Multiscale surface-texture characterisation of heritage paintings using reflectance transformation imaging and completed local binary patterns

Reflectance transformation imaging provides a non-contact means of recording the illumination-dependent appearance of cultural heritage objects and can assist in the visual examination of brushwork, cracks, surface irregularities, and material-texture variations. However, the large number of directional-light images produced during RTI acquisition creates challenges for extracting stable and interpretable surface-texture information. This study presents a multiscale computational framework for characterising the visible surface features of heritage paintings using physically acquired RTI image sequences. The proposed framework first normalises the directional-light observations and generates relightable representations under controlled illumination directions. Discrete wavelet transform (DWT) is subsequently employed to decompose the RTI observations into multiscale frequency components representing coarse surface structures and fine textural variations. Completed local binary patterns (CLBP) are then extracted from the wavelet sub-bands to describe local intensity differences, neighbourhood relationships, and directional texture characteristics. The resulting descriptors are combined to produce surface-detail maps and quantitative texture representations for different regions of the paintings. The framework is evaluated using real multi-light RTI datasets and compared with conventional local binary patterns, wavelet-only descriptors, gradient-based methods, and standard texture-analysis approaches. Evaluation considers descriptor repeatability across lighting directions, texture separability, edge preservation, local contrast, computational cost, and robustness to illumination variation. The framework is intended for computational surface visualisation and texture characterisation. The study provides a reproducible and physically grounded approach for analysing visible painting-surface characteristics from real RTI measurements and establishes a basis for subsequent evaluation with conservator-annotated heritage datasets.