Visual digital humanities: using image data to derive approximate metadata

Digital humanities researchers often collaborate with computer scientists, but most commonly with those computational researchers who work on the analysis of words and texts. Where collaborations have evolved around imaging, they tend to be on the capture of images, rather than analysis. Computer vision researchers spend their days extracting meaningful information from images and video, but there has been little work applying these techniques in the digital humanities field. In this chapter we describe preliminary work which collab oratively creates an approach to digital humanities that can deal with pictures as pictures , by analysing the visual properties of an image. This emerges through the development of a computational approach to modelling stylistic change, tested in a study of the work of Sir John ‘Kyffin’ Williams, a nationally renowned and prolific Welsh artist. Using images gathered from catalogues and online sources, we evaluate image-based descriptors that represent aspects of the paintings themselves: we investigate colour, edge orientation, and texture measures. We go on to estimate metadata from these descriptors using a leave-one-out methodology to classify paintings by year. We also investigate the incorporation of expert knowledge within this framework by considering a subset of paintings chosen as exemplars by a scholar familiar with Williams's work. This work shows a new avenue of research: analysing artefacts using their pictorial features and using this analysis to group and to classify the work directly. Such work is only possible, however, if the underlying data is openly accessible and suitable for analysis by emerging computational tools and methods. Introduction Digitally enabled research in the humanities creates new knowledge through the use of digital content, using tools and ICT-based methods for the analysis and interpretation of this data, and communicating the results of this work to the widest possible audience using traditional and non-traditional publishing methods, allowing greater engagement with research and research data than was previously possible. This has been called e- Wissenschaft , reflecting that the best examples of digital humanities are a new intellectual practice with elements that distinguish qualitatively the practices of intellectual life in this emergent digital environment from print-based practices.