Table 2 Dataset specifications and preprocessing parameters.
Parameter category | Specification | Details |
|---|---|---|
Image acquisition | ||
Original resolution | 2048 × 2048 pixels | Professional museum digitization |
Color depth | 24-bit RGB | sRGB color space |
File format | TIFF (archival), PNG (training) | Lossless compression |
Dataset composition | ||
Total images | 9700 | 8500 historical + 1200 contemporary |
Training set | 7760 (80%) | Stratified by pattern category |
Validation set | 970 (10%) | Stratified by pattern category |
Test set | 970 (10%) | Stratified by pattern category |
Preprocessing | ||
Training resolution | 512 × 512 pixels | Bicubic downsampling |
Color normalization | Histogram equalization | Per-channel normalization |
Contrast enhancement | CLAHE | Clip limit = 2.0, tile size = 8 × 8 |
Data augmentation | ||
Horizontal flip | Probability 0.5 | Preserves symmetry characteristics |
Random rotation | ± 15 degrees | Maintains compositional integrity |
Random crop | Scale 0.8-1.0 | Preserves pattern completeness |
Color jitter | B ± 0.1, C ± 0.1, S ± 0.1 | Simulates lighting variations |
Annotation quality | ||
Annotators | 3 experts + 1 senior reviewer | Folklore, art history, conservation |
Cohen’s Kappa | 0.82 (95% CI: 0.79–0.85) | Substantial agreement |
Fleiss’ Kappa | 0.78 | Multi-rater consistency |
Cultural tags | 237 categories | Hierarchical organization |
Visual descriptors | 1,856 terms | Standardized vocabulary |
