|
This study sets out to analyse the artistic materials used in the maritime Southeast Asian manuscript collection at the British Library. To gain a full understanding of how artistic practises may have developed over time and changed between regions, it is necessary to perform large scale scientific analysis. Visible/NIR spectral imaging is an efficient method of collecting spectral reflectance data which can be used to distinguish different materials. Recent advancements in automatic data collection have meant that the volume of data collected has greatly increased, making traditional approaches to data analysis impossible to perform in a timely manner. Machine learning provides a viable solution to this as it can be used to automatically cluster millions of spectra into smaller, more manageable numbers of distinct spectral groups. Self-organising Maps are used as the building blocks of an algorithm which can perform clustering of large collections of spectral imaging data. Spectral reflectance alone is often not enough to perform pigment identification, consequently other complementary techniques are required. Advances in spectral imaging mean that each of these complementary techniques has a corresponding imaging modality. The machine learning approach developed in this project can be adapted to allow for the clustering of multimodal spectral imaging data including VIS/NIR hyperspectral imaging, macro-X-Ray fluorescence mapping, macro-Raman mapping, and Fourier transform infrared mapping. For multimodal clustering, each modality can be clustered individually and then brought together to produce a single cluster map which is a more refined representation of the material distribution than that produced from any individual spectral imaging modality. A visualisation tool has also been developed for the easy interpretation and interrogation of spectral imaging data cubes and cluster maps for entire collections. Both the visualisation tool and clustering method will be made accessible to the cultural heritage community through an online DIGILAB platform.
|
