In the July/August 2009 issue of PT, we discussed the Ideal Tone- Reproduction Curve, a product of research conducted more than a half century ago that identif ies, for a scene element of any luminance value, the shade of gray (ref lection density) at which it is “best” reproduced in a black-and-white print. We also raised the question of how readily this ideal tone-curve is achieved in a purely digital workf low—the subject of the present article.
With the magic of Photoshop, any kind of tone reproduction can, of course, be realized with exacting precision. But what is inherent to digital in the absence of any manipulations? To answer this, we created a tone- reproduction curve for comparison with the ideal, using an inexpensive digital camera, inkjet printer, and photo-quality inkjet paper—no adjustments to the JPEG prior to printing, no printing profiles, just a straight print of an as-captured image. Creation of this digital tone- reproduction curve first requires construction of a camera’s characteristic curve as explained in the March/April 2009 edition’s Characteristic Curves for Digital Cameras. That article described how to characterize a digital camera in terms of the relation between camera exposure and brightness value (BV, on a scale of 0 to 255) of the resultant
image on screen. The curve derives from photographing a uniform card, such as an 18% gray card, over a wide range of exposures, spaced in one-stop increments, suff icient to create a series of frames ranging from BV= 0 to BV= 255 and plotting their BVs in one-stop increments on an arbitrary Log E axis. This graph serves as a reconnect between the world of digital and the characteristic curves we knew in the silver era. In the digital world it is also known as an opto-electronic conversion function (OECF) and is discussed extensively in ISO 14524:1999(E).
Figure 1 illustrates the characteristic curve unique to the camera we used for the experiments described in this article. We then used the data from this curve to construct a new image in Photoshop, illustrated in Figure 2. This image contains a series of patches having the same BVs as the data points in our camera curve (Figure 1). Select a printer, paper, and printer settings, print this image, and read the reflection density of each patch with a densitometer. Plot those densities against the same arbitrary Log E axis used previously and the result is a tone-reproduction curve for that specif ic combination of camera, printer, paper, and both camera and printer settings. Note that in this and the following two graphs we have subtracted Dmin from all density readings prior to plotting them. We did this to avoid the visual confusion of graphs being vertically offset due to differences in Dmin.
So does it look anything like the Ideal Tone- Reproduction Curve that resulted from that decades-old research in the wet darkroom? Plot the ideal curve on the same graph as the new digital-tone curve. Adjust its position horizontally such that the actual and ideal curves overlap as much as possible through the middle grays. This is what we have done in Figure 3. We were frankly rather astonished at the agreement, especially given that we had used a combination of a rather inexpensive camera, printer, and paper to generate our digital-tone curve. The treatment of midtones, ideal versus digital, is almost indistinguishable. Digital highlights are lighter than ideal as is to be expected from an inexpensive camera of limited dynamic range. There are differences in the shadow end of the scale, which are largely attributable to the difference in maximum densities for the two curves. As noted in the previous issue’s discussion of the ideal curve, such differences are of little significance.
In truth, this was not the only camera-printer-paper combination we looked at. But it did afford tone reproduction closer to the ideal curve than any other combination. Several others were moderately close to the result in Figure 3, but a disturbing number, even some incorporating the printer and paper manufacturer’s recommended printing profiles, were really rather bizarre.
Could the good fit illustrated in Figure 3 be made even better? And would it serve as a path to better prints? Indeed, it is possible to make a single adjustment in Photoshop’s Curves dialog to secure a tone reproduction that is a nearly perfect match to the ideal. To do this we need one more piece of information, that being the relation between screen image BV and print density.
The former is the same set of numbers (BVs) that appear in Figure 2. The latter is the set of print densities that resulted from printing Figure 2. Plot one against the other as illustrated in Figure 4 to create a tool that allows any density mismatches, actual vs. ideal, to be converted to BV mismatches, the values needed to create a correction curve in Photoshop. Carefully compare the two curves, actual and ideal, of Figure 3, jot down any density mismatches, and use Figure 4 to convert them to BV mismatches. Open an image as well as a Curves dialog box in Photoshop and insert some points on the curve. Type an original BV (the one that yielded the “wrong” print density) in a point’s Input box, and in the Output box enter the new value of BV that will force the correct print density. Save this curve. Applied to any image captured with the same camera, printed with the present printer and paper, it will always produce a print with “ideal tone reproduction.” We applied this correction curve to our image of Figure 2 and reprinted it. The results are displayed in Figure 5. The solid curve is once again that for Ideal Tone Reproduction; the data points are the densities read off of the corrected print.
Curves and their consequences
At this point we were eager to set the curves aside and explore their pictorial consequences. We selected an assortment of outdoor images, both pure scenics and people pictures, made with the camera referenced here in auto- program and normal contrast modes. After conversion to grayscale, we printed them both with and without benef it of the adjustment curve and compared the pairs side by side. Conclusion? We felt the prints made with the Curves adjustment to force the ideal tone reproduction def ined by those researchers of yesteryear were consistently superior to unadjusted straight prints.
Of course, who is to say that original ideal tone curve from so many years ago is really ideal? The ability of those researchers to illustrate various curves in their quest for the ideal was constrained to a degree by the photographic materials available to them. Perhaps with a film or paper of slightly different curve shape they would have come to a slightly different conclusion? Then there is the whole issue of cultural preferences. A few years back we attended a photo show in Moscow and were struck by how dark and ponderous all the exhibition prints were to our eyes. The classic ideal curve may well be very close to ideal for many, but it is likely not a universal standard.
These thoughts prompted us to explore some variations on the presumably ideal curve with attention focused on the midtones those earlier researchers noted as especially significant. Figure 6 again displays the ideal curve (dotted), and it is flanked by a pair of hand-drawn curves that slightly raise and lower the middle portion of the ideal curve. Using the procedures described above, we created curves in Photoshop that would force these modif ied tonal relations on prints and again reprinted the image of Figure 2. The solid curves in Figure 6 are the ones we sketched by hand; the data points represent the actual densities read off our new pair of prints. Satisf ied we could force these alternative tonal responses with our new adjustment curves, we applied both of them to the same assortment of images as before. Prints representing Figure 6’s lower curve were, in our estimation, uniformly inferior to those ref lecting the ideal curve. Prints with the upper curve, however, suggested that something between it and the ideal would furnish better prints than the ideal for about half of the images. We recognize that our limited evaluation involved only a few prints, whereas the original research of long ago involved 170 scenes, so we do not suggest that a curve slightly higher in the midtones truly represents an all-purpose improvement.
So what do we take away from this exercise?
• The Ideal Tone-Reproduction Curve as defined many years ago remains perfectly valid today, in our eyes.
• A digital workflow, absent all image manipulation,can, even with inexpensive equipment, deliver prints surprisingly consistent with the ideal curve.
• At the same time, we looked at many printer-paper combinations which were disappointingly far removed from the ideal.
• To the extent a particular camera- printer-paper combination misses the ideal for tone reproduction, a correction curve can be created in Photoshop to force the ideal tonality.
New correction method
Lastly, the correction curves we built to compel ideal tone reproduction with various cameras and papers open the door to an alternative way of working. Our usual workflow with a new image ultimately destined for printing is to open it in Photoshop, immediately apply adjustments as we visually deem appropriate, then print it—hoping WYSIWYG (what-you- see-is-what-you-get) prevails and we like the print. Our new alternative is to open the new image, apply the appropriate standard correction curve to it, make a print, and use our assessment of this print’s appearance to decide what kind of further image manipulation in Photoshop is indicated. This approach feels more like the good old days when we would make a properly exposed straight print of a new negative and mark it up with a grease pencil to identify areas for burning, dodging, hot spotting, and so on.
Readers who might like to pursue these tactics with their own cameras, printers, and papers need a copy of the Ideal Tone-Reproduction Curve to serve as a reference. In the table in Figure 7 we present a listing of data points for construction of this curve. This example is specific to a curve with a maximum density of 1.67. Should you prefer a different value of Dmax, our July/August 2009 column noted that the ideal curve pertained to a variety of Dmax values, the f irst seven data points of the table being common to all of them.