Color Homogenization in 3D Reconstruction of Castles from Heterogeneous Imaging Sources

Abstract. This study, conducted as part of the Rhineland Castles 3D Modeling Project, addresses the problem of color inconsistencies in 3D reconstructions generated from diverse imagery sources, including drone footage and terrestrial photography. Variations in lighting conditions, camera sensors and acquisition settings often result in photometric discrepancies that degrade the quality of the textured 3D model. 
Two datasets from Ramstein and Oedenbourg Castles in France were analysed, each presenting characterised by distinct photometric conditions. The Ramstein dataset exhibited relatively uniform lighting, allowing the application of statistical color transfer methods. Among the methods evaluated, Mean-Lab Transfer produced the most photometrically consistent and visually faithful results. In contrast, the Oedenbourg dataset presented significant photometric variability, characterised by extreme variations in saturation, luminosity, and contrast. To effectively manage these challenging conditions Here, a two-step enhancement approach was adopted, combining gamma correction (in HSV color space) and with CLAHE to balance brightness and preserve color identity. This paper details the applied methodologies, evaluates their effectiveness in achieving color consistency, and highlights the importance of adopting emphasizing the need for dataset-specific processing pipelines.