Table 2.
Performance of the different model settings, as evaluated on the manually classified test set.
| Classifier | Compression | Data normalization | High-quality MSE | Content | MSE | SSIM | TSS | Accuracy |
|---|---|---|---|---|---|---|---|---|
| ✓ | 8 | ImageNorm | 0.0044 | 0.37 | 0.013 | 0.03 | 0.95 | 97.7% |
| ✓ | 8 | ContrastNorm | 0.0017 | 0.72 | 0.042 | 0.08 | 0.97 | 98.5% |
| ✓ | 1 | ImageNorm | 0.0044 | 0.37 | 0.006 | 0.02 | 0.95 | 97.7% |
| ✓ | 1 | ContrastNorm | 0.0019 | 0.45 | 0.003 | 0.01 | 0.95 | 97.7% |
| ✗ | 8 | ImageNorm | 0.0038 | 0.12 | 0.005 | 0.02 | 0.82 | 89.3% |
| ✗ | 8 | ContrastNorm | 0.0016 | 0.18 | 0.004 | 0.04 | 0.92 | 96.1% |
| ✗ | 1 | ImageNorm | 0.0042 | 0.18 | 0.006 | 0.03 | 0.93 | 96.5% |
| ✗ | 1 | ContrastNorm | 0.0018 | 0.47 | 0.028 | 0.09 | 0.93 | 96.5% |
| Baseline KSO quality classification | – | – | – | – | 0.30 | 64.2% | ||
Notes. Content, MSE, and SSIM refer to the difference between the median losses of the two quality distributions for the according metric (see Sect. 3.6). Compression refers to the number of channels used for the quantized representation. Correct classifications are determined by the criteria introduced in Sect. 3.6 and are quantified in terms of accuracy and TSS. The evaluation of the original KSO labels is given in the last row. The results of the model yielding the best performance (CLASS-q8-CONTR) are marked in bold face.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.