Non-invasive analysis and large-scale imaging of murals at a UNESCO World Heritage Site

-Mogao caves, Dunhuang, China-
1/10
Overview

The UNESCO World Heritage Site of the Mogao caves, along the ancient Silk Road, consists of 492 richly painted Buddhist cave temples dating from the 4th - 14th century. The 45,000 square metres of wall paintings in nearly 500 caves are an immense resource for the study of the history of art, architecture, religion, technology, politics and cultural exchange in an area controlled at various points in history by the Chinese, Tibetan, Tangut and Mongol empires. 

 

Previous work undertaken on the Mogoa Caves site had used scaffolding and colour digital photography to photograph paintings in the caves, and encountered several practical challenges; the use of scaffolding in a constrained space, an increased number of personnel on site for safety, and the need for time-consuming post-processing of individual photos.

 

Our collaboration with the Dunhuang Academy was initiated in 2011. Since then, the ISAAC mobile laboratory has visited the site of the caves several times, bringing a range of specially developed portable, in situ and non-invasive equipment to study the murals. Used together, this equipment does not only image the murals; it is capable of characterising the paint deposits as well. 

Addressing the Challenge

Our research was mainly focused on the analysis of murals in cave 465. This cave, located at the northern end of the site, is unique in its Indo-Tibetan tantric Buddhist style. The date of its construction is still under debate. 

 

Because the wall paintings of this historic site are extremely vulnerable, any analysis needs to be non-invasive and non-contact. The geography of the caves and the large size of the murals also means that the use of scaffolding to analyse the paintings is extremely inconvenient. Our ground-based PRISMS visible/near infrared spectral imaging system, which is capable of remote, standoff analysis giving high spectral resolution at a distance of tens of metres, has been successfully deployed on similar projects for several years. Fitted with a telescope for image capture, it enables both large scale surveys, and safe analysis of otherwise inaccessible areas such as ceilings.   

Mogao_instruments.png

The ISAAC mobile laboratory in cave 465, Mogao caves

Large scale surveys in Cave 465

The PRISMS spectral imaging system was used to automatically capture large areas of wall paintings in Cave 465. This generated a huge volume of data; the eastern ceiling of Cave 465 covers an area of approximately 10 square metres, corresponding to a total of around 5000 individual image 'cubes' and over 100 individual spectra. However, this does not represent 100 different original materials; the exposure of the wall paintings to natural environmental conditions over the years has resulted in material changes such as chemical degradation and weathering, which cause many spectral variations in the large dataset.

 

The large volume of data collected during the automatic PRISMS analysis cannot be interpreted manually as it would be far too time-consuming. Instead, we have applied a computer algorithm known as Kohonen Self-Organizing Map (SOM) to process the data. This is a machine learning technique, helping us to automatically identify and 'cluster' pixels which have similar spectral reflectance and are therefore likely to be composed of the same material. These can be then viewed on a computer image of the mural in the form of a false colour map, where each distinct cluster of similar materials is assigned a unique colour on the image.  

 

After clustering, the number of unique spectra over the east ceiling of Cave 465 was narrowed down to 960 clusters with around 300 of them corresponding to areas of physical damage such as cracks or exposed substrate under partially delaminated paint layers. Even though some of these clusters are not useful for a survey of material content, they are useful to inform a conservation survey. 

(a) Colour image of the east side of the main hall of Cave 465; The bright green dots on the ceiling indicate areas in different parts of the east ceiling that share the same spectral information; (b,c) Colour images derived from the mosaic of PRISMS spectral imaging data collected from some of these areas assuming CIE illuminant D65, 1931 2° standard observer, along with (d,e) their corresponding cluster maps. Each cluster is given a unique false colour in the cluster maps. The two false colour cluster maps show several common clusters.

Point analysis in Cave 465

While the reflectance spectrum of each cluster of similar pixels gives a preliminary identification of the materials present, more detailed analysis is needed to confirm the material identification. The false colour maps generated by the automatic clustering of the PRISMS data were used to show us where to conduct point analysis using complementary spectroscopic techniques such as Raman, XRF or FORS. Because point analysis is generally conducted much closer to the surface of the deposit being analysed, we can't routinely perform point analysis in inaccessible areas such as the ceiling. However, we can instead conduct point analysis at a more accessible region of the mural closer to the ground. Where PRISMS automatic clustering has shown spectral similarity between two deposits, however far apart they are on the mural, we can be confident that the materials present will be the same. 

Fig 4 2020 Mogoa paper.webp

(a) Colour image of the eastern panel of the southern wall of the main hall of Cave 465 showing Mahāmāyā and his consort Buddhaḍākinī in union; (b,d) Two zoomed-in areas from the top and bottom of the wall; (c,e) False colour spectral imaging cluster maps showing areas of b and d which share the same spectral information derived from PRISMS spectral imaging data; (f) XRF spectrum and (g) Raman spectrum of a region on the bottom figure (marked by a yellow circle) in one of the clusters represented by the dark green false colour in the cluster maps.

Making a Difference

In the context of this project, a methodology for the automatic analysis of large-scale spectral imaging data was developed, using false colour spectral imaging cluster maps to show areas which have spectral similarities. This rapid classification allows point analysis to be undertaken on some representative areas, but crucially not all, allowing a much quicker and more manageable full characterisation of large painted murals. Moreover, for the first time, a total non-invasive and in situ examination of various degraded paint layers was performed, revealing their original composition. The scientific examination of the murals provided valuable information about the history of Cave 465, placing its dating in the period between late 12th to the 13th century.

References

Kogou, S., Shahtahmassebi, G., Lucian, A., Liang. H, Shui, B., Zhang, W., Su, B. and van Schaik, S. 2020. From remote sensing and machine learning to the history of the Silk Road: large scale material identification on wall paintings. Sci Rep 10, 19312 (2020). https://doi.org/10.1038/s41598-020-76457-9

Liang, H., Lucian, A., Lange, R., Cheung, C.S. and Su, B., 2014. Remote Spectral Imaging with Simultaneous Extraction of 3d Topography for Historical Wall Paintings. ISPRS Journal Of Photogrammetry And Remote Sensing, 95, pp. 13-22.

Lange, R., Zhang, Q. and Liang, H., 2011. Remote Multispectral Imaging with Prisms and XRF Analysis of Tang Tomb Paintings. Proceedings of SPIE, 8084, 80840y

Liang, H., Keita, K., Vajzovic, T., and Zhang, Q., 2008. PRISMS: Remote High Resolution In Situ Multispectral Imaging of Wall Paintings. In: International Council of Museums, Committee for Conservation (ICOM-CC) Triennial Conference, New Delhi, 2008, New Delhi.

Liang, H., Keita, K. and Vajzovic, T., 2007. PRISMS: A Portable Multispectral Imaging System for Remote In Situ Examination of Wall Paintings. Proceedings of SPIE, 6618, 661815

people

Academic Investigator:   

Professor Haida Liang (Nottingham Trent University)

Co-Investigator: 

Professor Su Bomin  (Dunhuang Academy​)

Research Fellows:            

Dr Sammy Cheung (Nottingham Trent University)

Shui Biwen (Dunhuang Academy​)

Zhang Wenyuan (Dunhuang Academy​)

Yu Zongren (Dunhuang Academy​)

Research Assistants:          

Rebecca Lange (Nottingham Trent University)

Andrei Lucian (Nottingham Trent University)

Research Students:         

Sotiria Kogou (PhD, Nottingham Trent University)

Alex Hogg (MSci, Nottingham Trent University)

Stuart Christian (BSc, Nottingham Trent University)