Machine learning for painting conservation: a state-of-the-art review

Abstract This review discusses recent machine learning applications in painting conservation, highlighting five themes: enhancement of scientific imagery, pigment analysis, damage detection, virtual restoration, and damage prediction. A persistent challenge is the scarcity of high-quality historical data, limiting model reliability and scope. While supervised learning remains widespread, unsupervised methods hold promise for revealing nuanced patterns within complex datasets. The review seeks to foster collaboration between machine learning practitioners and conservation professionals.