Multimodal data processing and deep learning in digital painting analysis

Summary form only given, as follows. The complete presentation was not made available for publication as part of the conference proceedings. Art investigation - an emerging and rapidly growing cross-disciplinary field of research - relies increasingly on signal processing and machine learning. Multimodal imaging is now routinely employed in order to support the technical study of art work, its conservation, or even presentation. Analysis of multimodal data often uncovers regions or patterns of interest that would otherwise remain unnoticed, thus enabling new insights and even support for certain decisions that are made during the conservation-restoration treatments. In this talk, we discuss recent advances in digital signal processing and machine learning for supporting the study and restoration of paintings. The focus will be on sparse coding, representation learning and deep learning, spatial context modelling and Bayesian inference in tasks such as crack detection, paint loss detection and virtual inpainting. Concrete examples will be shown from the ongoing conservation-restoration treatment of the Ghent Altarpiece.