Table 1 Candidate features, and their resulting weight after optimization.
From: A fragment-based approach for computing the long-term visual evolution of historical maps
Method type | Method | Interpretation | Weight |
|---|---|---|---|
Color distribution | RGB Histogram | Distribution of color intensities across the red-green-blue channels. | – |
RGB Peak color | Most frequent color intensity in each channel. | – | |
RGB Standard deviation | Dispersion of color intensities. | – | |
RGB Skew | Asymmetry of color intensity distribution. | – | |
RGB Kurtosis | Acuteness of color intensity distribution. | – | |
Morphology | HOG | Histogram of local edge orientations, relatively to the dominant edge orientation. | 1. |
Texture | LBP | Texture patterns (e.g. hatched, grid, plain, etc.), based on the local arrangement of pixels. | 0.1 |
Graphical Load | Otsu’s dark pixel proportion | Density of pixels with dark content. | – |
ED2 | Square root of the density of edges. | 0.15 | |
Num Connected Components | Number of graphically connected features in the image. | 0.25 | |
Line width | Skeleton ratio | Content ratio before and after thinning the mapel to single pixel width (skeletonization). | 0.25 |
Orientation | HOG orientation | Dominant edge orientation of the mapel, before its neutralization. | 0.05 |
Binned HOG pattern | Regroup the edge orientations into discrete bins*. | – |
