Scene-adaptive RTI acquisition guided by dataset quality analysis

Reflectance Transformation Imaging (RTI) is widely used for the documentation and visual analysis of cultural heritage artefacts. However, RTI acquisitions are typically performed using fixed or heuristic light configurations that do not account for the reflectance and geometric properties of the observed scene. As a result, datasets may contain redundant sampling in some regions of the light hemisphere while remaining insufficient in others.This paper proposes a framework for analysing the quality of RTI datasets in order to guide scene-adaptive acquisition strategies. The approach starts from a sparse pilot acquisition. We interpret RTI acquisitions as samples of the reflectance response of surfaces across illumination space. By analysing the structure of these reflectance responses, the proposed framework identifies sampling imbalances and suggests adaptive illumination directions.The framework performs two complementary analyses. First, a global analysis is conducted in the light-position space, where the Sampling Balance Indicator (SBI) quantifies the relationship between angular distances of neighbouring light directions and their corresponding image differences in order to detect sampling imbalance. Second, a local analysis is performed in the image domain, where pixel-wise reflectance responses are analysed using Self-Organizing Maps (SOMs) to identify heterogeneous surface behaviours across the scene.Based on these analyses, the framework proposes global and local light-position suggestion strategies aimed at improving illumination sampling and enabling scene-adaptive RTI acquisition. Experiments on both synthetic and real multi-light image collections show that the proposed analysis reveals significant variations in sampling behaviour across different surfaces and provides practical guidance for designing improved RTI acquisitions.