Analysing Deterioration Artifacts in Archival Material
have been developed offering the user greater choice in technical specification.
Data in this study was collected using a Mu-SIS system by Forth Photonics. Pages can be imaged rapidly facilitating real-time examina-tion and a tunable monochromator means that the optimum wavelength for examination can be easily selected. MuSIS contains a sensitive photodetector reducing the amount of illumina-tion required to fall on the object during image capture. Multispectral imaging with a MuSIS system can take photographs at 32 different wavelengths, ranging from the ultra-violet to the near-infrared at 20 nm intervals (420 nm – 1,000 nm) as shown in Fig. 1.
Software: Multispectral imaging extracts infor-mation from a sequence of digital images. There are several interactive visualisation tools avail-able for handling multispectral imaging. Histori-cally most software that processes multispectral data is specific to astronomical or remote sensing applications. Software ranges from commercial (ENVI from ITT Visual Information Solutions), to freeware (MultiSpec (Biehl, Landgrebe 2002)).
These tools were designed to help solve practi-cal problems faced by conservators in libraries, museums, and archives for character segmenta-tion, monitoring of degradasegmenta-tion, evaluation of cleaning methods, enhancing manuscript text, visualisation of palimpsests and for pigment identification.
Images collected with the MuSIS system were
analysed using HSI Labs (Joo Kim, Zhuo, Deng, Fu, and Brown 2010). HSI Labs is an imaging soft-ware program designed in collaboration with the Nationaal Archief of the Netherlands (NAN) and Art Innovation, a manufacturer of hyperspectral imaging hardware. The software is specifically designed for use on vulnerable historical docu-ments where visualisation and analysis methods are required to determine the state of the col-lection item. Features available with HSI ma-nipulation include interactive spectral selection, spectral similarity analysis, time-varying data analysis and visualisation and selective band fu-sion (Seon Joo Kim et al., 2010).
Results
Pigment Identification and Monitoring: Each of the 32 spectral images produced by the MuSIS system is displayed as a monochromatic image representing the percentage of spectral reflec-tance at each pixel for this band. The change in the value of the pixel’s spectral reflectance across the 32 bands can be plotted, and this cor-responding plot is characteristic of the material analysed. This information allows the user to dif-ferentiate between various pigments which may be unknown and compared to those which are known.
Fig. 2 displays three panels showing different representations of the Renaissance illuminated manuscript Add. Ms. 45722: Leaf from Sforza Hours. To the left is the original colour image.
Using HSI Labs spectral data is used to generate
Fig. 2
similarity maps. Similarity maps between the mean of the spectrum of the marked area and the other points in the data are computed by us-ing the entire spectral bands, visible bands and selected bands in the near-IR. The central and right images in Fig. 2 show similarity maps in greyscale and jet colour respectively.
Fig. 3 shows the spectral plot of the points marked 1-8 in the colour image of Fig. 2. The points were chosen based on colour differences observed visually, and were intended to capture a wide range of pigments used in the illumination.
A plot of this data allows the user to compare spectra of different image points to determine their similarity (or dissimilarity). This has appli-cations for measuring the corrosion or ink-bleed severity and separating foreground artifacts from the background of the image or document under analysis.
Photographic Degradation: Historical photo-graphs form an important part of cultural heri-tage collections as their examination allows for the improved understanding of most subjects of interest. They capture moments in time and allow observers to connect with characters and places in the past. Photographs are damaged by direct sunlight, insects, degrading adhesives, nearby sulphur compounds and high humidity which encourages mould growth.
Multispectral analysis has been used recently on daguerreotypes (Goltz, Hill 2012). Degradation of daguerreotypes results in the formation of tar-nish on the highly polished silver surface which
can obscure the graphic content of the image. It was found that the light absorption properties of a photograph with tarnished and untarnished ar-eas had significant differences. These differences allowed for the near-IR camera to image through dirt and heavily tarnished areas. Multispectral analysis can therefore be used as a means of vi-sually showing the conservator how much and which parts of the photograph have the potential to be recovered. Spectral analysis could be uti-lised as a means of monitoring changes to tar-nish and other photographic artifacts allowing preventative measures to intervene immediately.
Fig. 4 shows a 20th century photograph of a girl making her Holy Communion. A written inscrip-tion in ink along the top records the date and event. Three bands (420 nm, 620 nm, 1,000 nm) of the 32 which are produced from the MuSIS instrument highlight the differences observed at different parts of the spectrum. Dirt and impuri-ties visible at 420 nm can be eliminated at 1,000 nm. The ink inscription along the top, which is faded to observers in daylight, appears enhanced and more legible at 620 nm.
From Fig. 4 it is evident that each waveband produced by the multispectral instrument contains different information. While it is sometimes useful to isolate these bands it is also advantageous to combine bands with and provide contextual details in the entire data vol-ume in a process known as fusion analysis. This technique aids in enhancing the legibility of the data. The fusion technique is also useful in that it can remove artifacts on the document such as ink-bleed, ink corrosion, and foxing for research purposes.
Conclusion
Multispectral images are useful for analysing deterioration artifacts in archival material and where possible should be considered as part of the standard condition assessment process. Mul-tispectral data can be used to analyse the effects of environmental aging. It is known that the effects of changes in humidity and temperature and exposure to light induce damage to docu-ments over time. These changes can be system-atically monitored with visualisation tools such as HSI Labs to track the exact process of aging.
Parchment reflectance can be monitored to de-tect degradation before it is visually observed.
Image processing is as important as image capture. Processing and analysing digital images
Fig. 3
offers a non-invasive approach to study and dis-seminate historical documents without the risk of damaging the primary source.
References
Biehl, L. and Landgrebe, D., MultiSpec – A tool for multispectral-hyperspectral im-age data analysis, Computers and Geosci-ences, Volume 28, Issue 10, December 2002, Pages 1152-1159.
Forth Photonics Website, MuSIS HS, http://musis.forth-photonics.com/prod-ucts.php.
Goltz, D. and Hill, G., Hyperspectral Imag-ing of Daguerreotypes, Restaurator, In-ternational Journal for the Preservation of Library and Archival Material, Volume 33, Issue 1, March 2012, Pages 1-16.
ITT Visual Information Solutions: ENVI.
http://www.ittvis.com/ProductServices/
ENVI.aspx, Jun 2010.
Joo Kim, S., Zhuo, S., Deng, F., Fu, C-W. and Brown, M.S., Interactive Visualisation of Hyperspectral Images of Historical Docu-ments, IEEE Transactions on visualisa-tion and computer graphics, Volume 16, Issue 6, December 2010, Pages 1441-1448.
Saunders, D. and Cupitt, J., Image Process-ing at the National Gallery: The VASARI Project, National Gallery Technical Bul-letin, Volume 14, 1993, Pages 72-85.
Figure Captions
Fig. 1: The spectral range of the MuSIS system extends from ultraviolet to near infrared including the visible part of the Electromagnetic Spectrum. From http://musis.forth-photonics.com/
Fig. 2: HSI Labs multispectral image processing of the Renaissance illumi-nated manuscript Add. Ms. 45722, Leaf from Sforza Hours. Similarity maps are a measure between the mean spectrum of the marked area and the other points in the data are computed by using the entire spectral bands, visible bands and selected bands in the NIR. Left: Original RGB image showing a variety of pigments used across the illumination. The num-bers represent spectral plots shown in Fig. 3. Centre: Grey colour similarity map of the Leaf from Sforza Hours, Right: Jet colour similarity plots of the Leaf from Sforza Hours. Multispectral data was captured with the MuSIS system.
Fig. 3: Spectrum plot of the Renaissance illuminated manuscript Add. Ms. 45722, Leaf from Sforza Hours showing the use of multispectral images in pigment iden-tification. 1 = blue clothing, 2 = green clothing, 3 = yellow hat, 4 = grey collar, 5
= baby skin, 6 = gold halo, 7 = brown roof, 8 = red clothing.
Fig. 4: Three bands of the 32 generated by the MuSIS multispectral instrument are shown. Left: At 420 nm the photograph appears dirty and the ink inscription along the top is faded. Centre: At 620 nm the impurities on the surface have been reduced and the ink inscription is enhanced. Right: At 1,000 nm the impuri-ties have are removed and the image is no longer obscured. The ink, however, is not visible at this band. A combination of the 620 nm and 1,000 nm band would provide the optimum fusion.
Author
Christina Duffy, The British Library, 96 Euston Road, London, NW1 2DB, United Kingdom, christina.duffy@bl.uk.
Fig. 4