Munsell Soil Finder e PyWALL: strumenti per la documentazione cromatica e il rilievo dei paramenti murari in archeologia

Andrea Fiorini

«Archeologia e Calcolatori» 2026, 37.1, 255-276; doi: 10.19282/ac.37.1.2026.13

Abstract

This paper presents two software applications for quantitative archaeological documentation: Munsell Soil Finder and PyWALL. Both tools were developed using AI-assisted natural language programming (Vibe Coding), demonstrating its viability for domain-specific archaeological software development. The two tools address recurrent operational bottlenecks: the standardization of color recording and time-intensive masonry documentation. Munsell Soil Finder maps pixel samples from calibrated photographs to Munsell (H V/C) notation via KD-tree nearest-neighbour search in CIELAB and ΔE2000 refinement, using 2734 reference samples plus interpolated values. PyWALL segments bricks and mortar in orthorectified images via K-Means clustering (Lab + edge/texture features) or a pre-trained DynUNet, extracts contours, and exports to DXF. Validation comprised: 1) chart tests with ColorChecker profiling (accuracy 63-100%, robustness ≥0.87); 2) human-versus-software and inter-observer comparisons (38.3% exact agreement with the software output; 28.9% among observers), highlighting the variability of visual assessment; 3) segmentation benchmarking (IoU 0.883-1.00; F1 0.938-1.00). A case study on Rimini’s medieval walls employs fuzzy c-means on combined chromatic and metric features to identify a consistent brick module (~27×12×5 cm), two clusters per brick orientation – separated primarily by CIELAB values rather than format, suggesting variability in clay composition or f iring conditions – and fragmentation rates of 26.92-41.14%.

Figures

Preview