the testing error.
| Algorithm | bpp | Proposed method | Method [14] | SRM | maxSRMd2 | 256 | 384 | 512 | 512 |
| WOW | 0.1 | 0.4144 | 0.4349 | 0.4577 | 0.4249 | 0.4056 | 0.3001 | 0.2 | 0.3235 | 0.3150 | 0.2917 | 0.3625 | 0.3204 | 0.2353 | 0.3 | 0.2375 | 0.2202 | 0.2160 | 0.2473 | 0.2501 | 0.1905 | 0.4 | 0.1900 | 0.1814 | 0.1847 | 0.2072 | 0.2092 | 0.1593 |
| HUGO | 0.1 | 0.4083 | 0.3688 | 0.3711 | 0.3957 | 0.3601 | 0.3103 | 0.2 | 0.2979 | 0.2731 | 0.2816 | 0.3070 | 0.2834 | 0.2396 | 0.3 | 0.2533 | 0.2304 | 0.2194 | 0.2316 | 0.2259 | 0.1990 | 0.4 | 0.1899 | 0.1741 | 0.1517 | 0.1746 | 0.1852 | 0.1641 |
| HILL | 0.1 | 0.4422 | 0.4403 | 0.4646 | 0.4500 | 0.4354 | 0.3776 | 0.2 | 0.3464 | 0.3344 | 0.3494 | 0.3642 | 0.3628 | 0.3151 | 0.3 | 0.2653 | 0.2581 | 0.2491 | 0.2683 | 0.3017 | 0.2655 | 0.4 | 0.2107 | 0.1937 | 0.1822 | 0.2083 | 0.2503 | 0.2192 |
| S-UNIWARD | 0.1 | 0.4680 | 0.4566 | 0.4511 | 0.4525 | 0.4075 | 0.3661 | 0.2 | 0.3624 | 0.3952 | 0.3319 | 0.3364 | 0.3247 | 0.2912 | 0.3 | 0.2781 | 0.2601 | 0.2511 | 0.2675 | 0.2584 | 0.2383 | 0.4 | 0.2348 | 0.2027 | 0.1845 | 0.2074 | 0.2052 | 0.1959 |
|
|