There was a question raised about whether to include color bars while digitizing aged photo albums and Stanley Smith posted a reply. I asked Ken Fleisher of my staff at the National Gallery (who does not belong to this list) to prepare a comment to Stanley to post. This is a core topic of an upcoming Mellon benchmark grant to the RIT School of Printing and Dr. Franziska Frey who will work with some of the museums in ImageMuse. Please come to the ImageMuse panel at MCN for more discussion.
Thanks, Alan Newman. Stanley?s comments are preceded by > and Ken?s replies follow each of the comments. > Sent: Mon 10/20/2008 5:44 PM > > Subject: Re: [MCN-L] photography, digitization, and a color/grey card? > > A couple of comments regarding the use of grayscale and color bars: > > - due to differences between dyes and/or pigments used to make the > color bars and the materials used to produce the actual artwork, > accurately rendering the grayscales in a particular image may NOT > produce the most accurate rendition of the artwork itself. I think it's important to begin introducing the distinction into common usage between scene-referred encoding and output-referred encoding (see definitions at the end of this message). The idea of accurately reproducing "any" target that was captured with the artwork on a printed output, even if there were not issues associated with metamerism (which there are), is in my opinion an incorrect goal to begin with. To accurately capture a target in the digital image implies a scene-referred encoding and is a valid goal in artwork reproduction. By recording the scene information, you have the opportunity to know more about the object itself and to properly transform it as necessary for various types of output. On the other hand, to accurately reproduce that same target on a printed output implies that the goal is to transfer the scene colorimetry to the printed page. This is an incorrect goal because to make the best possible reproduction, it will be necessary to alter the tone reproduction curve (TRC), if nothing else, to reflect the differences in viewing condition associated with a printed image on a white page. For example, there is often a more limited gamut on the printed image so a TRC adjustment will be needed to compensate for the difference in appearance, the white page surrounding the image makes further TRC adjustment necessary, the level of illumination for viewing the printed image is likely different from how the artwork was illuminated during photography which, once again, means additional TRC adjustment is needed. With that out of the way, I have one more comment which I hope to bring into public discussion and awareness. That is the idea of creating an "accurate" reproduction . I understand why everyone says this, and I'm as guilty of it as anyone else, but the fact is that there is no single "accurate" reproduction. There are many renderings of the artwork which can be considered accurate for a given viewing condition. The appearance will be different if you view the artwork in gallery lighting versus 5000K lighting. Which is more accurate? They both are equally accurate. The same holds true even if you keep the same lighting and place the artwork in a white matte and hang it on a white wall versus using a beige matte on a dark gray wall. The artwork will have a different appearance, but both are "accurate". So which one do you reproduce? It is my hope that we will all start using the term "appearance preserving reproduction" rather than "accurate reproduction" and that we will have a defined viewing condition associated with the appearance that we are trying to reproduce. This may seem like a subtle point to some, but I feel it is an important distinction in helping remove ambiguities from our discussions. Further, and more related to the question at hand, it makes it more clear that an appearance preserving reproduction of a target is more important, and more appropriate, than a colorimetrically accurate one. Back to the comment above. It is correct to state that due to differences in pigments, the use of a color target to capture scene-referred colorimetry is a bad idea and will only have limited success. Therefore it is not recommended to use a color target for any capture (Note we are talking about standard targets like the ColorChecker--a case could be made for specialized targets that are known to be spectrally relevant to the object being photographed). On the other hand, an idea I adapted from the conservation guidelines is to include a single color patch simply to indicate that the capture is in color rather than grayscale. With regard to a grayscale target, something like the Kodak Q-13 is not appropriate for capture because it is not made from spectrally neutral materials (it is printed on photographic paper). However a set of step wedges that are spectrally neutral, such as the gray patches of the GretagMacbeth ColorChecker, are very useful to include when capturing artwork, assuming you are aiming for a scene-referred image. To take this to the logical conclusion, a capture of the gray steps of a ColorChecker should NOT be reproduced so that the TRC matches the original scene. In this case, the TRC needs to be adjusted for output to produce an appearance preserving reproduction and therefore I do not feel these gray steps should be included on output. > - If there is some visual editing to tweak color into place (hopefully > under calibrated viewing conditions with direct comparisons to the > original artwork), then another operator who may be printing the image > later will most likely UNDO those edits if they assume that the image is > rendered correctly by printing a neutral grayscale. This is exactly correct, even if we are only talking about spectrally neutral gray patches. Due to metamerism (i.e. the difference in pigments and the difference between how a camera/scanner sees color versus how humans see color), it is almost always necessary to make color adjustments to a digital capture of artwork, even when using the common methods of calibration, and these corrections will skew the patches of a target from their original values. In my opinion, it is still important to capture the spectrally neutral gray steps, but they should only be used a guide for tone reproduction in the scene-referred image. They should not be printed on the output and they should not be used to judge neutrality or correctness of color in the image. > - There is some interest in the production of "virtual" grayscales and > color bars. These would be digitally dropped into the edited and color > corrected image with the same colorspace as the original image. Then > other users would achieve better color when they match the grayscale and > color bars. I am still undecided on the usefulness of this approach, but I am keeping an open mind. On the one hand, if the virtual grayscales look greenish on your print, then this can be a very good indicator that something went wrong. However, there are many things that can go wrong with the color and yet still maintain perfectly neutral grayscales with correct TRC. Therefore, making the assumption that "if the grayscale looks correct on your print then the color is correctly reproduced", this can be misleading and give you a false sense of security. It is my opinion that most people will end up using this as a definitive test for accuracy, rather than only as "one" test for identifying a possible problem. In other words, it can only indicate when something is wrong but will not confirm that everything is right. To understand my point, open an image in Photoshop and convert to sRGB. Add a grayscale, or gradient (it doesn't matter). Now assign (don't convert, "assign") a different colorspace such as ProPhoto RGB. Do you see how the colors have changed radically but the grayscale remains neutral? Yes, the tones shifted slightly, but if the colorspace has the same TRC as sRGB, such as Adobe RGB (1998), you won't even see the tones shift, yet the color is now incorrect. In this case, including "virtual" color bars might be helpful, but I'm also not convinced of this use either, since necessary adjustments to the output will depend on the color gamut and other characteristics of the output device and will again skew the target colors from their original values. This is really getting into the realm of process control now, which may be too much to ask of a limited set of "virtual" color patches. If you go this route, I recommend using a carefully selected set of colors such as the FOGRA media wedge (you must purchase this) instead of the ColorChecker colors. I do, however, see some potential in using artificial color bars, along with gray bars, as a guide to correct reproduction, but it will require an educated user on the printing end who understands how to correctly measure and interpret the results. This is not a visual assessment--the color patches must be measured and compared to the expected Lab values. When the average delta-E is within a defined tolerance, the print is considered to pass. > - If it is your decision to incorporate grayscales and color bars, then > it is really not good practice to only do it once for a "batch". It is > too difficult to mate up the correct grayscales with subsequent images-- > especially years hence. Put them in every image. I agree with this statement completely. It's not hard to include the target on every image, so why not? It's my recommendation however, to include only spectrally neutral gray patches and not more than one color patch (which is not to be used for judging color accuracy) and to only use them as I described above. ========================== The following definitions are taken from the document "ICC_white_paper5glossary.pdf" which is found at <http://color.org/whitepapers.xalter> output-referred image data image data which represents the colour-space coordinates of the elements of an image that has undergone colour rendering appropriate for a specified real or virtual output device and viewing conditions. [ISO 12231] NOTE 1 The output referred image data is referred to the specified output device and viewing conditions. A single scene can be colour rendered to a variety of output-referred representations depending on the anticipated output viewing conditions, media limitations, and/or artistic intents. NOTE 2 Output-referred image data may become the starting point for a subsequent reproduction process. For example, sRGB output-referred image data is frequently considered to be the starting point for the colour re-rendering performed by a printer designed to receive sRGB image data. scene-referred image data image data which represents estimates of the colour-space coordinates of the elements of a scene. [ISO 12231] NOTE 1 Scene-referred image data can be determined from raw DSC image data before colour rendering is performed. Generally, DSCs do not write scene-referred image data in image files, but some may do so in a special mode intended for this purpose. Typically, DSCs write standard output-referred image data where colour rendering has already been performed. NOTE 2 Scene-referred image data typically represents relative scene colorimetry estimates. Absolute scene colorimetry estimates may be calculated using a scaling factor. The scaling factor can be derived from additional information such as the image OECF, FNumber or ApertureValue, and ExposureTime or ShutterSpeedValue tags. NOTE 3 Scene-referred image data may contain inaccuracies due to the dynamic range limitations of the capture device, noise from various sources, quantization, optical blurring and flare that are not corrected for, and colour analysis errors due to capture device metamerism. In some cases, these sources of inaccuracy can be significant. ISO 17321-1 specifies a DSC/SMI (DSC Sensitivity Metamerism Index), which can be used to estimate the amount of inaccuracy resulting from capture device metamerism. NOTE 4 The transformation from raw DSC image data to scene-referred image data depends on the relative adopted whites selected for the scene and the colour space used to encode the image data. If the chosen scene adopted white is inappropriate, additional errors will be introduced into the scene-referred image data. These errors may be correctable if the transformation used to produce the scene-referred image data is known, and the colour encoding used for the incorrect scene referred image data has adequate precision and dynamic range. NOTE 5 Standard methods for the calculation of scene-referred image data from raw DSC image data will be specified in ISO 17321-2. NOTE 6 The scene may correspond to an actual view of the natural world, or a computer-generated simulation of such a view. It may also correspond to a modified scene determined by applying modifications to an original scene to produce some different desired scene. Any such scene modifications should leave the image in a scene-referred image state, and should be done in the context of an expected colour rendering transform. ========================== I welcome any and all comments. I hope this will become a useful discussion. Ken Fleisher, Photographer National Gallery of Art On 10/20/08 5:44 PM, "Stanley Smith" <SSmith at getty.edu> wrote: A couple of comments regarding the use of grayscale and color bars: ? due to differences between dyes and/or pigments used to make the color bars and the materials used to produce the actual artwork, accurately rendering the grayscales in a particular image may NOT produce the most accurate rendition of the artwork itself. ? If there is some visual editing to tweak color into place (hopefully under calibrated viewing conditions with direct comparisons to the original artwork), then another operator who may be printing the image later will most likely UNDO those edits if they assume that the image is rendered correctly by printing a neutral grayscale. ? There is some interest in the production of "virtual" grayscales and color bars. These would be digitally dropped into the edited and color corrected image with the same colorspace as the original image. Then other users would achieve better color when they match the grayscale and color bars. ? If it is your decision to incorporate grayscales and color bars, then it is really not good practice to only do it once for a "batch". It is too difficult to mate up the correct grayscales with subsequent images-- especially years hence. Put them in every image. Stanley Smith Manager, Imaging Services J. Paul Getty Museum
