Big Data in Cultural Heritage Imaging Incubator - Day 1
Francesca Gabrieli, National Gallery of Art
| Jason Zeibel, US Army Night Vision & Electronic Sensors Directorate, discussing IR instrumentation and proccessing methods. |
For the last 20 years, advanced Imaging techniques have been applied to the study of art objects revealing incredible novelties and helping conservators and art historians in their understanding of the composition and history of the objects.
The OSA Incubator Big Data in Cultural Heritage Imaging – hosted by John Delaney, National Gallery of Art (United States); Martin Fischer, Duke University (United States); and David Saunders, British Museum (United Kingdom) – provides the opportunity for scientists from the cultural heritage community, the remote sensing community, and the sensor engineering community to come together to discuss the state of the art in using imaging techniques for cultural heritage applications. The hosts also hope to explore new technologies available in other fields, with the possibility of using those techniques to improve the quality of art object analysis. Ultimately, the participants will be brainstorming on what kind of information is missing in this field and how questions of both conservators and scientists can be answered. (In honor of the topic, we stylized the photos of our hosts.)
As the host John Delaney pointed out, the goals of this meeting are to overview the macro and micro imaging techniques used today in the field of cultural heritage, to describe how the imaging instruments are actually designed and produced, and to understand how to manipulate the extremely large data sets that are produced by these instruments.
Kicking off the meeting was Warren S. Warren (Duke University) with a welcome talk on Wednesday evening before dinner. He showed the how a powerful pump-probe imaging technique can be used in various fields, including medical analysis for melanoma characterization, forensic investigation, and in the study of art object surfaces.
The first session on Thursday was focused on instrument availability and cultural heritage imaging capabilities. John Delaney and Koen Janssens, University of Antwerp, summarized the achievable information level in cultural heritage that could be gained by exploiting the entire optical spectral range with macro imaging techniques; they showed how to apply hyperspectral molecular imaging (visible and near-infrared spectral range) and elemental imaging (MA-XRF and XRD) for the study of paintings. The visible-NIR spectral range is largely used to identify and map inorganic pigments, organic binders and organic dyes. It can also be used to study underdrawings and changes made by the artist, as Delaney showed. XRF and XRD imaging are used to help the conservator in making choices during the cleaning process by mapping how much of an original paint layer is still present underneath the overpaint and also give information about degradation processes.
Later talks were focused on the description of certain imaging cameras either as commercial products or in-house developed instrumentation. James Jafolla, president of Surface Optics, provided an overview of commercial broad-band hyperspectral cameras made by his company, some of which are currently being used by many museums all over the world. Surface Optics is selling instruments that are being used in medicine, geology, art and forensics. Jason Zeibel, US Army Night Vision & Electronic Sensors Directorate, described the IR hyperspectral cameras used for applications of interest to the US Army, such as gas detection, vehicle tagging, and explosives detection. Those same instruments have been used in collaboration with the National Gallery of art for studies of artwork in both reflective and emissive IR wavelengths.
| Ingrid Daubechies, Duke University, explores image analysis for the Ghent altarpiece |
Following the talks, the discussion turned to how to manipulate and manage the large data sets coming from hyperspectral cameras or other imaging modalities. For example, the acquisition of hyperspectral data, combined and registered with high-resolution visible images over the full area of a large work of art represents an extremely difficult processing problem and generates a huge combined data set. This represents a big issue in the cultural heritage field.
Different big-data analysis tools were the focus of the next session. David Messinger from the Rochester Institute of Technology and the Bodleian Library in Oxford, showed a novel approach for the separation of material making up the pigments of a famous, ancient map of England, known as the Gough Map. He showed it was possible to separate multiple compositions of reds or greens on the ancient map, without chemical identification but showing how many different materials were used and spatially identify them. Kathryn Dooley, from the National Gallery of Art, described a method that involved constructing false color images and applying statistical methods, such as PCA (principal component analysis) to visualize changes in a drawing or painting and separating materials that have different reflection properties. For the chemical identification of pigments or binders, it’s necessary to rely on the spectral information; different chemical compounds can be mapped using endmembers (spectra which show a characteristic spectral feature for a high number of pixels) found in an n-dimensional cluster using the spectral-angle mapper algorithm.
A number of speakers stressed how important it is to have any 2-D or 3-D spectral data sets of an art object registered accurately with a high-resolution visible image of the object. This allows both scientists and conservators to correctly interpret the elemental and molecular maps created. A Georgetown University research group under the direction of speaker Murray Loew has worked on the development of a registration tool for this purpose. It provides a graphical user interface for using different algorithms to mosaic and register spectral and other data onto visible images. Discussion about the manipulation of data across different methodologies, which can work or not work depending on individual case studies, was particularly interesting for scientists attending the conference and who were interested in exploiting hyperspectral imaging techniques for both conservation and research purposes.
As the Incubator continues, we will explore micro-imaging techniques to reveal other details about painting layers! Stay tuned for tomorrow’s blog!
Thanks for reading!
