More information regarding techniques employed in virtual restoration may be found in the DIAMM publication: Image Restoration Workbook (58Mb)

The use of digital images to explore the possibilities of restoration in a virtual environment (i.e. without intervention in the primary source) has been a possibility for some time, though mainstream image processing software has only recently reached a level of sophistication coupled with cost-effectiveness for the processes to become practically available to a wide user-base. The aim of the DIAMM project in developing restoration techniques is to use public domain software that many potential users of the techniques may already own, and others could afford to buy.

There are a number of image-processing applications currently available, but the most widely-used is Adobe Photoshop. The advantage of mainstream software like this is that it is commercially economic for the developers to continue to upgrade and improve the product, and the cost of obtaining it will remain within a range accessible to a high percentage of potential users. Because of its wide customer-base many of the functions available in Photoshop are designed towards a creative graphical end, irrelevant to virtual restoration (VR). However, three tools are of specific value in VR (layers, levels and selective colour), with a host of smaller routines offering additional subtlety.

Firstly, one of the recent additions to Photoshop is the facility that allows any process applied to the original image to be added as a Layer. This superimposes the alteration in much the same way as placing a succession of transparent overlays which can be added or removed:

But this is far more than simply a sophisticated 'undo' feature. As well as being able to switch the visibility of an overlay on and off, it is possible to adjust its opacity and also the manner in which it affects the underlying material.

Sequential overlays can also be shuffled to achieve optimum results. Once an ideal finish is achieved the overlays can be permanently 'flattened' onto the original image, and the result saved alongside the untouched original as a TIFF or other format file.

However, because of the marked variance in restoration needs across even single leaves, one particular 'overlay' may improve one section of a document while disimproving another section, and the interactive fluidity of adding, removing and adjusting layers can become an essential part of the process which cannot be discarded. Photoshop allows an image to be saved (in it's proprietary format) with the layers intact, so that users can continue to switch visibility of layers on and off or adjust their opacity. This is an extremely powerful feature, but has the disadvantage that resulting files can become very large, depending on the number of layers being saved. Individual layers can usually be merged into each other, which can save some space.

Layers can be imported from other documents and superimposed in the same way as layers created internally: ultra-violet images when overlaid on a faded original have proved successful, and in a similar way, performing multiple major alterations to a document, saving it separately and re-introducing the altered image as a semi-transparent layer to the original has produced unexpectedly good results.

Level Adjust is probably the most basic tool offered by Photoshop, and has yielded more impressive results than a whole host of sophisticated filters. Brightness and Contrast adjustments are available and can sometimes be sufficient for sources which are simply badly faded. Level adjust is a more subtle approach to the same problems. The level adjust tool offers a histogram which graphically represents the wavelengths of the colours present in the image. Peaks and troughs indicate colour density:

Here, the full RGB spectrum is covered. By moving the sliders at the base of the diagram inwards, it is possible to spread a narrow colour range over a much wider bandwidth, thus darkening the deeper and lightening the paler colours. The middle slider allows a subtler adjustment of mid-range colours alone. Most often, adjusting the full RGB spectrum is the most useful approach, but the drop-down menu at the top of the dialog box allows the three channels to be adjusted individually. By performing Level Adjust actions within an added layer, the results can not only be switched on and off, but can also be re-adjusted at a later stage, a feature not available when level adjustment is applied directly to the 'background' image.

Finally, Photoshop's Colour Range... selection tool (from the "Select" menu) enables the user to select pixels of a specific colour and also to capture pixels of a similar colour range. Using the pointer to indicate the colour pixels to be selected, the degree of similarity can be specified by the user as value from 1 to 200: the selected area appears as a 'negative' white in the preview box. The selected colour can then either be replaced by a neutral background tone, or faded back using level adjustment.

This has proved the most important tool in lifting layers of dirt, glue or later over-writing from a document by selecting a specific colour, mounting that selection as a layer and applying level adjust to the specific colour selected. The process nearly always has to be repeated several times, and the colours selected more accurately by enlarging the image repeatedly to a point where the pixels begin to break up. Repeating the process with limited selections avoids any colour range extending to material that you wish to remain unaffected.

The same process is used to darken writing that has become too pale to read - using 'select colour' the colours of the note-shapes in this example were isolated, and then darkened by using level adjust in a layer.

The example above was achieved after four successive colour selections and blanket level adjust, but as the aim was readability of the underlying material the results are more dramatic than would be expected with the majority of documents. Where readability is more important than an attractive finish, the ability to separate and manipulate colour channels is a useful tool and serves to emphasize how much hidden information there is in high-resoultion images that the naked eye cannot perceive.

There is always an ethical problem in applying this process:

  1. The 'restorer' has to make decisions from the start about what information he wishes to discard, and what he wishes to retain: thus to some extent the restorer must know (or have decided) in advance which elements on the page are later accretions, and which are the original notation. Inevitably different readings will be obtained by different scholars.
  2. A restored version misrepresents the original source, so care must be taken to ensure the user of the image knows that this image has been digitally processed. The current policy of DIAMM regarding restoration is to make it absolutely clear that any material presented as 're-found' notes cannot be mistaken for the original: notes darkened to improve readability are therefore coloured in a bright primary colour. There is thus no doubt that the material has been 'added', and in fact the material is often easier to read when the colours are less closely approximated to what is thought to be the original.

    Images ©2002 Chapter of San Lorenzo, Florence; enhancement work © 2004
                    DIAMM
    Images ©2002 Chapter of San Lorenzo, Florence; enhancement work © 2004 DIAMM

    Images ©2002 Chapter of San Lorenzo, Florence; enhancement work © 2004
                    DIAMM
    Images ©2002 Chapter of San Lorenzo, Florence; enhancement work © 2004 DIAMM

  3. It is unusual to find a method of enhancement in which some desirable material is not lost along with the undesirable. Often there is no 'optimum' result, and a reading can only be obtained by using a collation of different restorations.

Other Projects using image processing technology are generally looking for repeatable algorhythms that can be applied to a group of sources. The nature of the medieval documents in the DIAMM collection means that no single process can be applied to all the sources with equally satisfactory results. Since colour selection also relies on the hand-positioning of a 'target' over a specific colour pixel, even repeating the same operations on a single page can produce startlingly different results.

More detailed and complex restoration processes are possible using image-processing software, but these are not discussed here. In general, an ability to look beyond the primary purpose of a piece of software and explore new processes is essential in digital restoration, since no one process, or group of processes, can be used on every source. Once familiar with the software and comfortable with the concepts behind digital restoration processes, most scholars can achieve satisfactory results.