Personalizing image enhancement for critical visual tasks: improved legibility of papyri using color processing and visual illusions

Abstract

This article develops theoretical, algorithmic, perceptual, and interaction aspects of script legibility enhancement in the visible light spectrum for the purpose of scholarly editing of papyri texts. Novel legibility enhancement algorithms based on color processing and visual illusions are compared to classic methods in a user experience experiment. (1) The proposed methods outperformed the comparison methods. (2) Users exhibited a broad behavioral spectrum, under the influence of factors such as personality and social conditioning, tasks and application domains, expertise level and image quality, and affordances of software, hardware, and interfaces. No single enhancement method satisfied all factor configurations. Therefore, it is suggested to offer users a broad choice of methods to facilitate personalization, contextualization, and complementarity. (3) A distinction is made between casual and critical vision on the basis of signal ambiguity and error consequences. The criteria of a paradigm for enhancing images for critical applications comprise: interpreting images skeptically; approaching enhancement as a system problem; considering all image structures as potential information; and making uncertainty and alternative interpretations explicit, both visually and numerically.

Supplementary information: The online version contains supplementary material available at 10.1007/s10032-021-00386-0.

Keywords: Color processing; Image enhancement; Image quality; Papyrology; Perceptual image processing; Script legibility.

Fig. 1

Euclid, Elements, Book II, Proposition 5; papyrus of the 3rd–4th century CE from Oxyrhynchus, Egypt. (Credit: P.Oxy. I 29, University of Pennsylvania Museum of Archaeology and Anthropology, CC-BY-2.5.)

Fig. 2

Overview and detail of the original papyri images used in the evaluation. For higher resolution images please refer to the article’s online Supplementary Material. Top: Notice the degradation of the physical documents, the differences in image quality, and the difficulty in comparing details within a limited display space. Bottom: Details scaled to equalize character height, demonstrating the difficulty associated with deciphering such documents. (Credits: (1) Columbia University Library, CC BY-NC 3.0; (2, 4, 5) University of Michigan Library, Public Domain Mark 1.0; (3) Yale University Library, Public Domain Mark 1.0; (6) Courtesy of The Egypt Exploration Society and the University of Oxford Imaging Papyri Project; (7, 8, 9) Istituto Papirologico Vitelli, by permission.)

Fig. 3

Enhancement with (left to right) DStretch options rgb and lbk, and the novel methods vividness and lsv (detail of papyrus 4 of Fig. 2)

Fig. 4

Enhancement results on papyri PSI XIII 1298 (15a) r 1 (top) and P.Oxy.XXII 2309 (bottom). From left to right: the original; five standard enhancement methods; and the new methods proposed in this article (italics). Note the different attenuation of the smudge (top row) and papyrus texture (lower row). The image labels refer to the eponymous methods described in Sects. 4, “Methods”, and 5.4 “Algorithms”, and are abbreviations of (from left to right) original image, stretch limits (stretchlim), histogram equalization (histeq), adaptive histogram equalization (adapthisteq), local Laplacian filters (locallapfilt), retina and cortex (retinex), lightness, saturation, and value (lsv), vividness (vividness), negative vividness (negvividness), and negative lsv (neglsv). Retinex and vividness are established names in the literature, while stretchlim, histeq, adapthisteq, and locallapfilt carry over the function names given in the MATLAB programming language, which was the development environment in the work presented herein. All ninety evaluation images are available in the article’s online Supplementary Material.

Fig. 5

Color values at the same location in an image with sRGB (left) and Adobe RGB (1998) (right) embedded color profiles (detail of Fig. 7).

Fig. 6

The sRGB color gamut volume represented within the Adobe RGB (1998) gamut (gray hull) in the CIELAB color space. The vertical dimension represents lightness, $L\text{*}$, while the horizontal axes correspond to the red-green, $a\text{*}$, and the blue-yellow, $b\text{*}$, opponent colors.

Fig. 7

Color distribution of a papyrus image with native sRGB color profile (top), and following conversion to Adobe RGB (1998) (bottom). (Credits of papyrus image: Sorbonne Université, Institut de Papyrologie, P. Sorbonne Arabe 201 1500a.)

Fig. 8

Legibility varies with foreground/background and background/surround polarity.

Fig. 9

Comparison of three enhancement methods described in this article. The vertical diagram axis represents the CIELAB lightness, $L\text{*}$, the horizontal axis the chroma, $C\text{*}$, and the diagonal the vividness, $V\text{*}$. The variation of color along these axes is illustrated in the bottom part of the figure. For each enhancement type, the diagrams show the location of four exemplary points in a vertical section of the CIELAB color space before and after enhancement. Thus, stretchlim displaces all points towards the extremities of the lightness axis; gamut expansion displaces the points along the chroma axis (except for those situated directly on the lightness axis); finally, vividness applies a rotation transform resulting in all points being collinear with the lightness axis

Fig. 10

The figure depicts an example of the color profile of a papyrus in the HSV color space (the “value” channel and the complement of the “saturation” channel), along with the object classes dominating specific bands of the spectrum (ink pixels, non-inked papyrus surface, and the imaging area surrounding the papyrus).

Fig. 11

Papyrus before (left) and after (right) enhancement with the lsv method (detail of papyrus 8 in Fig. 2).

Fig. 12

From left to right: original, lightness channel $L\text{*}$, and hue-shifted papyri; all have negative lightness polarity (detail of papyrus 4 in Fig. 2).

Fig. 13

Hue shift (left) vs sign change (right), plus negative polarity (detail of papyrus 4 in Fig. 2).

Fig. 14

Clockwise from top left: a Original image, with three faint annotations indicated by an index. b Enhancement by retinex in the RGB color space. c Retinex enhancement in the CIELAB color space. d Blue-shifted and negative of the SE image (detail of papyrus 8 in Fig. 2).

Fig. 15

Example of legibility enhancement by cross-spectral colorization. a Papyrus imaged in visible light; b infrared image; c the infrared, colorized with the chromatic channels of the visible light image and enhanced with the gamut expansion and stretchlim methods; d vividness enhancement of the image (c); e negative of image (d); f MSRCR-RGB retinex of image (c); g negative blue-shift of image (f); h and the decorrelation stretch of the blue, red, and infrared bands.

Fig. 16

Spread of enhancement method rankings and ratings, measured with spatial flatness SF and Shannon entropy H

Fig. 17

The shapes in the center of the squares are perfectly legible, but ambiguous: do they represent the letter “B”, the figure “13”, or something else? Coloring helps differentiate between alternatives. Such shapes are said to be “multistable stimuli” [101], and have been the source of well known illusions, e. g. Arcimboldo’s face made of books [, p. 131]

Fig. 18

The graphical user interface of the Hierax software for papyri legibility enhancement, which implements the methods described in this article. Method selection and parametrization (left); overview of all processed images (center); image comparison and interaction panel (right)