Investigation into the impact of tone reproduction on the perceived image quality of fine art reproductions Susan Farnand* a, Jun Jiang a, Franziska Frey b a Munsell Color Science Lab, Rochester Institute of Technology, Rochester, NY 14623 b Preservation and Digital Imaging, The Harvard Library, Cambridge, MA ABSTRACT A project, supported by the Andrew W. Mellon Foundation, evaluating current practices in fine art image reproduction, determining the image quality generally achievable, and establishing a suggested framework for art image interchange was recently completed. (Information regarding the Mellon project and related work may be found at www.artimaging.rit.edu.) To determine the image quality currently being achieved, experimentation was conducted in which a set of objective targets and pieces of artwork in various media were imaged by participating museums and other cultural heritage institutions. Prints and images for display made from the delivered image files at the Rochester Institute of Technology were used as stimuli in psychometric testing in which observers were asked to evaluate the prints as reproductions of the original artwork and as stand alone images. The results indicated that there were limited differences between assessments made with and without the original present for printed reproductions. For displayed images, the differences were more significant with lower contrast images being ranked lower and higher contrast images generally ranked higher when the original was not present. This was true for experiments conducted both in a dimly lit laboratory as well as via the web, indicating that more than viewing conditions were driving this shift. Keywords: image reproduction, tone reproduction, workflow, perceptual image quality INTRODUCTION Research has been conducted previously on the capture stage of the art image reproduction process in which it was found that current capture processes are variable and that these varied processes can result in significant differences in the color appearance and image quality of digital images 1. It was of interest to understand the impact these differences are having on the printed and displayed reproductions of fine art. To explore this question, the perceptual image quality attained by the reproduction workflows in use today, with an emphasis on the output end of the art image interchange cycle, was evaluated. Toward this objective, a series of experiments were conducted, sponsored by The Andrew W. Mellon Foundation, in which cultural heritage institutions captured a variety of artwork targets. Using the capture files created by the participating institutions, prints and images for display were generated and used in psychometric experimentation. The perceptual differences between the hard-copy and soft-copy image quality attained by current cultural heritage imaging workflows in their efforts to reproduce the original artwork were evaluated both with and without the original present. Both situations were considered because museum personnel are interested in ensuring that the reproductions are acceptable representations of the artist s work. However, the most common experience that people will have with fine art reproductions is without the original artwork available for direct comparison. People may be buying printed materials in the museum shop, looking at exhibition catalogues in their homes, or browsing a museum s collection at websites or in databases online. The results of the experiments indicated that, while the presence of the original generally did not significantly impact the rankings for the printed reproductions, having the original present did impact how people ranked the reproductions on computer displays. In other words, reproduction and preference were not significantly different for the printed reproductions but they were for the displayed reproductions. Understanding this difference may be important when determining how to render images on websites and online databases. In this study, we delve further into the differences in the reproductions that were highly ranked when the original was present but not as highly regarded in its absence and vice versa. The resulting analysis was further tested in a brief web-based experiment. Image Quality and System Performance IX, edited by Frans Gaykema, Peter D. Burns, Proc. of SPIE-IS&T Electronic Imaging, SPIE Vol. 8293, 82930H 2012 SPIE-IS&T CCC code: 0277-786X/12/$18 doi: 10.1117/12.907485 SPIE-IS&T/ Vol. 8293 82930H-1
EXPERIMENTAL METHODOLOGY The experiments in this study involved creating target artwork, having participating cultural heritage institutions put this artwork through their established imaging processes, and visually assessing the renditions produced in this exercise. The artwork used in the experimentation conducted in this study included six pictorial works intended to provide a variety of images for exploring the impact of scene content and various image media while keeping the number of test stimuli manageable. All artwork was commissioned or otherwise acquired from sources that allowed for unrestricted reproductions rights. The pieces included an aquatint entitled Still Life, a platinotype photograph, an acrylic painting entitled Firelight, a watercolor, entitled Mountain, and two oil paintings, Daisies and Monet s Bridge, Figure 1. The aquatint and photograph were included to expand the range of media and to represent monochromatic art such as early photography and works on paper, which can be surprisingly difficult to reproduce. [Private discussion] The photograph was obtained from the collection of photographs at the Image Permanence Institute at the Rochester Institute of Technology. The Daisies painting is a relatively light, or high-key, scene. In contrast, the Firelight painting was commissioned to provide an original that was relatively dark overall but covering a wide dynamic range. Many fine art works, particularly the early Dutch masters, fit this description. The Bridge painting was created with a mixture of blues and purples. These colors were included in the study because research has shown some blues are difficult to reproduce due to the differences in the responses of the camera and the human eye to certain blue pigments, especially cobalt blue 2. At least in part for this reason, Monet s Waterloo Bridge has been cited as a painting that has been difficult to acceptably reproduce. [Private discussion] The Mountain painting was commissioned to have the medium of watercolor represented in our set. Again, blue and purple were featured in this work. Fig. 1: The artwork included in the experimentation from top left: Mountain, Inspired by Monet s Waterloo Bridge, Photograph, Daisies, Still Life and Firelight. Objective targets were also included to provide measurements of key image parameters. Included in the set was the Universal Test Target, or UTT, from Image Engineering for assessing color, tone, sharpness, and uniformity at capture, Figure 2. This target is a commercially available target developed to evaluate the quality of the image capture. (Other commercially available test targets used for assessing image quality at capture in fine art reproduction include the Golden Thread target from Image Science Associates.) The pictorial and objective targets were circulated among 17 participating cultural heritage institutions contributed. These institutions included a range of sizes, from the Yale Gallery of the British Arts to the Metropolitan Museum of Art as well as a range of print producers from the Northeast Document Conservation Center, which rarely conducts print SPIE-IS&T/ Vol. 8293 82930H-2
runs of its works, to the National Gallery of Art, which provides of plethora of printed products of its collection and other exhibited artwork. Each institution delivered digital files and information regarding its workflow. Two museums provided multiple sets of files; one provided files for two different cameras and one provided files made in different color spaces. The files were generally provided in the AdobeRGB color space, though there were a few each in the ProPhotoRGB space and the ecirgb color space. Fig. 2: The Universal Test Target produced by Image Engineering. From the files provided, renditions of the artwork and targets, with the exception of the UTT, were printed on the Heidelberg Speedmaster at the Rochester Institute of Technology s Print Applications Laboratory using Prinergy Workflow 3.0.2.2 and Kodak Thermal Gold plates on a VLF 5080 Quantum platesetter and following the ISO 12467 protocol. The same press operator conducted all of the print runs. All prints were made on NewPage Sterling Ultra 80# Matte Text paper with Biolocity inks. The digital files obtained from fifteen of the seventeen participating institutions were used to generate the soft copy images. One institution supplied files for two different cameras; both of these files were included. The images delivered were generally encoded in srgb or AdobeRGB color space, although a few were in ecirgb or ProPhotoRGB space. Because the gamut of the Apple 30 Cinema Display used in the experiments was smaller than that of the AdobeRGB space, images in either the ecirgb or ProPhotoRGB space were first converted to the AdobeRGB color space before being shown on the display. All the images were resized to be of the same size such that two would fit side-by-side on the display. Using the prints and images for display made from the delivered digital files, psychometric experiments were conducted to generate relative visual ratings of image quality with and without the original present. The hard-copy experiments followed a rank order protocol. The soft-copy experiment followed a modified paired comparison procedure, in which the generated images were shown to the observers in pairs and the observers were asked to determine the better reproduction of the original or their preferred image, depending on whether the original was present or not. For both experiments the rankings provided by the observers were translated into z-scores following procedures outlined in Engeldrum 5 and based on Thurstone s Law of Comparative Judgments Case V 6. For further details on the experimental procedure, please refer to earlier publications. 7-9 The experiment was conducted in the Perception and Display Lab in the Color Science building at RIT. The hard-copy experiments took place in a viewing booth under D50 lighting conditions in the Display and Perception Lab in the Color Science building at RIT. The correlated color temperature of this booth is 4911 o K, as measured with a PhotoResearch 650. The experimental setup is shown in Figure 3. The soft-copy set up is shown in Figure 4. To ensure accurate colorimetric reproduction, the display was characterized using an LMT 1210 colorimeter and following the procedure outlined by Day, Taplin and Berns. 4 The display white point and luminance were adjusted to match with those of the light booth by using a Halon perfect reflecting diffuser (PRD). Additionally, the luminance and chromaticity of the background of the light booth were measured using a PhotoReserach-650 spectroradiometer. The background of the SPIE-IS&T/ Vol. 8293 82930H-3
software interface was adjusted to match these settings. The colorimetric performance of the display was evaluated. The mean and maximum color differences were 0.62 and 1.42, respectively. Given that the white point of both the AdobeRGB and srgb spaces is D65, the same as that of the display, no chromatic adaptation transform was performed. Display digital counts were generated from tristimulus values of test images using the display model. When the original was present, in was shown in a D65 booth adjacent to the display. The general setup is shown in Figure 4. Following the hard-copy and soft-copy testing, a web-based experiment was conducted. For this experiment, a web application that included a graphic user interface and a back-end database was implemented to allow testing of image preference of fine art reproductions in an uncontrolled environment. Observers from around the world were invited to take part in the experimentation. Anyone with adequate Internet speed and a web browser (and half an hour) could take part. The interface was similar to that used in the soft-copy experiment without the original, although in this experiment, observers were asked to select their preferred image by to clicking in the area of the image that most influenced their decisions. Fig. 3: The hardcopy experimental setup. Fig. 4: The softcopy experimental setup. The observers who participated in the experimentation generally consisted of individuals in the cultural heritage community. The observers included librarians, photographers, curators, art teachers, conservators, and imaging science students and staff with an interest in fine art reproduction. Observers were tested for color vision anomalies with the Ishihara plates. A total of 24 observers participated in the experiments with the original image present and 14 observers participated in the experiments without the original including 10 that had participated in the workflow experiment and, therefore, were well-acquainted with the original artwork. The remaining four observers, however, had only limited knowledge of the originals. Eighty-eight observers participated in the web-based experiment. EXPERIMENTAL RESULTS AND DISCUSSION The experiments yielded soft-copy (on screen) and hard-copy (in print) results both with and without the original present during testing. For the hard copy experiments with and without the original present, the results averaged over all of the artwork, Figure 5, indicate that the presence of the original had limited impact on the ranking results for most of prints. The results are highly correlated for most of the workflows. Two workflows that ranked above average with the original present ranked below average without the original available (gold squares in Figure 5). In the soft-copy experiment, the presence of the original had a more significant effect on the relative ratings, Figure 6. While for many of the workflows, the impact of the presence of the original was minimal (blue diamonds in Figure 6), the relative ratings for three workflows that ranked above average with the original present are lower, sometimes much lower, when the original is not present (gold squares). In fact, the workflow rated the highest when the original was present was ranked lowest when the original was absent and the second highest workflow when the original was present was ranked below average when it was not. SPIE-IS&T/ Vol. 8293 82930H-4
Z-score without original 0.4 0.2 0-1.5-1 -0.5-0.2 0 0.5 1-0.4-0.6-0.8-1 -1.2 Z-score with original Fig. 5: The hard-copy results with and without the original artwork present. Z-score without the original 2 1.5 1 0.5 0-2 -1-0.5 0 1 2-1 -1.5-2 -2.5 Z-score with the original Fig. 6: The soft copy Z-scores results with and without the original present. At least three issues may have contributed to the changes observed with and without the original present. Neutrality was an issue for the less chromatic originals. The Still Life aquatint, the photograph, and the Daisy painting each had a slight hue. Prints that matched this hue were ranked relatively well when the originals were present. However, without the original available, many observers preferred prints having an overall hue that was closer to neutral in both the hardcopy and soft copy cases. Texture was another issue. Observers tended to rank prints depicting some degree of the texture of the original paintings, whether the impasto or the canvas itself, higher when the paintings were present then when the originals were absent, especially for the Bridge, which had a high degree of texture, and the Daisies and Firelight paintings. Perhaps the most interesting issue, however, driving the changes in rankings with and without the original present appeared to be contrast. With a focus on the three more chromatic paintings, Bridge, Firelight, and Mountain, Figure 1, the objective data gathered at capture using the UTT, Figure 2, was examined to better understand why the images for some workflows changed substantially in the softcopy experiment when the original was not present. Among the various image characteristics measured in captures of this target are colorimetric tone reproduction data. In the hard-copy experiment, it was determined that ΔL at capture was the best predictor of relative ranking for most of the workflows. (The ones that were not well-predicted either included extensive manual adjustment to the individual images SPIE-IS&T/ Vol. 8293 82930H-5
or had different photographers for the targets and the paintings.) For further details, please consult previous publications. 7,9 These results, along with visual examination of the images for the workflows that changed so significantly, suggest that contrast may be the underlying cause. Figure 7 shows the tone reproduction curves, measured from the UTT capture, for the two workflows that had the most significant changes. In these graphs, the green area represents values within the acceptable range for tonal reproduction. The red-circled points are out-of-specification. Note that both of these graphs have black values that were insufficiently dark. Also, the tone curve for the workflow that went from the highest rated to the lowest rated (left) was too light throughout nearly the entire dynamic range. The tone reproduction curves for two of the workflows that improved in the rankings when the original was not present are shown in Figure 8. The tone curve on the left indicates a substantial increase in contrast with the darker tones being much darker and lighter tones being slightly lighter. The curve on the right indicates that the mid-tones for this workflow were slightly darker. Generally, workflows with higher contrast and slightly darker tones performed better when the original was not present in the soft-copy testing while workflows with lower contrast dropped substantially in the ranking without the original available. Fig. 7: The tone reproduction curves as measured on the UTT for the workflow that went from highest to lowest ranked (left) and the workflow that went from second highest to below average (right). Fig. 8: The tone reproduction curves as measured on the UTT for workflows that increased in ranking for some of the originals. The tone reproduction curve for the workflow that went from the third highest ranked when the original was present to slightly below average when it was not is shown in the left graph in Figure 9. From this graph, it is evident that the tone reproduction for this workflow at capture essentially matched the aim curve. One might speculate, then, that the tone reproduction that is optimal for print reproduction may not be as satisfactory for displayed reproductions where the original is not present. While this may be true, there were three other workflows that also essentially matched the aim tone reproduction curve. One of these workflows incorporated post-capture sharpening that increased the local contrast SPIE-IS&T/ Vol. 8293 82930H-6
of the images relative to other workflows and one of the workflows included substantial manual adjustment to each of the individual reproductions. However, the third, Figure 9 right, generally followed a similar workflow to that used by the institution with the tone curve shown in the graph on the left of Figure 9. Both employed limited post-capture image processing. This workflow resulted in images that were generally highly ranked hard- and soft-copy, with and without the original, for the chromatic paintings. For the more neutral artwork, the reproductions for this workflow went from below average to about average with and without the original, respectively. The difference between these workflows may be that a systematic adjustment was applied to the images after capture in the workflow for the institution represented by the graph on the right. So, although post-processing was minimal, the one change that was made was an adjustment to the tone curve, which seemed to work well for the chromatic images and move the more neutral images even closer to neutral. Fig. 9: The tone reproduction curves as measured on the UTT for the workflow that went from being highly ranked to being ranked slightly below average (left) and a workflow that increased in ranking for all of the originals (right). The result that lower contrast images are less appealing when shown on displays in a dimly lit laboratory is not surprising. Imaging scientists have known for decades that photographic transparencies intended for viewing in a dark room need to have substantially higher contrast relative to photographic prints intended for viewing in a well-lit room. 10 What was more interesting was that the soft-copy testing done via the web yielded similar results to that done in the lab; that higher contrast reproductions were preferred over their lower contrast counterparts. It is unlikely that the participants in the web experiment were generally doing the experiment in darkened rooms. To determine whether it was the tone reproduction that was having this impact, a brief verification test was conducted similar to the soft-copy experiment previously described, except that the laboratory lights were not dimmed. In this experiment, four of the originals were used: Mountain, Bridge, Photo, and Firelight. Five new renditions of these originals were generated by boosting the contrast of the lower contrast images, lowering the contrast of images for the workflow represented by the graph on the left of Figure 8 and lightening the images for the workflow represented by the graph on the right of Figure 8. Specifically, the images for the three workflows that dropped in ranking and two that had jumped in ranking were modified using Adobe Photoshop as follows: Workflow 2, Left Figure 7: decrease contrast 25 Workflow 7, Right Figure 7: increase Photoshop brightness 25 Workflow 8, Left Figure 8: increase contrast 25 Workflow 14, Right Figure 8: decrease Photoshop brightness 15/increase contrast 35 Workflow 16, Left Figure 9: increase contrast 25 The unadjusted images along with the images for six other workflows were included in the experiment, for a total of 16 images for each artwork. The results of this verification run are shown in Figure 10. The adjusted image for Workflow 2, which was reduced in contrast relative to the original, shows a sharp drop in performance. The adjusted images for Workflows 14 and 16, which were increased in contrast, show a distinct increase in performance. In other words, the images that started either high or low in contrast had a significant shift in performance with the change in contrast. The SPIE-IS&T/ Vol. 8293 82930H-7
images that had tone reproduction curves close to the aim, Workflows 7 and 8, did not shift significantly when these images were adjusted. The relatively large adjustments that were made in these images had only a minor impact, if any, on ranking. (It should be noted that the Workflows that were not adjusted showed a shift toward average performance in the verification run relative to the original experiment. This may have resulted from the elimination of two of the lowest ranked images in the verification run.) Something other than the overall tone reproduction was causing the shift in ranking performance with and without the original present. For Workflow 7, this may have been inaccuracies in the bluish colors (the Bridge images showed the largest improvement in ranking of the more chromatic originals) and a shift toward neutral for the monochromatic images. For Workflow 8, it may have been the result of a general reduction in color saturation. Further testing would be required to confirm the cause. Z-score 2 1.5 1 0.5 0-0.5-1 -1.5-2 -2.5 1 2 6 7 8 9 11 13 14 15 16 Workflow Exp1 Exp2 adj Fig. 10: The soft copy Z-scores results without the original present before and after adjustments were made. Six workflows that were not adjusted were included for comparison. CONCLUSION Understanding which workflows provide both acceptable representations of the originals as well as pleasing images will be important to understand for those who are striving to provide images of fine art. The results of the experiments performed in this study indicate that there was not much difference in the rankings for the printed renditions when viewed with and without the original present. There was a greater degree of difference with and without the original when viewed in the soft-copy mode. The two workflows most often selected as providing the best representations of the originals were considered the worst, on average, when the original was not present. Investigation into the tone reproduction achieved at capture suggests that people prefer a slightly darker (possibly more saturated) and higher contrast reproduction when displayed on a computer screen. This may indicate that files being prepared for the web may need to be processed differently from those being prepared for print. However, one workflow in the study achieved relatively highly-ranked results for print and display for the chromatic originals. ACKNOWLEDGMENTS The authors wish to thank the Andrew W. Mellon Foundation for supporting this work and the observers for taking the time to help make this project a success. SPIE-IS&T/ Vol. 8293 82930H-8
REFERENCES [1] Berns, R. S., Frey, F.S., Rosen, M.R., Smoyer, E.P.M., and Taplin, L.A., Direct Digital Capture of Cultural Heritage Benchmarking American Museum Practices and Defining Future Needs - Project Report, MCSL Technical Report, (2005). [2] Berns, R. S., The science of digitizing two-dimensional works of art for color-accurate image archives concepts through practice, MCSL technical report (2000). [3] McCamy, C. S., Marcus, H., and Davidson, J. G., A Color-Rendition Chart, Journal of Applied Photographic Engineering, 2, 95-99 (1976). [4] Day, E. A., Taplin, L., Berns, R. S., Colorimetric Characterization of a Computer-Controlled Liquid Crystal Display, Color Research and Application. 29(5) (2004). [5] Engeldrum, P. G., Psychometric Scaling: [A Toolkit for Imaging Systems], Imcotek Press, Massachusetts, (2000). [6] Thurstone, L.L., A law of comparative judgment, Psychological review, 273-286 (1927). [7] Frey, F.S. and Farnand, S.P., Benchmarking Art Image Interchange Cycles, Final project report (2011) [8] Farnand, S.P., Jiang, J., and Frey, F.S., Comparing Hardcopy and Softcopy Results In the Study of the Impact of Workflow on Perceived Reproduction Quality of Fine Art Images, Image Quality and System Performance VIII (Proceedings Volume), SPIE Vol. 7867, (2010) [9] www.artimaging.rit.edu [10] Fairchild, M.D., Considering the Surround in Device-Independent Color Imaging, Color Research and Application. 20(6) (19954). SPIE-IS&T/ Vol. 8293 82930H-9