While earlier generations usually studied hard copies of the PAM images without any modification, high-quality digitized photos are now available and regularly pass through the PCs of authors and editors before they appear as part of an edition.1 With graphic software readily available for every scholar, it is essential to define best practices in the manipulation and enhancement of images. This chapter is an attempt to define lines of proper conduct in two ubiquitous practices of working with DSS images, which are also employed in the present volume: enhancing images by means of digital filters and the repairing (or “patching”) of images to restore their original state.2
1 Enhancement and Manipulation
Good science requires reliable data, and therefore every published image should be an accurate representation of the actual data. On the one hand, scholars should make sure not to misrepresent the data, that is, not to create new data that does not exist on the physical fragment, while on the other hand they should not lose data during manipulation. Another factor to be taken into account is cognitive perception: the minds of various observers would conceive differently of the level of legibility of one enhancing method over another. In this section we offer some reflection and survey of research methods, and finally list several rules that should be followed for appropriate image manipulation.
The use of simple or advanced filters, together with advanced imaging technology, has brought about several exciting discoveries in manuscript studies, and is now indispensable in our field.3 Given the state of the DSS fragments, significant parts of the data will only be discernible after some manipulation, most commonly the adjustment of brightness, contrast, and sharpness of the image. Methods for legibility enhancement are available and have been used for a long time for various artifacts.4
The legitimacy of image enhancement is a matter of dispute in various disciplines, from the life sciences and medicine to forensic science and intelligence. One can find two general attitudes to this question. On the one hand, scholarship in the field of biology, especially DNA sequences, has been quite conservative. Thus a rather strict protocol was defined for the use of images and adopted by leading journals.5 On the other hand, papyrologists working with Egyptian and Greek materials have been freely indulging with various such algorithms for a long time.6 Proponents of this attitude would say that the preference for the “original” image is no more than a dogma, for “digital reproductions are inherently ‘manipulated’ images, and an image at the exit point of an imaging system is no more faithful than, say, an image calibrated during post-production.”7 Methodological difficulties have been pointed out, however, together with ways to check the skewing effects of various filters.8 Noise reduction, for example, is not always beneficial. While it may hide disturbing factors such as flaws in the skin, dirt, or uneven edges, this very act may also decrease the validity of a reading: one small stain unintentionally removed by a filter may make a difference in the identification of a letter or word. According to the conservative method, while it is acceptable to adjust the overall brightness and contrast of a whole image, such adjustments should not obscure or eliminate any information present in the original and should not introduce new information into it.9 This matter was raised in a systematic treatment of imaging by Bruce Zuckerman.10 As he rightly notes, the computerized act of prioritizing visual elements and constructing a distinct image based on them is not different from the same procedure that takes place in the human eye and mind as part of our “objective” human sight. Furthermore, if the alternative to these enhanced photos is the hand-copies produced by expert human paleographers, then these drawings are no less subjective than the image filters. In the words of Zuckerman:
The graphic representation […] no matter how realistic it may appear to the eye – is no less and no more reliable than a drawing of this section of the text would be. Nor should it be expected to carry more legitimate weight than a more conventional scholarly drawing of an ancient inscription.11
Admittedly, however, the subjective element is more readily recognized in hand-copies made by paleographers than in processed pictures, which still retain an aura of authenticity.
Even the same image will not be identical when observed in two different computers. For example, the operation systems found on Mac and on Microsoft Windows make different assumptions about the gamma settings in the monitor display.12 Such differences, or others like them, would usually arise between the natural eyesight of two human observers. This problem should be kept in mind to qualify what we intuitively define as information present in the original. In light of this, we proceed to using filters – i.e., algorithms for image enhancement – while attempting to establish parameters for regulating this use.
A game-changer is the multispectral technology used by the Leon Levy Dead Sea Scrolls Digital Library and its wide availability. While many of the known enhancements prove useful for the earlier generation of images, it is often the case that the “original” multispectral image supersedes the manipulated products of earlier images. In regular scrolls and fragments, that do not involve a palimpsest or extraordinary damage to the ink, the use of simple tools such as Contrast or Clarity provided by Microsoft Windows should give good results, sometimes even better than advanced algorithm-driven manipulations.13
Rule 1: Adjustments should be applied to the whole image, rather than to a single section of it. Adjusting a section in order to highlight a letter or a feature of the skin will skew the relation between that particular feature and the rest of the fragment and may ultimately create a wrong reading or reconstruction. With readings in DSS editions often depending on a tiny speck of ink, every small defect in the enhancement could be meaningful. Specific features of a fragment should rather be pointed out to the readers using other means, such as drawing arrows or circles on the image surface.
Rule 2: Prefer linear to non-linear adjustments, i.e., those adjustments in which the same change is applied to each pixel according to a linear function. Tools such as Brightness or Contrast14 are therefore more legitimate for enhancing the reading of fragment due to their very nature of linear enhancement. Other filters available through common software such as GIMP of Photoshop alter the pixels according to a nonlinear function, for example by affecting the intensity of specific regions of the image. Such common filters as Sharpness or Clarity function by enhancing the mid-tones of an image, thus yielding stronger contours for the shapes that are otherwise represented in a blurred way. In the biological sciences where work is done with very high-resolution images of minute particles in microscopes, the use of filters from commercial software is not recommended, as they may inadvertently create new factors or eliminate other factors mistakenly deemed less important by the author.15 The situation is different in DSS studies, where even the minutest trace of a letter or a flaw in the skin is significantly larger than the entities observed by biologists. In our experience, using a mild filter such as Clarity to sharpen a given letter does not obscure or eliminate other essential factors of the image, and is therefore legitimate.
Rule 3: Use filters in a complementary way. Recent experiments in the psycho-physics and perception of images have shown, as can be expected, that no one single filter can be embraced as an optimum for reading ancient papyri.16 Every pairing of an individual user with a specific document under specific conditions may give priority to one method or another. Various filters should thus be explored, and the results constantly compared.
Rule 4: Reversibility and accountability. All transformations applied to an image must be reversible. Readings produced by means of a filter will be accompanied by a notation of the software version and filter name. In extreme cases, the filtered image should be displayed next to the original one (in DSS studies it is usually the IR image). In an edition that involves multiple such cases, at least one example should be fully represented and explicated in the introduction.
Rule 5: Practice caution when merging discrete images into one. Such a move is needed when pasting the shape of a letter in a lacuna (cloning) or when joining several pieces into one “fragment.” It is important to verify that the regions were not separately scaled or enhanced before pasted in the artificially constructed image. The reader should be informed of any such action, the source(s) of the respective images, and the steps applied to each of them.
Rule 6: Control the changes of size and resolution in images. Images arrive in a certain resolution, e.g., 300 dpi. An image may be resized without altering its resolution, with the features of size and resolution traded off: the larger the size, the smaller the resolution.17 Some programs allow enlarging the resolution of an image without changing its overall size. When this is done, the computer needs to generate data that are not contained in the original image, thus creating unreliable results. Unmonitored resizing or compression often occurs when converting files to PDF format or when pasting them in a PowerPoint presentation. These steps should therefore be avoided in files intended for publication.
In every case of non-linear adjustment, the best practice is full disclosure of the details and logic of the adjustment. In general, scholars will do well to indicate the program and procedure used for every published image if it is not a 100% reproduction of the original.
One characteristic of DSS studies that eases difficulties which may arise from image manipulation is the public availability of the basic, raw images through the LLDSSDL website and now on the SQE website. While many journals and publishers now require authors to submit their raw data for peer review together with the processed images in the manuscript, this is not necessary in DSS studies since readers can easily check the originals on the web, or submit requests to the IAA staff for further images. Editors who use other photos than those publicly available should report it to their readers accordingly.
2 Digitally Repairing the Fragment
We now discuss a different aspect of image manipulation employed in the present book. The goal of restoration is to produce a single image that resembles the original shape of the fragment as much as possible, in order to support and improve readings and material reconstruction. Since it is no longer feasible in most cases to restore the physical fragment, scholars should achieve it by digital means. The procedure of digital restoration seeks to restore as many pieces of the fragment back into their proper alignment, fixing any damage caused to the fragment over the years. The digital process can fix and restore broken pieces back into their original location, unfold folded edges, repair uneven joining, etc. Information about the fragment’s original shape is obtained from its various images, each image with its own merits, as well as by closely analyzing the deterioration of the fragment since its initial discovery.
Before initiating any digital procedure, its ethical consequences must be considered. Modern digital means can easily change any image, leaving nothing but faint traces of the graphic manipulation. The border between restoration and intervention is not always clear. One must therefore make sure that any graphic action does not result with a “new” fragment, one which never existed in reality. While repairing fragments, the rules delineated above for filtering should be kept in mind. More specifically, changes should be applied evenly to the entire piece of skin at hand: moving the fragment and rotating it are legitimate, but stretching parts of it that have shrunk will yield an unreliable representation of reality. Uneven scaling across various images in a canvas, or uneven scaling within the same fragment, i.e., obstructing the height-width ratio, are illegitimate.
In addition, it is crucial to be as transparent as possible. One is required to keep an exact log of all graphic steps taken in the process and duly report them, making the procedure fully reproducible.18 Repairing (or “patching”) a fragment is subject to the above noted rules with regard to filters and manipulation, as well as with regard to the import of images from various plates into one “new” fragment. There is nothing in the repairing process in general, however, which inherently runs counter to the rules. If properly done and documented, it is a legitimate and helpful pre-processing step.
The restoration begins with collecting and comparing all the graphic documentation of the fragment. It is recommended to digitally locate all images of the fragment side by side on one sheet, in order to diagnose conspicuous changes. The goal of this action is to choose the preferable image to use as the basic image.
The basic image is the image that works best for reconstruction, depending on the state of the fragment and its unique problems. It will usually present the most complete fragment, upon which all manipulations and additions will be performed. The new LLDSSDL images bear several advantages since they were taken in a controlled environment, under documented conditions of illumination, from the same distance and from the same angle. They provide a large amount of visual data which in many cases allows for better readings and better assessment of the fragment’s material properties. However, in numerous cases the old PAM images that were taken closer in time to the discovery of the fragments preserve data that is otherwise lost. In some instances, the newer images document fragments that have entirely deteriorated and crumbled, broken into tiny pieces or simply blackened, rendering their reconstruction not worthwhile. In other instances, when the main goal is creating a physical join, the loss of small pieces from the contours of a fragment may be significant for validating or disproving the suggested join. The process of choosing the basic image, therefore, should be done for each individual fragment.
In his various studies, Zuckerman discussed this procedure, which he calls “Patching.”19 This procedure is employed when the various pieces that comprise the fragment have been moved, deliberately or by mistake, during the various stages of preservation and photography. His main emphasis was to record the stages of patching as distinct layers in Photoshop and to abstain from smoothing the patch, so that the change would be apparent for the observer. The procedure suggested here is similar, with some nuance. The fragment should first be scaled and its background removed (see chapters 4 and 5). If the fragment comprises more than one piece, and the various pieces are improperly placed, the following steps should be followed.
Cut each piece of the fragment from the basic image and paste the pieces back as separate layers. Make sure to keep the pieces at the same orientation as they have been imaged.
Choose one of the pieces as an anchor piece. If possible, this anchor should keep its original position as documented in the original image. It is preferable to choose pieces whose orientation is certain, i.e., a complete line of writing or a piece with right, bottom or top margin, which can be easily and securely aligned.
Adjust all other pieces, piece by piece, according to the anchor. Pay attention to produce straight lines of writings. While in most cases one should make sure that the fragments do not overlap, in some cases the skin has split, showing its flesh and hair sides. In such cases, the hair side of one fragment in the join must cover the flesh side of the other fragment. This is accomplished by defining the hair side as the upper layer in GIMP. When folded or twisted parts of the fragment appear only on the verso side, one should cut the twisted/folded piece from the verso image and paste it as a separate layer into the restored image of the recto.
The basic image should now be compared with all other images. The following questions should be considered: Does the basic image contain all the smaller parts represented on all images? Are there parts whose orientation changed vis-à-vis the old images? Are there signs of shrinkage compared to the old images? Did the restoration affect the shape of the basic image? Are there any folded edges of twisted pieces on the verso? (Such phenomena are especially common in papyrus fragments.)
If the answer to any of the above questions is positive, then the missing or damaged parts should be integrated into the image as separate layers from an image where they are more clearly visible, less deteriorated, or otherwise better fit for the join. The newly added parts should then be adjusted to the main piece according to the guidelines drawn above.
Adhering to the above drawn procedure is important for providing reliable images as infrastructure for later stage of the reconstruction. In this chapter we drafted best practices for the appropriate manipulation of images, specifically in two aspects: the application of digital filters and the repair of composite fragments using graphic software. In the next chapter we will define the need of properly scaled images and the methods for attaining them.
In this volume we use the images supplied to us by the LLDSSDL. Most images correspond to the composite color and IR images that are available on the LLDSSDL website, only with higher resolution. Raking light images of the scrolls are now available on the SQE website.
This section benefitted from the advice of Prof. Roger Easton, Rochester Institute of Technology (May 2019). For a definition of algorithms for image enhancement see Richard Szeliski, “Image Processing,” in Computer Vision: Algorithms and Applications (London: Springer, 2011), 99–204. See updates of this volume in
See for example Roger L. Easton, William A. Christens-Barry, and Keith T. Knox, “Spectral Image Processing and Analysis of the Archimedes Palimpsest,” in The 9th European Signal Processing Conference (EUSIPCO 2011), 1440–44; and the impressive harvest of publications resulting from the “Sinai Palimpsest Project” (
Such is for example the D-Stretch tool (
See especially Mike Rossner and Kenneth M. Yamada, “What’s in a Picture? The Temptation of Image Manipulation,” Journal of Cell Biology 166.1 (2004): 11–15. A more detailed discussion is found in Douglas W. Cromey, “Avoiding Twisted Pixels: Ethical Guidelines for the Appropriate Use and Manipulation of Scientific Digital Images,” Science and Engineering Ethics 16 (2010): 639–67.
See for example Melissa Terras, Image to Interpretation: An Intelligent System to Aid Historians in Reading the Vindolanda Texts (Oxford: Oxford University Press, 2006); Ségolène Tarte, “Papyrological Investigations: Transferring Perception and Interpretation into the Digital World,” Literary and Linguistic Computing 26 (2011): 233–47; Anna Tonazzini, “Color Space Transformations for Analysis and Enhancement of Ancient Degraded Manuscripts,” Pattern Recognition and Image Analysis 20 (2010): 404–17.
Vlad Atanasiu and Isabelle Marthot-Santaniello, “Personalizing Image Enhancement for Critical Visual Tasks: Legibility Enhancement of Papyri Using Color Processing and Visual Illusions: A Case Study in Critical Vision,” International Journal on Document Analysis and Recognition 24 (2021).
Jin Chen, Daniel Lopresti, and George Nagy, “Conservative Preprocessing of Documents Images,” International Journal on Document Analysis and Recognition 19.4 (2016): 321–33.
Rossner and Yamada, “What’s in a Picture,” 12.
Zuckerman, “The Dynamics of Change,” 19–21.
Zuckerman, “The Dynamics of Change,” 20.
Cromey, “Avoiding Twisted Pixels.”
Our preliminary impression from the use of advanced tools such as the Hierax website with LLDSSDL images is that they do not offer substantial improvement, as the images are nearly optimal to begin with. Further experimentation is required.
Rossner and Yamada, “What’s in a Picture,” 14. In this volume we worked mostly with the simple and accessible application Microsoft Photos, which is standard for handling images in Windows 7 and 10.
Cromey, “Avoiding Twisted Pixels.”
Atanasiu and Marthot-Santaniello, “Legibility Enhancement.”
Rossner and Yamada, “What’s in a Picture,” 15.
Zuckerman (“Every Dot and Tiddle,” 188–89; “The Dynamics of Change,” 6–7) defines two levels of graphic manipulation: “invasive” and “noninvasive.” As he claims, the former needs to be reported with a “higher critical profile.” At this time, many academic journals provide guidelines for proper image manipulation. See, for example, ‘PloS One’ figure preparation checklist:
Zuckerman et al., “A Methodology for the Digital Reconstruction,” 48–49; Zuckerman, “The Dynamics of Change,” 5–6; Stökl Ben Ezra, Qumran, 58–60.