University Archives

Unpaved forest roads present significant challenges for classical image-processing algorithms due to highly variable illumination, irregular textures, and dense vegetation. This work develops a modular sequential segmentation process consisting of linearization, Region of Interest (ROI) extraction, Hue-Saturation-Value-based shadow mitigation, Retinex normalization, color clustering, and texture-based segmentation. The system demonstrates that the combination of illumination normalization with both chromatic and texture domain segmentation vastly improves the stability across shadowed, gravel road variations. Muitl-Scale Retinex proved especially critical in reducing luminance discrepancies before clustering.