ROAD FEATURES ENHANCEMENT IN SENTINEL-2 IMAGERY USING CONDITIONAL DILATION AND LINEAR STRUCTURING ELEMENTS

This study evaluated the application of the conditional dilation operator, and its erosion based variant, for enhancing road centerlines in Sentinel-2 multispectral imagery. The analysis involved multiple spectral bands (b2, b3, b4, b8) and derived indices (NDVI, NDBI, NDWI), using a dataset of approximately 890 image patches per band. Results demonstrated that a structuring element length of 380 pixels provided the best trade-off across most bands, although no significant improvements were observed for band b4. The conditional dilation operator (δB,X ) achieved strong results in NDBI and NDWI bands, while the variant operator (δ′ B,X) performed better for b8 and NDVI. The use of the opening-by-reconstruction operator as a preprocessing step had limited influence on segmentation performance, and in some cases, subtly degraded results. Future research should explore more selective marker definitions for reconstruction and combine multiple morphological strategies to improve road segmentation robustness.