Appearance Segmentation and Documentation Applied to Cultural Heritage Surfaces

This paper describes the application of a novel supervised segmentation technique used for conservation documentation based on visible appearance changes of Cultural Heritage (CH) metal surfaces. The technique employs a linear discriminant analysis model to classify Reflectance Transformation Imaging (RTI) reconstruction coefficients. The Hemispherical Harmonics (HSH) reconstruction coefficients for each pixel are first calculated and then normalized. This normalization enhances the robustness and invariance of the application, making it possible to apply it for documenting different surfaces at different time intervals. This paper presents three cases related to surface appearance changes due to corrosion that provokes low or high topographic changes. Tarnishing of silver and filiform corrosion on steel are examined. For each case study, a supervised data set is constructed, teaching the algorithm to recognize as distinct a specified appearance characteristic (such as corrosion, metal, etc.) by comparing it to the reconstruction coefficients of each pixel through Linear Discriminant Analysis (LDA). A simplified color map visualizes the segmented information. The calculated results are afterward verified by visible inspection from conservation-restoration experts. The method can segment
surfaces with changes in micro-geometry, creating accurate cartographies of the object’s condition. However, limitations are met on surfaces with minimal topography and high specularity.