Direct Digital Capture of Cultural Heritage Benchmarking American Museum Practices and Defining Future Needs. Project Report

Transcription

1 Direct Digital Capture of Cultural Heritage Benchmarking American Museum Practices and Defining Future Needs Project Report Roy S. Berns, Franziska S. Frey Principal Investigators Mitchell R. Rosen, Erin Smoyer, and Lawrence A. Taplin Researchers Rochester Institute of Technology April 2005 Supported by a grant from THE ANDREW W. MELLON FOUNDATION i

2 Table of Contents I....Executive Summary... 1 II....Document Current Imaging Practices Using an Online Survey... 5 III....Develop a Quantitative Procedure to Benchmark Image Quality and Evaluate the Procedure at Several Institutions IV....Benchmarking Conference V....Publications and Presentations VI....Personnel VII...For More Information ii

3 I. Executive Summary Many museums, archives, and libraries (hereafter called museums for simplicity) are engaged in direct digital image capture of cultural heritage. This heritage encompasses works on paper, paintings, manuscripts, sculptures, and photographs, among others. For this project, the digitization of photographic collections is excluded. There is a range of technology used for image capture as well as a range of operator expertise. Many systems use components re-purposed for this specific application. As a consequence, there is a wide range of quality of these various image archives. Thus the purpose of this research project has been to benchmark the quality of imaging practices used in representative American museums. The benchmarking was both documentary and experimental. The program commenced July 2003 with financial support from the Andrew W. Mellon Foundation. The project had nine major components: 1) online survey of institutional photography departments; 2) interview of key digital imaging personnel from a selection of departments; 3) compile and summarize documentary standards on imaging quality; 4) develop a quantitative testing procedure; 5) administer the test at representative institutions; 6) organize and hold a conference; 7) fully analyze information and document this program; 8) disseminate information through publications and presentations; and, 9) synthesize a report that clearly explains the findings. Reports stemming from this project and material from this and related projects can found by visiting the home site of the Museum Imaging Research Team: and Document Current Imaging Practices Using an Online Survey More than 50 cultural heritage institutions from throughout the United States participated in the American Museums Digital Imaging Benchmark Survey investigating the use of digital photography for direct capture of artwork within US museums, libraries and archives. The survey was available online for a year. Facts and opinions were sought to learn the history, current status and future of digital imaging within the institutions. Composed of 78 questions, the survey built a comprehensive picture of the staff, equipment, software and workflow of the organizations and captured the attitudes of the respondents. The key findings were: a. Over half of the respondents took at least 90% of their photographs digitally last year. b. 94% of the respondents are increasing the use of digital photography. This goes hand in hand with investing in new equipment, staff hires to support digital, and applications for grants to support transitioning to a digital workflow. c. More than half of the respondents report a perceived lack of knowledge about digital, however, a majority feels comfortable with their institutions embracing digital photography. d. To make collections accessible over the Internet, to include digital images in a collection management system, and to produce printed reproductions are the top three reasons for taking digital images. e. The majority of responding institutions are using high-end cameras capable of delivering the same quality as a traditional 4x5 studio camera. f. More than 50% of the participants are regularly calibrating their camera systems g. Most of the institutions are processing their images. The three categories to choose from were: visual editing, retouching, and sharpening. Very few use automated processing. h. 66% of the respondents are using the TIFF format for their master files. i. Close to half of the institutions are keeping the original camera raw file. 1

4 j. 88% of the respondents answered that they routinely backup their files. However, this also means that 12% do not. k. It is remarkable that close to 50% of the institutions are using media that are known be very temporally unstable and are not recommended for longer term storage. 23% of all institutions report that at some point they lost a digital master that was unrecoverable. Recommendations for the future are: a. Educate photographers in cultural heritage institutions on various issues of digital imaging. b. Educate users on long-term effects of imaging for a specific output. This includes color, spatial image processing, and compression. c. Make the community aware of the shortfalls of current archiving strategies. d. Build a community including scientists, photographers, conservators, and curators, allowing for a rich and unique interdisciplinary exchange. Develop a Quantitative Procedure to Benchmark Image Quality and Evaluate the Procedure at Several Institutions A testing procedure was designed for characterizing both the color and spatial image quality of trichromatic digital cameras, which are used to photograph paintings in cultural heritage institutions for the purpose of creating archival quality digital master images. The testing procedure was target-based, thus providing objective measures of quality. The majority of the testing procedure followed current standards from national and international organizations such as ANSI, ISO, and IEC. The procedure was developed in an academic research laboratory and used to benchmark four representative American museum s digital-camera systems and workflows. The four museums were chosen because they were early adopters of digital-image archiving. The nine quality parameters tested included system spatial uniformity, tone reproduction, color reproduction inaccuracy (spectral sensitivity and target-based), noise (image and color), dynamic range, spatial cross-talk, spatial frequency response, color-channel registration, and depth of field. In addition to the characterization testing, two paintings were imaged and processed through each museum s normal digital workflow. The results of the four case studies showed many dissimilarities among the digitalcamera systems and workflows, which caused a significant range in the archival quality of their digital masters. These differences point out the need for standardization of digital imaging in American museums, libraries, and other cultural-heritage institutions. The key findings were: a. Quality standards exist to evaluate spatial uniformity, tone reproduction, spectral sensitivity, noise, dynamic range, spatial cross-talk, and spatial frequency response. b. Standards do not exist to evaluate color-reproduction accuracy, color-channel registration, and depth of field. c. It is possible to develop a single experimental procedure to evaluate color and spatial quality. d. There were considerable differences in color and spatial quality among the tested institutions. e. There is not a common workflow among the tested institutions. f. Museums often include visual editing on a computer display as part of their workflow from object to image archive. 2

5 g. Libraries rarely include visual editing on a computer display as part of their workflow from object to image archive. h. Visual editing added significant time to the entire workflow. i. Visual editing tended to increase the chroma of the archived image rather than improve its color accuracy. j. Some camera systems used routinely for digital photography intrinsically have low color quality and high spatial quality. Recommendations for the future are: a. Develop a standard practice to characterize a digital imaging system s color and spatial quality. This standard practice would be unique to cultural heritage digital imaging. b. Use this standard practice to leverage improvements in commercial digital cameras. c. Eliminate visual editing from the museum imaging workflow. d. Standardize the meaning of a raw image file. Benchmarking Conference RIT hosted the American Museums Digital Imaging Survey Benchmarking Conference from September 21 to September 23, 2004 for two main purposes. The first was to disseminate the results of the online survey, the test method to quantify color and spatial quality of digital-imaging systems used for the direct capture of cultural heritage, and the results of using the test method to benchmark four representative institutions. The second purpose was to bring together American leaders and practitioners of digital imaging of cultural heritage and have a dialog to document their opinions and list their needs for the future. The conference consisted of invited speakers, panel discussions, moderated discussions, and many opportunities for the program personnel and the participants to interact informally. The key findings were: a. Image quality was often assumed to correlate with file size (and by extension the number of pixels) rather than objective metrics. b. Raw images are often processed rather than the digitized signals from the sensor. c. Long-term data storage is a hidden cost when evaluating a digital workflow. d. Many imaging departments (particularly museums) are driven by output catalogs and prints, not developing an image archive. This has consequences for the workflows chosen. e. Quantitative metrics of color and spatial quality are in most cases not the primary selection criteria for a camera purchase. Rather, human-factor criteria and technical support are viewed as more important. f. The ideal digital photographer has in-depth knowledge in information technology and art history and expertise in photographing cultural heritage. g. The cultural heritage field has insufficient leverage to affect advances in technology. Recommendations for the future are: a. Standardize the definition of a raw file. b. Orient imaging practices towards creating an image archive rather than printed documents. This can only be done with the involvement of the publications department. c. Better education of how to define color and spatial quality. d. Education on how to use benchmarking tools. 3

6 e. Build an independent third party service offering quality testing of equipment for institutions. f. Find other industries to partner with to leverage commercial camera manufacturers to develop products with better intrinsic properties for cultural-heritage applications. g. Continue to support research that advances digital photography enabling it to be used as a true surrogate of a work of art. 4

7 II. Document Current Imaging Practices Using an Online Survey More than 50 cultural heritage institutions from throughout the United States participated in the American Museums Digital Imaging Benchmark Survey investigating the use of digital photography for direct capture of artwork within US museums, libraries and archives. The survey was available online for a year. Facts and opinions were sought to learn the history, current status and future of digital imaging within the institutions. Composed of 78 questions, the survey built a comprehensive picture of the staff, equipment, software and workflow of the organizations and captured the attitudes of the respondents. A complete report on the survey results and related papers can found at the team research site: Survey Introduction The American Museums Digital Imaging Benchmark Survey went live on the RIT web site on October 20, 2003 and terminated a little less than 12 months later. The survey was designed to complement other worldwide efforts 1-4 in deriving a comprehensive picture of how the cultural heritage community is using digital photography. Over the survey period, close to 60 responses were received. Non-U.S. institutions and multiple surveys from the same institution were filtered out, leaving a total of 52 surveys from 21 states. The distribution from throughout the United States is shown in Figure 1 and Table I. Figure 1: States that provided at least one response. Table I: Response frequency by state. State Respondents State Respondents CA 6 MN 2 CO 2 MO 3 CT 3 NC 2 DC 3 NM 1 GA 2 NY 6 FL 1 OH 2 IL 1 PA 4 IN 1 TX 2 MA 6 VA 1 MD 2 WI 1 MI 1 Total 52 5

8 The 78 questions of the survey took about one hour to complete. Participants included many of the major American museums, along with libraries, archives, imaging studios and consultants. That so many took the time to respond to the questions is testimony to a strong need and interest by photographic service providers for a better understanding of digital imaging. The community is looking for answers to the challenges of making initial image captures that are suitable for a wide variety of purposes and long-term documentation. 5,6 The survey questions were aimed at learning about digital imaging for direct capture of artwork. This included the digital photography of paintings and sculpture, but did not include scanning of photographic prints, negatives or chromes. Questions were divided into 10 categories as follows: I. About You: General contact information, II. More About You: Respondent background, III. About Your Staff: Staff backgrounds, IV. Use of Digital Photography: Attitudes on the new technologies, V. Imaging Workflow: Workflow details, VI. Digital Imaging Studio Setup and Equipment: Descriptions of up to 5 studios, VII. Image Editing: Workflow image modifications, VIII. Color Management: Color control in workflow, IX. Digital Master Files: Maintenance of archives, X. Final Questions: Info. sources and comments. Survey Responses Survey Question 20 asked for the year that digital imaging was first used at the institution. Figure 2 is a histogram of the responses. Considering the slope of the cumulative frequency in the plot, 1995 appears to be the year that digital imaging began to see a considerable increase in cultural heritage users. Since that time there has been consistent growth. The timeline illustrated in Figure 2 is consistent with the growth of digital photography commercial studios. In this respect, the museum community has been representative of the greater professional photographic market. median: 1999 median: 90% Figure 2. Histogram of the year institutions began use of digital imaging. Solid curve is cumulative frequency. Figure 3. Histogram of the percentage of photography performed digitally last year 6

9 The Figure 2 histogram shows the median year of digital introduction to have been When asked for the percentage of imaging performed digitally in the previous year, a median of 90% was reported as illustrated in Figure 3. Over half of the 52 respondents performed at least 90% of their photography digitally last year demonstrating an extraordinary shift in an average of only five years. Figure 4 shows average percentage of photography performed digitally last year compared with when digital program was initiated. The data show that within the first five years a majority of photographs was digitally generated. A 3% growth trend of digital usage per year is seen once digital photography is adopted. Percent of Photography Performed Digitally Last Year Average Percentage to 5 2 to 6 3 to 7 4 to 8 5 to 9 6 to 10 7 to 11 8 to 12 9 to to to 15 Years Since Digital Program Began Figure 4. Comparing year of introduction of digital to the percentage of photography performed digitally last year. Trend line shows an average 3% growth per year. Further documentation for fast growth of digital photography within these institutions: 49 of the respondents, 94% of the total, reported that they are increasing their use of digital photography; a count of 32 surveys, 62%, said their departments were investing in new equipment; 14 responses at 27% of the survey population describe recent new staff hires to support the digital photography programs; and 41% of the organizations, 21 in total, have applied for grants to support their transition to a digital workflow. Other indicators are that many reported that they were buying new software, that they were receiving requests for digital images from new users within the institution and that often these new customers came from departments not previously served by the photographic studios. Investment in staff retraining was commonly mentioned on the surveys. Figure 5 looks at the relationship between length of photography experience and digital photography experience for individuals in the survey population. In that graph, respondents are grouped in five-year blocks according to how long they have been practicing photography. For each group, the average years of digital photography experience was calculated and displayed (dark curve). Separately, the average percentage of years spent in the digital medium was also calculated and displayed (light curve). The latter data shows that those who entered the photography field in the last 15 years have spent, on average, at least half that time using digital technology. There is a steep increase for those entering more recently. Those who have five years or less experience have spent almost all of that time in digital photography. 7

10 Comparing Length of Photography Experience to Digital Photography Experience Average Years Digital Experience % 75% 50% 25% 0% Average Percentage of Years Using Digital Average Years Digital Experience Average Percentage of Years Using Digital Years Photography Experience Figure 5. Comparing length of respondent s total photography experience with experience in digital photography. Average years of digital experience: dark diamonds, average percentage of photographic years spent with digital photography: light squares. Survey Question 13 asked whether the respondents felt they knew enough about digital imaging on a scale from 1 to 5 where 1 was assigned to I do not know enough and 5 to I definitely know enough. 56% responded with a 3 or less. See Figure 6. This indicates that many have a perceived lack of knowledge about the new technologies. This may not be surprising since the population reported a median of only five years experience with digital photography whereas it had a median of approximately twenty years of overall photographic experience. On top of this, Figure 4 had demonstrated a fast pace of digital adoption at the institutions. It is encouraging that in spite of uncertainty, lack of experience and fast changes, respondents strongly indicated that they are at peace with digital photography. This can be seen in Figure 7 where 75% chose one of the top two positive answers when asked for their comfort level with the digital direction. 56%: negative to neutral 75%: above neutral Figure 6. Histogram of respondent self-described knowledge of digital imaging. Figure 7. Histogram of respondent comfort with the institution s digital photography direction. 8

11 Table II summarizes the demands being met by digital photography within the museums, libraries and archives represented in the survey population. This is from Survey Question 30 at the beginning of the Section V workflow questions. While much of the purposes fall within traditional documentation and publication workflows, it is interesting to note that the largest responses were in areas not possible without digital images: making collections accessible over the Internet and placing images into a collections management system. Both categories were selected by close to 90% of the institutions as reasons for their use of direct digital photography of their artwork. Table II: Reasons for Taking Digital Images of the Collection (Question 30). Reason Percent To protect vulnerable originals from use 67% To produce printed reproductions 78% To make collection accessible over the Internet 88% To include in a collection management system 87% To document conservation treatment 60% Other 29% The survey left space for five complete studio descriptions from each institution. A total of 92 digital studio configurations were described. On examining the studio descriptions, the researchers divided them into the three categories of high-end, medium-level and low-end. Point-and-shoot systems fell into the bottom category. Systems that delivered the same or better quality than a traditional 4x5 studio camera were classified as high-end. All others fell into the medium category. Several systems were not sufficiently described. Studio system summary is found in Table III. Table III Digital Studio System Categories (Section VI). System Number High-end camera 53 Medium-level camera 27 Low-end camera 8 Camera not sufficiently described 4 Table IV lists the manufacturers of the cameras found in the high-end systems. At this time for American museum imaging departments, it appears that four manufacturers dominate the field: Better Light, Sinar, Phase One and Leaf. Only one of these companies, Sinar, comes out of traditional photography. The other three companies have been solely digitally based since their inception. Table IV: Dominant Camera Systems in High-End Studios (Section VI) scan=linear CCD, area=2-dimensional CCD. Manufacturer Number Better Light (scan) 14 Sinar (area) 10 Phase One (scan) 9 Leaf (area) 7 Phase One (area) 6 Other/ambiguous 7 9

12 Although there is a small number of camera manufacturers represented on the list, there has been steady growth of museum digital studios. As the museum community continues to come together and understand its requirements for camera systems, those camera systems that most closely respond to such demands will likely find the marketplace open to them. The calibration behavior of participants was also captured within the Section VI questions on the survey. More than 50% of the studios were described as following a regular calibration practice. Most indicated the use of a GretagMacbeth Color Checker, either the traditional 24-patch target or the Color Checker DC. Other targets popularly mentioned in survey responses included the Kodak gray scale and the Kodak Color Separation target. Calibration frequencies ranged from several times a session to every 3 to 6 months. Although, as already mentioned, only approximately half of the studios perform calibration, 91% of the studios capture and save targets along with their artwork. Sections VII and VIII included questions on processing applied to photographs once the cameras have delivered data to the system. Respondents were asked to describe manipulations made within their workflow. For Survey Question 49 with results illustrated in Figure 8, the following categories were used: visual editing defined as global changes such as contrast and color balance; retouching defined as local changes and sharpening. Most images produced by the survey community undergo some form of digital processing. Frequency How Do You Process Your Images? 79% Visual editing (e.g., making global changes such as contrast, hue, etc.) 46% Retouching images (e.g., making local changes by making masks, fill-in missing parts, etc.) 58% Sharpening Figure 8. Histogram of ways in which images are processed. Of great interest to the researchers of this study was gaining understanding of the manner in which these processes are applied. Most of the tasks are performed manually. Only 20% of the institutions reported any form of automated processing. Of these, seven of the institutions have developed these time savers by themselves. Without exception, all those using automated processing are still spending some time manually processing images as well, some for as long as 40 minutes on average per image. Figure 9 shows a histogram of the amount of time spent in post-processing images by all institutions. The spread is wide. More than half spend an average of over 12 minutes per image with one out of five of the population investing an average of a half an hour or more to each image for postcapture processing. 10

13 54% of the respondents reported using color management. Of these, 80% said they built their own profiles. Only 14 of the 28 institutions that use color management reported using color measurement instrumentation to check the validity of their profiles. Questions were answered concerning the choice of rendering intents, working spaces and storage color spaces. The most telling answers, though, were in response to the question of whether the respondents know enough about color management. A histogram of responses to Survey Question 14, Do you know enough about Color Management? is found in Figure % categorized themselves as neutral to I do not know enough. It is worth noting that only 2 of the respondents gave themselves highest marks. This negative bias of respondents comfort with color management deserves attention within the community. Figure 9. Histogram of average time spent editing images. Figure 10. Histogram of respondent self-described knowledge of color management. Table V describes where sharpening first occurs. Almost 10% of the imagery is sharpened at capture and another 20% of the highest quality images are sharpened prior to saving. Table V: When is Sharpening First Applied in the Workflow (Question 49). When Sharpening is First Applied Percent At capture 9 Before printing 12 Highest quality image (digital master) is sharpened before saving 6 Other (please specify) 9 Do not know 2 At capture 7 Pro and Con Responses In Section X, survey-takers were rewarded for having gotten through 77 previous questions with an unstructured area to enter impressions of the pros and cons of digital imaging. 45 of the respondents took advantage of this opportunity and offered their thoughts. There was a broad range of topics covered. Example topics that showed up on both the pro and con side of the discussions include the cost of digital imaging; the value of the digital archive; occupational and environmental safety; image 11

14 quality; ease of transmission; and instant turnaround. Other example positive topics dealt with collection repurposing and comprehensive search ability. As negative arguments respondents mentioned steep learning curves; limited server storage space; concurrent demand for film; software bugs; and, lack of standards. One respondent s pro/con comments that cover much of the ground described above are captured in Table VI. Table VI: Example Pro/Con Comments From One Respondent (Question 78). Pro 1) No chemicals killing off photographers, and no more polluting the sewer system with spent chemicals!! 2) Very cost effective (In the long run) compared to buying film, processing, pickup and delivery charges to the lab, and time lost if film returns and is not sufficient for use, causing re-shoots. 3) The ability to capture entire collections for study and internet access is invigorating for museum professionals, knowing that we can share our holdings that previously were inaccessible to the public. 4) I can photograph three times as many objects than with film and have total control of color and management of image files. Con 1) Mind numbing production for those color correcting and cataloging many thousands of images for multiple applications. 2) The perception from administration and other museum managers that digital processes equate to magic...thinking any digital image can be reproduced in an instant, and exist as a thumbnail and as large as the side of the building. 3) The perception that digital processes have no costs associated with production of images for , prints and CD's. Time is still a commodity last time I checked. 4) Image management!!! Far and away, after color management and correction, the most difficult and time consuming aspect concerning production of digital images. Survey Conclusions The American Museums Digital Imaging Benchmark Survey has proven to be a useful exercise for uncovering information about how the photography studios in U.S. museums, libraries and archives are using digital photography for direct capture of artwork. Most of those running these departments have five years or less experience with digital photography and yet over 90% of all photographs taken last year within these institutions were digital. Respondents showed they were still lacking knowledge about the new systems and about critical aspects such as color management, but that they are comfortable with the direction the technology is taking them. A small number of camera manufacturers dominate the high-end of the field. One of the open questions on the capture side is the capture and saving of technical metadata. The results of this survey might help push forward solutions in this area since camera manufacturers need to know what the users are doing and need to implement automatic metadata capture in their systems. 7-9 Many of the digital images taken are destined for traditional workflows but new purposes for these digital images are dominating their uses. In most cases images are being taken to be available for long-term use, making the correct capture and storage an even stricter requirement 10. Photographers are spending a tremendous amount of time editing and manipulating their images. A majority of respondents reported spending 12 minutes or more on each captured image with some institutions spending up to an average of one hour. It seems clear that there 12

15 will be much to gain in improving the imaging process so that required quality is delivered directly, removing the need for such extensive post-processing time and effort. References 1. Kljin, E., and de Lusenet, Y. In the picture; preservation and digitisation of European photographic collections. PUBL/pdf/885.pdf (2000). 2. Sitts, M., Handbook for digital projects: a management tool for preservation and access, Northeast Document Conservation Center, Andover (2000). 3. Rieger, O., Implementing a digital imaging and archiving program: technology meets reality, Proc. IS&T Archiving 2004, pp (2004). 4. Tsai, S. and Lin, W., The workflows of painting and calligraphy on digital archives an example of Hwa-Kang Museum, (in Chinese), Proc. of the Digital Archives Conference on Workflows and Quality Management Proceedings, pp (2004). 5. Berns, R.S., Sneaking scientific validity into imaging tools for the masses. Proc. IS&T First European Conference on Color in Graphics, Imaging, and Vision, pp. 1-2 (2002). 6. Berns, R.S., The science of digitizing paintings for color-accurate image archives: a review, J. Imaging Sci. Technol. 45 pp (2001). 7. Waibel, G. and Dale, R. (2004). Automatic exposure: capturing technical metadata for digital still images. Proc. IS&T Archiving 2004, pp (2004). 8. NISO Z39.87, Technical Metadata for Digital Still Images. (2002). 9. Klijn, E, SEPIADES, Recommendations for Cataloguing Photographic Collections, European Commission on Preservation and Access, Amsterdam (2003). 10. F.S. Frey and S. Süsstrunk, Digital photography - how long will it last?, Proc. of IEEE ISACAS 2000, pp. V-113, (2000). 13

16 III. Develop a Quantitative Procedure to Benchmark Image Quality and Evaluate the Procedure at Several Institutions Introduction For decades, museums, libraries, and other cultural-heritage institutions (referred to as museums in this report) have been using analog photography as a means for documenting their collections and producing reproductions of their artifacts. Through the years, these institutions developed best practices for the process of documentation and reproduction, which included photographing the object, storing the image, and cataloging, so that a high quality image archive could be obtained and maintained for many years. Now that digital photography is well established and comparable to analog photography both in price and image quality, these cultural-heritage institutions have a choice of whether to continue imaging the traditional way or start imaging using digital technology. A number of procedures for testing the quality of digital cameras have been established in the recent past. Unfortunately, there has been no unifying attempt to collect them into a cohesive package for the purposes of comprehensive evaluation of studio imaging environments, particularly those used in museums for the direct digital capture of artwork. The ultimate goals of this research were twofold. First, it is beneficial to the cultural-heritage community because it might provide a possible guideline for high-quality-digital imaging and second, it benchmarked four camera systems and procedures currently used for digital imaging by the cultural-heritage community. Although the saying, You get what you pay for typically applies in the acquisition of imaging systems, there is no substitute for the careful and thorough testing and benchmarking of digital-imaging systems. 1 Benchmarking systems help to compare different camera systems, giving better information than the manufacturers provide, and should lead to a better understanding of the whole imaging process. 2 The aims of the testing procedure were to follow current digital-photography standards to the greatest extent possible, provide only objective measures of image quality by imaging test targets, and be as automating as possible with the use of The MathWorks MATLAB programming language analysis software. The outcome of this procedure was an extensive quantitative description of the digital-image-quality parameters, which characterized four museum digital cameras and procedures used for the direct-digital capture of cultural heritage paintings. Case Study Descriptions The cameras and lights used in the case studies in each of the four museums were different. The four museums were not chosen as case studies for this reason; they were chosen because they were early adopters of digital-image archiving. Table I summarizes the camera descriptions and imaging set-ups of the four museum case studies. 14

17 Table I. Case study camera and imaging set-up descriptions. Case Study I Case Study II Case Study III Case Study IV Camera Leica Phase One Sinar Better Light Camera Body Leica (integrated) TTI 4 5 (on copy stand) Horseman 4 5 Sinar 4 5 CCD Array Tri-Linear Tri-Linear Area Tri-Linear Maximum Native Resolution Lens 100mm f/2.8 Leica 150mm Schneider 100mm f/4 Rodenstock 210mm f/5.6 MC enlarging Apo Sironar digital HR Sinaron SE Filter Illumination Software Leica daylight balancing/ir cut-off 4 Lowel Scandles (approx. 5000K) Silver Fast Leica S1 (ver ) Phase One tungsten balancing/ir cut-off 2 TTI Reflective Lighting tungsten lights, each w/ 4 OSRAM 250W Quartz Halogen photo optic bulbs (approx. 3000K) Phase One Image Capture (ver ) Sinar IR cut-off 4 Speedotron Xenon strobe imaging lights in a 202VF light unit w/ UV correction filter over bulb (approx. 6700K) Sinar CaptureShop (ver ) Better Light daylight balancing/ir cut-off 4 Broncolor HMI F 1200 bouncing the light off of white walls and a 12 ceiling (approx. 5000K) Better Light ViewFinder (ver ) Image Area Cropped Cropped Uncropped Cropped Case Study Testing Procedure and Museum Workflows There were two main parts of the case study testing procedures. The purpose of the first part was to learn about each museum s digital imaging workflows. The purpose of the second part was to characterize their camera systems. In the first part, two paintings (see Figure 1) which were painted with Gamblin Artist Oil paints, were imaged and processed through each museum s normal digital imaging workflow. 15

18 Figure 1. Flower (left) and fish (right) paintings used for the analysis of each museum s digital imaging workflow. Uniform areas of pigment are marked with a white circle. In each case study, the photographer was asked to image the pair of paintings as he would a typical painting in his day-to-day imaging at the museum. This included everything from the setup of the camera and lights to the processing of the image for storage as a digital master. In Case Study I, only a digital master image was created, whereas in Case Studies II IV, the photographer created both a digital master image and a visually corrected image. Figure 2a-d shows the digital imaging workflows performed by the photographer in each case study. Leica Scandles Profile PPT non-linear ProPhoto RGB Digital Master Camera Assign camera profile Apply tone curve Convert to working space profile 16-bit RGB Figure 2a. Flowchart of a.) Case Study I, b.) Case Study II, c.) Case Study III, and d.) Case Study IV digital imaging workflows. A green background in the diamond means that the action was performed in the capture software. A red background in the diamond means that the action was performed in Adobe Photoshop. 16

19 linear curve w/ top at 211dc and bottom at 19dc Phase One Tungsten Camera Camera Apply tone curve Assign camera profile 16-bit RGB -2 red w/ eyedropper on w [232], g [130], k [22] of Qp101 target Digital Master HUAM-RGB- D gamma-7-01 Adjust color balance Adjust levels Convert to working space profile 16-bit RGB -1-2 Visually Corrected Adjust hue Adjust saturation Adjust red curve 16-bit RGB Visual correction under 4963K lights on Barco monitor (6031K white point) Figure 2b. 17

20 Sinar 54H made using CCDC target correct illumination non-uniformity using n-picker in camera soft. Camera Assign camera profile Apply shading reference Neutralize on gray card increase contrast Digital Master ProPhoto RGB amt: 300% radius: 0.3p threshold: 0levels Adjust levels Convert to working space profile Apply un-sharp mask 16-bit RGB +10 Visually Corrected Convert image to 8- bit Adjust saturation 8-bit RGB Visual correction under 5407K lights on Sony Artisan monitor (5198K white point) Figure 2c. 18

21 non-linear curve created using CC grayscale Digital Attributes_22_ Space Digital Master Camera Apply tone curve Convert to working space profile 8-bit RGB Visual correction under 5086K lights on Mitsubishi Diamond Pro monitor (6258K white point) amt: 100% radius: 0.4p threshold: 0levels Adjust tone curve Adjust hue/saturation Apply unsharp mask orange, red, blue-purple, dk. green, flower bkg, entire image Adjust selective color Visually Corrected 8-bit RGB Figure 2d. The colorimetric accuracy of the digital master and visually corrected paintings images were evaluated using 11 uniform areas of pigment (circled in Figure 1) on each painting and compared across the four museums. In addition, spectroradiometric measurements, which were made from the CRT monitors in each case study of the 11 uniform areas of the visually corrected paintings images, were also analyzed. 19

22 The images that were analyzed in the second part of the case study testing procedure were representative of digital masters. In this part of the case study testing procedure, there were nine quality parameters tested. The first one, system spatial uniformity, which assesses the amount of system spatial non-uniformities that can be caused by such things as uneven illumination of the scene and/or lens fall-off, was tested using a uniform gray card target. The second is tone reproduction, which was tested using an ISO standard grayscale target (see Figure 3a) and analyzed in the form of an opto-electronic conversion function, or OECF. The third is color reproduction inaccuracy, which is fundamentally caused by the inherent lack of correlation between the camera s spectral sensitivities and those of the average human observer. Spectral sensitivities were determined by imaging a monochromator instrument (see Figure 3b). Also, nine different color targets were imaged and analyzed (see Figure 3c-j). In addition, observer metamerism was evaluated between the camera and photographer using a tool called the Davidson & Hemmendinger (D&H) Color Rule (see Figure 3k), unfortunately no longer manufactured. The fourth and fifth parameters are noise (image and color) and dynamic range. Image noise and dynamic range were both tested using an ISO standard noise target (see Figure 3l), imaged eight times at the same exposure level. Color noise was tested using selected patches of the Macbeth ColorChecker (see Figure 3c). The sixth image quality parameter, spatial crosstalk, otherwise known as image flare, was tested using an IEC standard target (see Figure 3m). The seventh, spatial frequency response, (SFR) which is used to characterize a camera s ability to reproduce detail, and the eighth, color-channel registration were both tested using the knifeedges of an ISO resolution target (see Figure 3n). Depth of field, the ninth quality parameter that was tested, was tested using a three-dimensional target (see Figure 3o) that had a total depth of 6. The test targets and paintings were approximately the same size, so the camera and lights setup remained consistent throughout the majority of the imaging process. The imaging of the monochromator instrument and depth of field target were the exceptions. Although the basic imaging procedure was consistent for all four of the museum case studies, they were each still unique because the photographer had the freedom to follow his normal imaging procedure. 20

23 a.) ISO OECF target b.) Monochromator Instrument c.) Macbeth ColorChecker d.) Macbeth ColorChecker DC f.) Cobalt blue target g.) Gamblin oil paint target e.) Esser Test Chart h.) IT8 target i.) Kodak Color Separation and Grayscale targets j.) BCRA target k.) D & H Color Rule l.) ISO Noise target m.) IEC spatial cross-talk target n.) ISO Resolution target o.) Depth of field target Figure 3. Test targets used in the characterization of each museum s digital camera system and imaging workflow. Paintings Analysis The paintings in Figure 1 were imaged (and visually corrected in Case Studies II through IV) in each case study as if they were one painting, because they both contained the same pigments. The circled areas of the paintings in Figure 1 were evaluated for colorimetric accuracy by comparing the digital master image data (and visually corrected image data in Case Studies II through IV) to the measurements made with a spectrophotometer. Also, in Case Studies II through IV, spectroradiometric measurements taken from the CRT monitor of the visually corrected image were compared to the spectral measurements of the paintings. Figure 4 21

24 compares the digital master paintings images of each of the four case studies and Figure 5 compares the visually corrected images of Case Studies II through IV. The differences in the colors of the digital master images between the four case studies in Figure 4 were mostly attributed to the cameras spectral sensitivities and the accuracy of the profiles used. Different photographers visually edited the digital master images in Case Studies II through IV, causing the visually corrected images to also look different between these case studies in Figure 5. Figure 4. Digital master paintings images of the four Case Study I, II, III, and IV from left to right. 22

25 Figure 5. Visually corrected paintings images of Case Study II, III, and IV from left to right. Figure 6a 6d shows the hue and chroma errors between the measured spectral data and the digital master image (and the visually corrected image and CRT measurement data in Case Studies II through IV) of the fish painting in the CIELAB color space. The longer the vectors, the more error there was. Case Study I Case Study II Figure 6a and 6b (description below). 23

26 Case Study III Case Study IV Figure 6c and 6d. Figure 6. CIELAB a* (green (-) to red (+)) vs. b* (blue (-) to yellow (+)) error vector plots of the fish painting uniform patch areas between the measured patch data (dot) and the 1. Digital master image patch data (point of black vector arrow), 2. Visually corrected image patch data (point of green vector arrow), and 3. Spectroradiometric measurements of the CRT monitor (point of blue vector arrow). General trends that can be found in a* vs. b* error vector plots are that the errors are in chroma when the error vectors are facing directly inward toward or outward from the origin and the errors are in hue when the error vectors are facing in other directions. Lightness errors can not be analyzed with this type of plot. For Case Study I, the errors were mostly in chroma (the digital master image was more chromatic than the painting) and there were also hue errors in the blue pigments. In Case Studies II through IV, where the digital master images were visually corrected and CRT measurements were made, it would be expected that the green error vectors, which represent the errors of the visually corrected image and start at the vector arrows of the digital master vectors, would be the same length as, but in the exact opposite direction than the black vectors, which represent the errors of the digital master image, that is, aiming back to the spectrophotometrically-measured values depicted by the colored dots. This would be expected, because the photographer visually corrected the digital master image to make it appear the same as the painting, whose measurements are represented by the colored dots. Also, it would be expected, if the monitor was calibrated accurately, that the CRT measurements blue error vectors would point to the same place as the visually corrected image green vectors, which would mean that the photographer would have seen the same colors on the monitor that are in the visually corrected image file. In Case Study II, the digital master image errors were minimal. In general, the green error vectors do not face toward the colored dots, but, instead, face outward from the origin, making the digital master more chromatic with the visual corrections. Also, in general, the blue error 24

27 vectors were facing the same direction as the green error vectors, implying that the monitor calibration was somewhat accurate. In Case Study III, most of the errors in the digital master image were in the red pigments. The visually corrected image, represented by the green error vectors, was more chromatic than the digital master image and painting. For the most part, the blue error vectors did not face the same direction as the green error vectors, which shows how inaccurately the monitor was calibrated. In Case Study IV, the digital master image errors were mostly in chroma. The visually corrected image only corrected the large errors of the digital master image slightly. In general, the blue error vectors did not face in the same direction as the green error vectors, which shows how inaccurately the monitor was calibrated. Table II lists the mean ΔE 00 3 (a CIELAB color difference metric) values between the spectral measurements and the digital master image, visually corrected image and CRT measurement data of both the flower and fish paintings. Color differences are used as a summary metric as they measure only magnitude and not direction. Table II. Case Study mean ΔE 00 of fish and flower paintings 22 total uniform patch areas between the measured spectral reflectance and the CIELAB data of the digital master image, visually corrected image, and the CRT spectroradiometric measurements of the visually corrected image. Digital Master Visually Corrected CRT Measurements ΔE 00 Case Study I 12.3 N/A N/A Case Study II Case Study III Case Study IV

28 Since the purpose of visual editing the digital master image was to improve its color accuracy (make it appear more like the original), it would be expected that the ΔE 00 of the visually corrected image would be less than the digital master image. In Case Studies II through IV, the ΔE 00 of the CRT measurements were slightly smaller than those of the digital master, which shows that the photographer did improve the color of the image on the monitor when he visually corrected the digital master image. Table 2 also shows that the ΔE 00 values of the visually corrected image were greater than the CRT measurements in Case Studies II through IV, which means that the visual improvements in color made by the photographer on the monitor did not get completely stored in the visually edited image file, due to the inaccuracy of the monitor calibration. Target-Based Analyses System Spatial Uniformity Spatial uniformity was analyzed using 6x6 evenly spaced patches of a target, which consisted of two gray cards made from Gray 6.5 Color-aid paper, which were placed side-by-side. The spatial uniformity analysis was similar to that described in the IEC and standards. Figure 7 shows the system spatial uniformity results of the four case studies. 26

29 Case Study I Case Study II Case Study III Case Study IV Figure 7. System spatial uniformity results case study comparison. Luminance factor is a measure of light relating to the perceptions of brightness and lightness is used to evaluate spatial uniformity. Luminance factor is equal to CIE tristimulus Y. These data for each of 36 (6 x 6) evenly spaced patches of the gray card target were compared to the mean image luminance factor of all 36 patches and a percent difference was calculated between them. In Case Study I, light metering was performed during the imaging system set-up. Figure 1 shows that there was slightly more light on the right side of the gray card target and slightly less light on the left and upper edges in comparison to the middle of the image, so the photographer did not light meter the entire image area thoroughly. In Case Study II, no uniformity correction was done, so there was a hot spot of light on the right side of the image and much less light on the left and upper edges in comparison to the middle of the image. In Case Study III, the nonuniformities were corrected using the image capture software, so even though the distribution of the spatial non-uniformity was slightly noisy, the overall system spatial uniformity was adequate. In Case Study IV, the uniformity was checked in the image capture software during set-up, which resulted in the left side of the gray card target being of higher luminance than the right side. From these data, it can be concluded that metering of the entire image area or doing a correction of non-uniformities using the image capture software before imaging helps to reduce system spatial non-uniformities. 27

30 Tone Reproduction During the case studies, the ISO OECF target 6 in Figure 3a was imaged at a nominal exposure, underexposed, and overexposed, so that the target patch image data over the full range of possible digital count values were obtained. The average image target patch values were determined for each exposure level and rescaled to match the nominal exposure level. The OECF functions for each channel are shown in Figure 8. They were fitted with gamma encodings. Some of the case studies had different OECF curves for each channel. The mean gammas of the three channels are listed in Table III for the four case studies. The OECF results from one case study was not necessarily better than that of any another case study. The gamma encoding could have been imposed on the images by the camera s profile or image software. They are all nonlinear, an important property to minimize visual artifacts such as banding and loss of shadow and highlight tonal detail. If the OECF or gamma encoding is known, it can show what the actual gamma encoding of each channel is and if there is any unwanted clipping. Figure 8a. Tone reproduction OECF curves of RGB channels case study comparison. Figure 8b. Case Study II. 28

31 Figure 8c. Case Study III. Figure 8d. Case Study IV. Color Reproduction Inaccuracy Most digital camera spectral sensitivities are not linear transformations of an average human visual system s spectral sensitivities 5,7. That is, through a linear mapping, the camera and a standardized observer see color identically. This is, perhaps, the main underlying reason why color inaccuracies exist in digital images. In the case studies, the monochromator instrument shown in Figure 3b was imaged 36 times from bandpass peaks of approximately 360nm to 730nm in 10nm increments with the imaging lights turned off. After the images were taken, the radiance of the same bandpass peaks were measured with a spectroradiometer. The average image values of the spot of monochromatic light in the centers of each of the 36 images were divided by the radiance values to obtain relative spectral sensitivities. Figure 9 shows the relative spectral sensitivities rotated to fit the CIE standard standard observer sensitivities 8. The lack of fit to the 2 observer can be summarized using a quality metric, µ-factor. 9 The µ-factor was calculated for each case study using the imaging illuminant and camera spectral sensitivities, a D50 viewing illuminant 8, and the 2 observer. These results for the four case studies are shown in Table III. The closer that this value is to unity, the better the correlation of the camera s spectral sensitivities to the 2 observer. A value of zero signifies no correlation. 29

32 Case Study I Case Study II Case Study III Case Study IV Figure 9. Spectral sensitivity case study comparison of relative spectral sensitivities (dotted lines) rotated to fit the CIE 2 standard observer (solid lines). Figure 10 shows a composite image, which is a visual comparison of the color differences of the digital master images of the Macbeth ColorChecker (see Figure 3c) converted into the srgb color space 10,11. Ideally, all four surrounding patches should match the central patch for each color of the ColorChecker. Similar results would occur for the other color targets. These targets included the Macbeth ColorChecker DC (see Figure 3d), the Esser Test Chart (see Figure 3e) 4, a cobalt blue pigment target (see Figure 3f), a Gamblin oil paint target (see Figure 3g), the IT8 target (see Figure 3h) 12, the Kodak Color Separation and Grayscale targets (see Figure 3i), and a target made from ceramic BCRA spectrophotometer calibration tiles (see Figure 3j). 30

33 Figure 10. Color reproduction accuracy comparison of the Case Study I (top left), Case Study II (top right), Case Study III (bottom left) and Case Study IV (bottom right) Macbeth ColorChecker digital master images to the measured data (center) rendered using illuminant D 50. For most of the color patches in Figure 10, Case Studies II and III match the central patch more closely than Case Studies I and IV. The color reproduction inaccuracies of the four case studies can be summarized using ΔE 00. The ΔE 00 value was determined between the average image data of each patch of nine color targets and the spectrophotometrically measured data. The mean 90 th percentiles of all of the patches of each case study are listed in Table III. (We find that the 90 th percentile is more indicative of quality than the maximum value.) The higher this value, the more color error there was. The amount of color difference errors resulting in all four case studies is mostly dependent on the spectral sensitivities of the camera system and to a lesser extent, the camera system s color management profile. Since the spectral sensitivities of the camera cannot be changed, except with the use of filters, the focus tends to be improving profile performance. Therefore, the profiles should be optimized using a target representing the pigments and materials being imaged with the camera. Observer metamerism occurs when two spectrally different stimuli viewed under the same light source match to one observer, but do not match to another observer. The D&H Color Rule tool (see Figure 3k) demonstrates how closely what the photographer sees resembles what the camera sees. It consists of two strips of metameric colors, one labeled alphabetically and one labeled numerically. When placed behind a mask with only one color of each strip showing at a time, these two color strips can be slid back and forth by an observer under a light source until the best match is made. In each case study, the photographer made a match under the illuminant used in the imaging of the targets (in Case Study III, the match was made under the fluorescent illuminant used to illuminate the paintings during the visual correction, because Xenon strobes were used as the taking illuminant). The camera s match in each case study was determined by comparing the CIELAB values between each of the patches of the alphabetic and numeric strips by calculating ΔE 00. The two patches (one from each strip) with the lowest ΔE 00 value were deemed the camera match. Figure 11 shows how the metameric camera and photographer matches differed from each other for each museum case study. There is no correct metameric match that can be made using the D&H Color Rule. The match depends on the light source under which the match is made and the inherent spectral sensitivities of the detector. If the metameric matches were the same for the camera and the photographer in each case study, 31

34 then this would indicate that there would need to be no or a small amount of visual corrections made to the images after digital capture, because the image would already appear and have CIELAB values that were similar or the same as the original painting. Case Study I Case Study II Case Study III Case Study IV Figure 11. Davidson & Hemmendinger Color Rule matameric matches made by thephotographer and camera in each of the four museum case studies. Noise The center three patches of the ISO Noise target 13, shown in Figure 3l, were used to evaluate the image noise. The total average noise of each case study are listed in Table III. These values were calculated according to the ISO standard 13. The lower this value, the less noise the image had. In order to produce images with a low amount of noise, at least one dark correction image should be subtracted from the digital master images. This was done automatically in the image capture software in Case Study III and not at all in the other case studies. Also, using a low ISO and short exposure time when imaging will help in the reduction of the image noise level. 32

35 Color noise is defined as image noise that is color dependent. It is seen as pixel variations in an image of a uniform patch of color. The patches of the Macbeth ColorChecker (see Figure 3c) used in the color noise evaluation were red (#15), green (#14), blue (#13), yellow (#16), magenta (#17), cyan (#18), white (#19), gray (#22), and black (#24). The color noise was evaluated using a metric called the mean color difference from the mean (MCDM) 14. First, the mean of the image pixel data of each color patch was determined. Next, the color difference of the each pixel of the patch from the mean was determined using ΔE 00. Then, the MCDM was calculated from the color differences of all of the pixels. The mean MCDM, or MMCDM, which is the mean of all nine of the selected Macbeth ColorChecker patches, is listed in Table III for each of the four case studies. The smaller the MMCDM value, the less color noise a digital camera produces. The dynamic range, otherwise known as tonal range, of a digital camera system is the capacity of the camera to capture extreme density variations. The darkest and second darkest patches of the ISO Noise target in Figure 3l were used to calculate the dynamic range as a luminance ratio according to ISO standard The log10 of this ratio was calculated to determine the dynamic range density values listed in Table III for all four case studies. It is desirable to have a high dynamic range. A density of 0.3 is equal to one stop of light (log 10 2). In order to obtain the most dynamic range achievable by a digital imaging system, the amount of spatial cross-talk or flare should be reduced as much as possible. Spatial Cross-talk In order to evaluate spatial cross-talk 4, the target shown in Figure 3m was imaged twice. In the second image, the target was rotated 180 so that, for example, a gray patch with a black background in the first image had a white background in the second image. The spatial cross-talk results listed in Table III for the four case studies are the relative maximum percent differences of the 30 gray patches between the two target image rotations. The lower this value, the less spatial cross-talk the digital masters had. In order to reduce the amount of spatial cross-talk, or image flare, in a digital image, the image area surrounding the painting being imaged should be as dark as possible. Spatial Frequency Response The central and upper left corner square horizontal and vertical knife edges, of the ISO Resolution target 15, shown in Figure 3n, were used to determine the digital master images SFR curves 15. The knifeedges were evaluated using Burns sfrmat2 program. 16 The better of the center vertical or horizontal knife edge SFR curves for each case study are plotted in Figure 12. For the scanning linear array cameras (Case Studies I, II, and IV), the SFR caused by the direction of the linear array itself was better than the SFR caused by the direction of the linear array s scanning. The higher the SFR curves in Figure 12 are across the frequency range, the better the preservation of detail at each frequency of the object being imaged. Differences in the SFR curves between the channels were caused by color misregistration in the image or a difference in the image processing that was performed on each channel. Half sampling is at a frequency of 0.5cycles/pixel. This is the Nyquist frequency, which is where aliasing starts to occur in an image. The mean areas under the SFR curves of the three channels across all four edges, which were normalized between zero and unity, are listed in Table III for the four case studies. The higher this value, the better the target s detail was preserved. Un-sharp masking was performed on the digital master in Case Study III, which is why the SFR area was very high in Table III and why part of the SFR curve in Figure 12 exceeds a value of one. However, sharpening increases image noise (the SFR curve for Case Study III is noisier than the other case studies SFR curves); the SNR was very low for Case Study III. The SFR results of the case studies could have been affected by the tool used for focusing the images before capture. In Case Study I, the photographer focused by looking through the ground glass, whereas in the other three case 33

36 studies, a magnification tool or frequency-focusing tool in the image capture software was used to focus the images. Case Study I (center vertical edge) Case Study II (center horizontal edge) Case Study III (center horizontal edge) Case Study IV (center horizontal edge) Figure 12. SFR plots of center edges of each case study. The horizontal or vertical edges that had the best SFR for each case study are plotted. Color Channel Registration The color channel registration was evaluated using the same four knife-edges as in the SFR analysis. It was also evaluated using Burns sfrmat2 program 17. The mean amounts of color channel mis-registration of all three color channels across the four knife-edges are listed in Table III. Mis-registration in images from both scanning and area array CCD cameras can be caused by such things as chromatic aberration of the lens or color filter array lenslets. Depth of Field Depth of field is the range of distance for which the subject is rendered acceptably sharp in an image. It increases as the lens is closed down (f-stop increases). It is greater for short focal lengths than for long 34

37 ones, and it increases with the subject distance. A digital imaging system should have a suitable depth of field when it is used to image paintings because a painting is a three dimensional object that has some depth and a large painting could be warped. The center column of the depth of field target, shown in Figure 3o, was focused on when the image of this target was taken in each case study. The other columns are a total of 3 in front of and behind the center column in 0.5 increments. The SFR of each square s knife-edge on top of each of the 13 columns was determined. The same edges (horizontal or vertical) were used that are plotted in Figure 12. Figure 13 shows the depth of field results of the four case studies as plots of the areas under the SFR curves from frequencies of 0.0 to 0.5cycles/pixel vs. distance on the target. The steeper the sides of the curves are, the less depth of field the case study images had. Also, if the peaks are shifted with respect to the focus aim point, then the focusing tool was not accurate. Case Study I Case Study II Case Study III Case Study IV Figure 13. Depth of field results comparison of the case studies. Summary of Analyses The color and image quality mono-numeric metric values of nine of the quality parameters discussed in the characterization analysis of the four case studies are summarized in Table III. The data listed in Table 35

38 III were obtained from 16-bit digital master images in Case Studies I, II, and III, and 8-bit digital master images in Case Study IV. Quality Parameter Table III. Case study characterization results of nine of the quality parameters and their aim values. Tone Reproduction Mean gamma Spectral Sensitivity µ-factor Target-based Color Reproduction Inaccuracy Mean ΔE th percentile of 9 targets Image Noise Mean total average noise of RGB channels (digital counts) Color Noise MMCDM Dynamic Range Density Spatial Cross-talk Mean relative maximum % difference of RGB channels Spatial Frequency Response Mean area under the RGB curves across all 4 edges from frequencies of 0.0 to 0.5 cy/pixel Color Channel Registration Mean registration shift of RGB channels and across 4 edges Aim Value 1.80 to 2.20 Case Study I Case Study II Case Study III Case Study IV (8-bit) 257 (16-bit) (8-bit) 4.80 (16-bit) The tone reproduction results of one case study are not necessarily better than any other. Typical gamma encodings range from 1.80 to It is preferable that the gamma encodings the same as the working space gamma encoding so that the tonal relationships remain accurate when the image is opened in the future. The working space used in Case Study I, ProPhoto RGB (see Figure 2a), has a gamma encoding of 1.80 for all three channels. The working space used in Case Study II, HUAM-RGB-D gamma-7-01 (see Figure 2b), has a gamma encoding of 2.20 for all three channels. The working space used in Case Study III (see Figure 2c) was the same as that used in Case Study I. The working space used in Case Study IV, Digital Attributes_22_Space (see Figure 2d), has a gamma encoding of 2.20 for all three channels. The spectral sensitivity µ-factor values of Case Studies II through IV were very similar. The µ-factor value of Case Study I was significantly lower than the aim value. The target-based color reproduction inaccuracy ΔE th percentile values are correlated with the color differences seen visually in the Macbeth ColorChecker composite image shown in Figure 10. Case Studies II and II performed well in comparison to Case Studies I and IV. The µ-factor value in Case Study I was low, which means that the camera s spectral sensitivities probably contributed the most to it s target-based color reproduction inaccuracies. The µ-factor value was high in Case Study IV, which means that it s target-based color reproduction inaccuracy was affected greatly by the inaccuracy of its camera system s color management profile. The image noise total average noise values exceeded their aim values in all four case studies. The noise performance of Case Studies I and IV was much better than the noise performance of Case Studies II and 36

39 III. The noise performance of Case Study III was probably affected by the use of the un-sharp masking tool during the imaging procedure. The color noise MMCDMs of the four museum case studies were small and very similar to each other. The color noise results are not correlated with the image noise results. The aim dynamic range density values were theoretical and calculated based on the number of bits in the image. These aim values are only used as a reference point, so the dynamic range of the imaging system can be higher or lower than this value. The dynamic range data of Case Studies I through III were much lower than the aim value. Case Studies I and II had the highest dynamic range values. The dynamic range of Case Study IV was higher than the aim value, which is acceptable. Case Studies II and III had the highest spatial cross-talk mean relative maximum percent difference of the RGB channels and Case Study IV had the lowest. The SFR areas correlate with what can be seen from the curves in Figure 12. Case Study III had the highest SFR, which is mostly caused by the un-sharp masking that was applied to the digital master files by the photographer. Case Study I had the lowest SFR of the four case studies. This could have been attributed to the fact that the photographer focussed the image through the groundglass, whereas in Case Studies II through IV, a magnification tool or frequencyfocusing tool in the image capture software was used to focus the images. The mis-registration errors in all for case studies were acceptably low. The results show that Case Study IV had the least amount of misregistration error and Case Study II had the most. Further Details This report summarizes the quantitative analyses of the four case studies. Greater details, in the form of a Master s thesis, 18 contains a review of standards, a detailed description of how each case study was performed at the museum, a detailed description of how the data were analyzed, the results of each case study, and a comparison of the four case studies. Conclusions The testing procedure described here can be used to provide objective measures of a range of performance characteristics of digital-camera systems and workflows, which are used in cultural heritage institutions to document archival quality digital master reproductions of their collections. Cultural heritage institutions can store future characterization data as metadata with their images. Also, digital camera manufacturers can use this characterization data to see where imaging systems need improvements for cultural heritage applications. As a result of these case studies, many differences were discovered among their current digital-imaging practices, which points out the need for standardization in American museums. None of the four museum case studies had the best results for all of the quality parameters tested. Ideally, a raw digital image should be captured and stored as a digital master with the characterization metadata of the digital-imaging system. This way, the digital information is as accurate as possible and if, in the future, there is an improvement in the way digital data are interpreted, the raw data and information about the means by which it was formed can be retrieved. Cultural heritage institutions should also document their digital-imaging workflows for future reference. When a work of art is digitized, an accurate archival quality digital file should be the goal of the photographer. In other words, the photographer should be careful not to image the painting with a specific reproductive purpose in mind, such a printing and publishing. After the art is imaged and the raw data stored, derivatives can then be made for various applications. 37

40 References 1. Conway, P., Overview: Rationale for Digitization and Preservation. Handbook for Digital Projects: A Management Tool for Preservation and Access; First Edition, 2000: Ch Kenney, A., Digital Benchmarking for Conversion and Access. Moving Theory into Practice: Digital Imaging for Libraries and Archives, 2000: Ch.3: p Luo, M. R., Cui, G., Rigg, B., The Development of the CIE 2000 Colour Difference Formula: CIEDE2000. Col. Res. Appl., 26, IEC : 2001, Multimedia systems and equipment - Colour measurement and management - Part 8: Multimedia colour scanners. 1 st Edition. 5. IEC : 2000, Multimedia systems and equipment - Colour measurement and management - Part 9: Digital cameras. 1 st Edition. 6. ISO 14524: 1999, Photography - Electronic still-picture cameras - Methods for measuring opto-electronic conversion functions (OECFs). 1 st Edition. 7. ISO : 2003, Graphic technology and photography - Colour characterisation of digital still cameras (DSCs) - Part 1: Stimuli, metrology, and test procedures. WD. 8. CIE Publication No. 15.2, 1986, Colorimetry. 2 nd Edition. 9. Vora, P. L., and Trussell, H. J., Measure of Goodness of a Set of Color Scanning Filters, J. Opt. Soc. Am. A, 1993: 10, p IEC : 1999, Multimedia systems and equipment - Colour measurement and management - Part 2-1: Colour management - Default RGB colour space srgb. 1 st Edition. 11. IEC Amendment 1: 2003, Multimedia systems and equipment - Colour measurement and management - Part 2-1: Colour management - Default RGB colour space srgb. 1 st Edition. 12. ANSI IT8.7/2: 1993, Graphic Technology - Color reflection target for input scanner calibration. 1 st Edition. 13. ISO 15739: Photography - Electronic still-picture imaging - Noise measurements, 2003, First Edition. 14. Berns, R. S., Billmeyer and Saltzman s Principles of Color Technology ; Third Edition, John Wiley & Sons, NY, ISO 12233: 2000, Photography - Electronic still-picture cameras - Resolution measurements. 1 st Edition. 16. Burns, P., Slanted-Edge MTF for Digital Camera and Scanner Analysis. Proc. IS&T PICS, 2000: p Burns, P., Williams, D., Using Slanted-Edge Analysis for Color Registration Measurement. Proc. IS&T PICS, 1999: p Murphy, E. P., A Testing Procedure to Characterize Color and Spatial Quality of Digital Cameras Used to Image Cultural Heritage, Rochester Institute of Technology, Rochester, NY,

41 IV. Benchmarking Conference RIT hosted the American Museums Digital Imaging Conference from September 21 23, The first two days were a mix of formal presentations, panel discussions, open discussions, and lots of time to tour MCSL, RIT, George Eastman House International Museum of Photography and Film, and exchange ideas. Approximately 150 people came to the conference from around the world. The third day was optional and hosted by the Image Permanence Institute. About 50 conference attendees stayed for the third day program. The final program schedule is shown in Figure 1. Figure 1. Benchmark Conference program schedule. The American Museums Digital Imaging Conference had invited speakers that included Francesca Casadio of the Art Institute of Chicago (The Use of Imaging in Conservation Science), Ricardo Motta from Pixim Inc. (Emerging Technologies), Peter Burns and Don Williams from the Eastman Kodak Co. (Criteria for Camera Development), Günter Waibel from RLG (Metadata Standards), and Alan Newman from the National Gallery of Art in Washington, DC (The Future of Museum Imaging). Our panelists included Erik Landsberg from the New York Museum of Modern Art, Barbara Bridgers from the Metropolitan Museum of Art, David Remington from the Harvard University Libraries, Andrew Gunther from the Harvard University Art Museums, and Christopher Gallagher from the Art Institute of Chicago. The panel discussion was designed to learn more about the imaging 39

42 practices in each institution, their philosophy towards digital archiving, their successes and hurdles, and needs for future improvement. Panel Discussion Summary The purpose of this panel discussion was to have each of the case study institutions present an overview of their imaging practices. This was followed by questions and answers. Because this report is keeping the case study institutions anonymous, this write up will consist of quotes and paraphrased comments. For several institutions, digital photography began using an outsourced consulting company. This has been replaced with in-house personnel. The role of consultants at each institution was not entirely clear, though several consultants participated in the conference. Image quality was in most cases defined by file size. As camera technology evolved, so had file size. File size was used as a metric of quality. There was a variety of raw files. Institutions have different definitions on what constitutes a raw file. Each institution had a different definition of digital master. This varied from a true raw file to sharpened, color-corrected, and working-space encoded files. The imaging goal is to shoot a work of art once. However, this concept was viewed as naïve because with significant technological advances, re-imaging may be desirable. Digital archiving ranged from CD to centralized file servers. All agreed that data storage was often a hidden cost, not considered in the initial move from chemical to digital imaging. We want the image to look like the work of art. (More about this below.) Photographers are color separators. This comment reinforces that much of digital photography is driven by catalog and print, not archiving collections digitally. Move from chemical to digital was evolutionary. It had to be shown to be equivalent. Small digital projects begat larger projects. In some institutions, only digital imaging is used. RGB and CMYK files are sent for printing. Thus color separation becomes the domain of the imaging department At one institution, a digital color target was included along with files sent for printing as an aid for color separation. At other institutions, the preference was to image a color target along with the work of art. The target would aid printing and provide an image calibration. The final image would crop out the target. There was a lengthy discussion about how to evaluate potential cameras. Subjective evaluation of the camera system including ease of use, feel, software, reputation of the vendor, technical support, compatibility with existing lenses and camera bodies, and allowing photographers to be photographers were the key criteria. Objective measures were of secondary importance. Number of pixels and file size were the main objective criterion in camera selection. This was the main justification of using scanbacks over area arrays. Vendors with proprietary color management, file formats, and so forth were not seen as desirable or/and advantageous. There is a desire for open and standardized procedures and a move towards industry-wide compatibility. Scanners were evaluated more objectively than cameras. Photographers have become information technology (IT) personnel. In many institutions, they are the only person worrying about data storage, backup, etc. There is a clear lack of expertise and IT experts are not being hired to service imaging departments. Institutions where IT issues have been resolved are part of larger computer-driven entities such as research universities. 40

43 There was a lengthy discussion about the required skill set for future hires in imaging departments. Commercial photographers, graphic designers, and computer scientists were not desirable despite their familiarity with the technology. A background in art history is very important; this enables the photographer to understand the needs of the curators. For the leading institutions, years of fine-art photography was necessary; this could be acquired at a museum or auction house. The successful photographer has a visceral understanding of the art; aesthetic judgment is critical. Aesthetic judgment in photographing works of art was discussed. What emerged as a main point was the difficulty in lighting. Because current practices are to create two-dimensional images of three-dimensional objects (even paintings), subjective decisions are required to either accentuate or de-accentuate topographic details. An understanding of how the object is understood as a work of art aides in making proper lighting decisions. This is one of the main reasons for wanting photographers with a background in art history. Digital photography is easier to use to change aesthetics. One aspect was the ease of changing tone and color reproduction characteristics. The pervasiveness of slide libraries with the curatorial and art historian communities is at odds with the move to digital. In some cases, imaging departments are using a digital film recorder to generate 35mm slides. There was a feeling that this was a top-down issue; that is, if museum directors moved to digital approaches and the use of software such as Microsoft Powerpoint, curators would follow. One limitation that remains is the use of dual projectors. Adjacent images in Powerpoint do not have the same quality as two projected slides filling two screens. Quality expectations and prejudices seemed a function of experience. Without question, there is a large range of computer-projector quality (as there are with slide projectors). Open Discussion on User Needs There was an open discussion about user needs with topics ranging from philosophy to practical. A number of practical needs were voiced. They are listed below: Automatic recording of metadata through capture software, Lens development for improved capture of large objects, Off-axis color improvement, Detector development for improved capture of large objects, Automated creation of derivative image, that is, from an archival image (synthesis independent), create images for different applications such as catalog and the web, Appropriate targets and calibration data for imaging cultural heritage, Universal file format for raw files, Definition of raw and what amount of processing by the camera is appropriate for a raw file. Better understanding of color management, Integration of color-profiling software for color management incorporated into camera capture software, Spectral imaging for conservation. Following these specific, practical needs, there was a discussion about how images are used by art historians. One aspect was that, ideally, the digital image should be a true surrogate for the object. We should be able to view it from different angles, to see its surface properties, to look at it in magnified 41

44 form, to see it rendered for different lighting, etc. Clearly, this was a manifesto for future research in imaging cultural heritage, 3-D rendering and image capture at multiple geometries, what is known in some circles as bi-directional reflection distribution functions (BRDF). This will require a significant effort of color, imaging, and computer sciences. Philosophically, there was a dichotomy between libraries and museums. Library-initiated imaging seemed to be oriented towards archiving, that is, the end product is the digital archive. Museum-initiated imaging was production oriented, the end product often a printed catalog. This two world scenario does make sense when taking into account the users of the images and the way the highest-quality images are presented to the users: websites in the case of the libraries, books and catalogues in the case of museums. The role of the curators and the publishing department in the museums has to be taken into account as well. They are the ones making the ultimate quality call. A second philosophical topic concerned the difficulty of museums to leverage products for their application. Certainly most of the high-resolution digital cameras are sold for advertising photography. Nonetheless, the museum and library communities represent a reasonable number of cameras and one that lends itself very well to marketing purposes. Perhaps to underscore the lack of leverage, we were unable to have representatives from the major camera vendors participate in the conference, Betterlight, Sinar, and Phase I. Suggestions for leverage included the medical imaging community and other biological applications. However, these applications, while requiring high spatial and color quality, do not require the high-resolution requirements. 42

45 V. Publications and Presentations Publications RS Berns, LA Taplin, M Nezamabadi, Y Zhao and Y Okumura, High-Accuracy Digital Imaging of Cultural Heritage without Visual Editing, Proc. IS&T Archiving Conference, in press (2005) EP Smoyer, LA Taplin and RS Berns, Experimental Evaluation of Museum Case Study Digital Camera Systems, Proc. IS&T Archiving Conference, in press (2005). MR Rosen and FS Frey, RIT American Museums Survey on Digital Imaging for Direct Capture of Artwork, Proc. IS&T Archiving Conference, in press (2005). MR Rosen, The RIT American Museums Digital Imaging Benchmark Survey, The Digital Archives Conference on Workflows & Quality Management Proc. (2004). EP Smoyer, A Testing Procedure to Characterize Color and Spatial Quality of Digital Cameras Used to Image Cultural Heritage, RIT Masters Thesis (2005). MR Rosen, RS Berns, E. Smoyer, LA Taplin, C. Desimone and FS Frey, American Museums Digital Imaging Survey Benchmarking Conference Program (2004). Presentations EP Smoyer, Experimental Evaluation of Museum Case Study Digital Camera Systems at Council for Optical Radiation Annual Meeting, May EP Smoyer, Experimental Evaluation of Museum Case Study Digital Camera Systems at RIT Center for Imaging Science Annual Industrial Associates Meeting, May EP Smoyer, Evaluation of Case Study Imaging Systems at American Museums Digital Imaging Survey Benchmarking Conference September FS Frey, Digital Preservation at American Museums Digital Imaging Survey Benchmarking Conference, September MR Rosen, Analysis of the Case Survey and Case Studies at American Museums Digital Imaging Survey Benchmarking Conference, September RS Berns, Introduction of the Benchmarking Project and Other Related Projects at MCSL at American Museums Digital Imaging Survey Benchmarking Conference, September MR Rosen, Spectral Reproduction Research for Museums at the Munsell Color Science Laboratory, invited talk at IS&T International Conference on Digital Printing Technologies, September

46 RS Berns, High-Accuracy Digital Imaging of Cultural Heritage without Visual Editing at IS&T Archiving Conference, April EP Smoyer, Experimental Evaluation of Museum Case Study Digital Camera Systems at IS&T Archiving Conference, April MR Rosen, RIT American Museums Survey on Digital Imaging for Direct Capture of Artwork at IS&T Archiving Conference, April FS Frey, Direct Digital Capture, mini-workshop at Museums and the Web 2005, April MR Rosen, The RIT American Museums Digital Imaging Benchmark Survey, invited keynote at The Digital Archives Conference on Workflows & Quality Management, December FS Frey and MR Rosen, Direct Digital Image Capture of Cultural Heritage in American Institutions, interactive session at Museum Computer Network, November

47 VI. Personnel Roy S. Berns R. S. Hunter Professor in Color Science, Appearance, and Technology Munsell Color Science Laboratory Chester F. Carlson Center for Imaging Science Color Science Building Lomb Memorial Drive Rochester, NY (585) Dr. Roy S. Berns is the Richard S. Hunter Professor in Color Science, Appearance, and Technology at the Munsell Color Science Laboratory and Graduate Coordinator of the Color Science master's degree program within the Center for Imaging Science at Rochester Institute of Technology. He received B.S. and M.S. degrees in textile science from the University of California at Davis and a Ph.D. degree in chemistry with an emphasis in color science from Rensselaer Polytechnic Institute. His research includes spectral-based imaging, archiving, and reproduction of cultural heritage; algorithm development for multi-ink printing; the use of color and imaging sciences for art conservation science, and colorimetry. He is active in the International Commission on Illumination, the Council for Optical Radiation Measurements, the Inter-Society Color Council, and the Society for Imaging Science and Technology. He has authored over 150 publications including the third edition of Billmeyer and Saltzman's Principles of Color Technology. During the academic year, he was on sabbatical at the National Gallery of Art, Washington, DC as a Senior Fellow in Conservation Science. During 2000, Dr. Berns was invited to participate in the Technical Advisory Group of the Star-Spangled Banner Preservation Project. He is currently involved in a joint research program in museum imaging with the National Gallery of Art, Washington, DC and the Museum of Modern Art, New York. He is also collaborating with the Art Institute of Chicago and the Van Gogh Museum in digitally rejuvenating paintings that have undergone undesirable color changes. 45

48 Dr. Franziska Frey Assistant Professor School of Print Media College of Imaging Arts and Sciences Rochester Institute of Technology 69 Lomb Memorial Drive Rochester, NY Phone (585) Franziska Frey is a Professor at the School of Print Media at Rochester Institute of Technology. She teaches courses in materials and processes for printing, image database design, and digital asset management, and is involved in research projects in the Sloan Printing Industry Center at RIT, and the Munsell Color Science Laboratory. Franziska is also a Faculty in the "Mellon Advanced Residency Program in Photograph Conservation" at George Eastman House, International Museum of Photography. Franziska received her Ph.D. degree in Natural Sciences (Concentration: Imaging Science) from the Swiss Federal Institute of Technology in Zurich, Switzerland in Before joining the faculty of the School of Print Media, she has worked as a research scientist at the Image Permanence Institute at RIT. Her work has primarily focused on establishing guidelines for viewing, scanning, quality control, and archiving digital images. Franziska publishes, consults, and teaches in the US and around the world on various issues related to establishing digital image databases and digital libraries. She is also involved in several international standards groups. 46

49 Mitchell R. Rosen Research Assistant Professor Munsell Color Science Laboratory Chester F. Carlson Center for Imaging Science Color Science Building Lomb Memorial Drive Rochester, NY (585) Mitchell Rosen is a member of the Munsell Color Science Laboratory and the Visual Perception Laboratory, both of the Center for Imaging Science at RIT. His research is in the areas of color management, spectral imaging systems, museum imaging and eye movement analysis. He teaches graduate courses on color systems and tutorials on color management and color reproduction. He joined the Munsell Lab in 1998 as Senior Color Scientist and was recently appointed Assistant Professor. Previously, Dr. Rosen spent a decade in the research labs of Polaroid working on design and support of digital cameras, scanners, printers and color management systems. He is Color Imaging editor of the Journal of Imaging Science and Technology and active in organizing international conferences on spectral imaging. Lawrence A. Taplin Color Scientist Munsell Color Science Laboratory Chester F. Carlson Center for Imaging Science Color Science Building Lomb Memorial Drive Rochester, NY taplin@cis.rit.edu Erin Smoyer M.S. Color Science Graduate, 2005 Rochester Institute of Technology Currently employed at Texas Instruments 47