Site map/guide to tutorials Contact | News Making fine prints in your digital darkroom Understanding image sharpness and MTF Image galleries / How to purchase prints Photographic technique Image editing with Picture Window Pro A simplified zone system Digital vs. film

View image galleries

Search WWW Search www.normankoren.com

Prints made from images captured on digital cameras can have extraordinary tonal quality, comparable to the finest full-toned traditional prints. But to reach their full potential they must be processed properly, which involves changing default camera settings and moving away from standard file formats. This page covers the key steps for achieving ultimate tonal quality, which include,

• Storing images in RAW format-- the image sensor's native format,
• Converting and editing images using file formats with a bit depth of 16 (48-bit color or 16-bit B&W), and
• Using curves to adjust tones.
  

Introduction: RAW conversion

---

Output luminance = (pixel level)^gamma

pixel level = (RAW pixel level)^1/gamma ~= exposure^1/gamma

Figure 1. Characteristic curves for the two conversion processes in digital printmaking.
	File pixel level (normalized to 1) as a function of exposure (raw pixel levels).	Output luminance (print or monitor) as a function of normalized file pixel levels.
Both shown with linear scales. The same data is displayed with logarithmic scales in Fig. 3.

Gamma correction is one of several functions of RAW conversion-- the process of converting the output of an image sensor to a standard file format, such as JPEG or TIFF. Depending on your camera's Storage or Quality setting, RAW conversion takes place inside the camera, immediately after the exposure, or afterwards on a computer.

It should be evident from these graphs that RAW conversion compresses pixel levels representing high luminance and expands pixel levels representing low luminance. This means that the converted file, with gamma = 2.2, has relatively fewer pixel levels in the highlights and more in the shadows. This turns out to be an advantage when human vision is considered.

RAW conversion performs several additional functions.

• Bayer array interpolation (de-mosaicing) is the most important function. Virtually all image sensors (except for the Foveon sensors used in the Sigma SD9/SD10) use the Bayer array pattern, where alternate rows of pixels represent RGRGRG... and GBGBGB..., but each pixel in standard image files (JPEG, TIFF, etc.) represents all three colors. RAW data is converted using a sophisticated process called interpolation, where data from the green sensors (twice as plentiful as red and blue) is used to enhance the resolution of the red and blue channels. The best Bayer interpolation routines-- the routines that result in the highest resolution and the fewest artifacts-- use iterative calculations, which are too slow to run on digital cameras, but present no problem for modern computers. See the section on Bayer sensors in Digital vs. film for links to technical papers.
• Sharpening.  Unsharpened digital camera images tend to look soft; they benefit strongly from sharpening. But images, especially in some inexpensive compact digital cameras, are often oversharpened, resulting in "halos" near edges. (Digital SLRs tend to be more conservative.) Oversharpened images look good in small enlargements straight out of the camera, but oversharpening creates artifacts that can be hard to get rid of. It's always best to sharpen late in the image editing process.
• White balance  Digital cameras are used with light sources that have a wide range of color temperature. This must be addressed when the RAW conversion is performed. Otherwise the image may look too blue or yellow. Most cameras have several custom White Balance settings in addition to an automatic setting that estimates the White Balance based on the spectral content of the image. Automatic White Balance algorithms work well most of the time, but they can fail for unusual subjects, for example, where a strong color dominates the scene. When you save an image in RAW format and convert it later, you can make use of your calibrated monitor to get White Balance just right.

The remainder of this page will show how performing RAW conversion on a computer, rather than in the camera, enables you to achieve optimum tonal quality.

Human vision and tonal levels

---

Delta L/L = 0.01

In a photographic print, which has about a 100:1 luminance ratio, the eye can distinguish between 100 and 200 discrete luminance levels-- fewer than the 256 available in 8-bit B&W or 24-bit color. Assuming 200 levels, this is an average of 30 discrete visible steps in each of the 6.6 zones that comprise the 100:1 luminance ratio-- fewer than the 70 indicated by the Weber-Fechner law. The reason for the discrepancy is that the eye distinguishes fewer luminance levels in the dark areas of a print due to visual interference-- mostly flare light-- from the light areas. The number of luminance levels the eye can distinguish decreases gradually from light to dark zones.

In a RAW, i.e., linear file, the pixel level is proportional to light intensity. An exposure change of one zone or f-stop involves halving or doubling the pixel level. Therefore, half the pixel levels are in the top (brightest) zone, a quarter are in the second zone, a fourth are in the third zone, etc. A linear file with a bit depth of 8 (256 discrete levels) would contain 128 levels in the top zone, 64 levels in the second, and so on. The top zone would have far more levels than the eye can distinguish, while the lower zones would have relatively few levels-- only 8 and 4 in zones 5 and 6, respectively. "Banding" would be visible in shadows. If much editing were applied to the image-- if shadow areas had to be lightened (dodged) or increased in contrast, the banding could become severe. In a color space with gamma = 2.2, the distribution of levels is much more uniform: there are 69 levels in zone 1 (close to Weber-Fechner's magic number), 50 in zone 2, and so on. Zone 6 would still have 14 levels. Banding is invisible unless considerable editing is applied. Gamma = 2.2 may well be the ideal value for files with a bit depth of 8 (8-bit B&W and 24-bit color).

A few things are evident from this chart.

• Levels are lost when converting from a 12-bit RAW format (4096 levels total) to an 8-bit file (8-bit B&W or 24-bit color). But image quality in an 8/24-bit file will be adequate, though just barely, if the exposure is correct and little editing is required. This is achievable in studio environments, but less often when using "natural" (i.e., uncontrolled) light.
• 16-bit files (16-bit B&W or 48-bit color) have plenty of levels (65,536 total). You can edit to your heart's content without fear of banding or other artifacts arising from limitations of 8-bit files.
• When RAW conversion is performed within the camera, i.e., when you save JPEG (or other standard) files instead of RAW files, you have little control over the process. The Canon EOS-10D has a contrast setting that gives a small amount of control, but it's awkward to access and it makes little difference.
• If you save files as RAW (the sensor's native format) and convert them later on a computer, you have enormous control over tones when you convert.

Dynamic range

---

We can gain some insight by looking at a simple imaging system-- a camera/printer combination that reproduces images with no editing or manipulation. The two gamma equations in the section on RAW conversion are applied exactly. The contrast of the print would be identical to the scene contrast-- a nice idea in theory, but one that rarely produces "fine" prints because a print can only reproduce a tonal range of about 100:1-- 6.6 f-stops. In this case the system dynamic range would also be 6.6 f-stops-- well under the camera's potential dynamic range.
 

The transfer curve (on a logarithmic scale) would look like the straight dashed blue line in Fig. 2. (Logarithmic scales correspond to human vision better than linear scales; they are discussed in detail below.) Camera manufacturers are faced with a classic tradeoff: contrast versus dynamic range. Most images look best with enhanced contrast, but if contrast is increased, dynamic range suffers. Dynamic range can be increased by decreasing contrast, but images tend to look flat. That won't sell cameras! One way around this tradeoff is to apply a curve, shown in solid black in Fig. 2. Contrast is reduced in the shadows and highlights, but increased in the important midtone area, which is responsible for perceived contrast.

Figure 2.

I don't know whether curves are applied in the RAW conversion algorithms used in cameras. There's no reason why they shouldn't, but if they are, camera manufacturers aren't saying. It would be part of the proprietary signal processing package that gives them a competitive advantage. Camera manufacturers could do even more-- they could make conversion adaptive. Low contrast scenes could have contrast expanded and contrasty scenes could have contrast compressed. Again, if any camera manufacturers are doing this, they aren't telling. Perhaps it will be next year's big "advance."

In the meantime you can apply curves manually with a flexible RAW conversion program like Capture One DSLR, or in an image editor. Both approaches are described below. One of the best uses of curves is to bring up dark levels that are lost when RAW conversion is performed inside the camera, as described above. This enables you to access your camera's hidden dynamic range.

The number of A-to-D converter output levels isn't the only factor that limits dynamic range. Noise is another, and it's extremely important. Unfortunately you won't find much about it in camera spec sheets or glossy brochures. It's difficult to quantify-- the amount of noise depends on the exposure and ISO speed, and it has a spectral distribution-- it's not a simple number. But it has a few important properties.

• Noise is strongly correlated with pixel size: the larger the pixel, the lower the noise. Digital SLRs, which have 6.8 micron or larger pixels, tend to have lower noise than compact digital cameras, which have 3.4 micron or smaller pixels. Hence they have a larger hidden dynamic range. This is one of the major advantages of DSLRs.
• Signal-to-Noise Ratio (SNR)-- the amount of noise relative to the signal, is worst in the darkest zones.
• The higher the ISO speed, the higher the noise. Digital cameras increase ISO speed by amplifying the signal-- moving a given exposure to a lighter zone. For example, an exposure that would fall in zone 9 (in Table 1) at ISO 100 would be fall zone 8 at ISO 200 and zone 7 at ISO 400, etc. Unfortunately, noise is amplified along with the signal, and if it's amplified enough-- if ISO speeds are high enough, noise can become objectionable, particularly in dark zones. This can be particularly significant if you use curves to expand dynamic range.

Exposure

---

• If you choose to save the images in a standard file format,  you should expose for tones that look good-- as close as possible to the tones you want in the final print. This works best if the scene dynamic range is close to the dynamic range available in a print-- about 5 to 7 f-stops.
• If you choose to save images in RAW format, you should expose to capture maximum information: to maintain as much highlight and shadow detail as possible, even if the middle tones aren't what you want in the final print. You should strive to capture all highlights except for bright light sources and specular reflections. You should expose enough to capture detail in large shadow areas. I mostly agree with the Luminous-Landscape.com article, Expose (to the) Right, which recommends setting the exposure to the maximum value that doesn't burn out highlights. (This applies only to images saved in RAW format.) However I wouldn't go too far. A little margin doesn't hurt; there are plenty of levels in 12-bit A-to-D converters. In extreme situations, you may want to make two exposures and combine them.

Shadow and highlight detail are extremely important in fine full-toned prints. To my eyes, a print with dead shadows or burnt out highlights looks amateurish. One of the things that distinguished the glorious prints of Paul Strand, Edward Weston, and Ansel Adams is the tonal detail in shadows and highlights, as well as middle tones. If you haven't seen original prints by these great artists, it's time to make a pilgrimage.

Although I repeatedly stress the importance of saving files in RAW format, there are instances where high quality can be achieved using JPEGs. Claude Jodoin of Michigan uses JPEGs for weddings and other high volume events. RAW conversion is too slow for his workflow. He uses high quality lighting, an ExpoDisc to set the White Point, and he knows how to expose properly, so images look good straight out of the camera. The bottom line is satisfied clients. This doesn't work well for landscape photography, where lighting is uncontrolled and scene contrast varies all over the place.

Comparing digital and film: gamma equals contrast.

---

For this reason it makes sense to use logarithmic curves to display the relationships between exposure, film density, pixel level, and luminance. This is routinely done for film and papers, where Film Density (-Log₁₀(transmitted light/incident light)) plays the same role as pixel level in digital images. Note that one unit on a Log₁₀ scale (such as Density) equals 3.32 exposure zones (f-stops); one exposure zone equals 0.301 Density units.

When the conversion curves in Fig. 1 are replotted on logarithmic scales, the characteristic curves become straight lines: If y = x^gamma, then log(y) = gamma * log(x). Gamma is the slope of the line; it is the change in the dependent variable (y-axis) for a given change in the independent variable (x-axis). Using the RAW conversion plot in Fig. 3 as an example, gamma is the change in Log (pixel level) for a given change in Log (exposure). In other words, gamma is contrast.

Figure 3. Characteristic curves for the two conversion processes in digital printmaking. Both are displayed with logarithmic (base 10) scales. In contrast to the linear display in Fig. 1, the transfer curves are straight lines.
	File pixel level (normalized to 1) as a function of exposure (raw pixel levels).	Output luminance (print or monitor) as a function of normalized file pixel levels.

The equivalence of gamma with contrast becomes strikingly apparent when we compare Fig. 3 to film and paper characteristic curves in Fig. 4. The RAW conversion plot (Fig. 3, left) is similar to the film curve (Fig. 4, left) and the print/monitor display curve (Fig 3, right) is similar to the photographic paper curves (Fig. 4, right). Log (pixel level) is analogous to Density = -Log (light transmission) in the film plot and Log exposure in the paper plot. Reflection density is the same as -Log (luminance). The average slope of the film and paper curves is called "gamma;" it is indeed the same gamma used to characterize digital color spaces and monitors.

Figure 4. Characteristic curves for the two conversion processes in conventional photographic (darkroom) printmaking.	Kodak Tri-X film Film density = -log10(fraction of light transmitted through the film) as a function of exposure.	Kodak Polymax photographic paper Print reflection density = -log10(fraction of light reflected by the print) as a function of log exposure (-film density).
The principal difference between the digital and conventional curves is in the "knee" regions near the response limits. The knees tend to be more abrupt for digital.

Measuring digital camera tonal response and dynamic range With the new Imatest program, digital camera tonal response and dynamic range to be measured easily and accurately using a transmission step wedge-- a piece of film with zones of incresing density. The details of the measurement are in the Imatest Q-13 tour. Here are the results for the Canon EOS-10D at ISO 400, converted from RAW format with Capture One LE. The upper left plot is the density response curve. The lower left plot is noise measured in f-stops-- a relative measure that corrresponds to the eye's response. Each step represents 1/2 f-stop (a density step of 0.15). The total dynamic range of the EOS-10D is 8.5 f-stops. Dynamic range improves slightly (but less than I expected) for 48-bit TIFF conversion and ISO 100. Noise is significantly lower for 48-bit TIFF conversion. The shape of the response curve is a strong function of the conversion software settings. The curve on the right for Canon Zoom Browser with Contrast set to Low is very different from Capture One LE (though both are half an "S" curve), but the dynamic range hardly changes. You can download an evaluation version of Imatest that allows up to 20 runs of individual modules, and you can purchase Imatest for a very affordable price.

Achieving ultimate tonal quality

---

What follows may seem a bit strange; please bear with me. Fig. 5 is the curve for Kodak Azo paper-- a silver chloride contact printing paper with an exalted reputation for fine tonal quality. Michael A. Smith and Paula Chamlee-- print quality fanatics in the Edward Weston/Ansel Adams tradition (they use 8x10 and 8x20 inch cameras)-- rave about it. "In our opinion, we have not seen contemporary photographic prints on any other photographic paper, including all platinum and other "alternative" papers, that consistently rival the depth, luminosity, and tonal range of photographs printed on Azo paper." Unless you plan to swap your DSLR for an 8x10, take a careful look at the curve. It's different from the Polymax curve in Fig. 4 and strikingly different from the digital camera curves in Fig. 3.

Figure 5. Azo characteristic curve.

The key difference is in the knee of the curve, indicated by the red arrow-- the highlight region where density starts increasing. The knee is much broader for Azo than for Polymax-- density increases more gradually. The knees in both film curves are in stark contrast to the digital curves in Fig. 3, which cut off abruptly at their limits. (The polarities of several parameters are reversed; the corresponding knee in Fig. 3, such as it is, is indicated by an arrow in the upper right.) Thanks to the knee, Azo holds highlight detail very nicely. It doesn't easily burn out. The knee gives Azo its superb luminosity. (Luminosity refers, of course, to highlights.)

The knee is not there by accident. It's the climax of an evolution that has spanned several generations of silver-based paper (and photographers as well). Digital printing has evolved enormously in the last decade, but it's still the new kid on the block. There are still some rough edges, or perhaps... sharp corners.

Digital photographers need not despair. The effects of the "knee" can be duplicated by Curve controls using one of two approaches.

1. Apply a curve control during RAW conversion. This requires a high quality RAW conversion program like Capture One DSLR. (I use C1 LE with my EOS-10D.) The crummy File Viewer Utility supplied free with Canon digital cameras won't do (though it would be a bargain at half the price).


2. Convert the RAW image to a file with a bit depth of 16: 48-bit color or 16-bit B&W and apply the curves with an image editor. This allows you to get away using Canon's File Viewer Utility, but you need an image editor that can support a bit depth of 16, like Picture Window Pro (my favorite) or Adobe Photoshop CS. Earlier versions of Photoshop had limited support for 16-bit editing and Elements doesn't support it at all. Results will be inferior with a bit depth of 8.

And there's one more difference with digital. You can fine tune the curve to match the needs of the image, precisely. You aren't limited by the curve of the paper. (In my old B&W darkroom days I would to adjust the curve by mixing standard and low contrast paper developers-- Dektol and Selectol Soft. Digital curves are easier.)

1. Curve applied during RAW conversion

---

Figure 7. C1 LE Levels tab with histogram The histogram, visible when the Levels tab is selected, is shown in Fig. 7. The levels tab allows you to set the minimum and maximum pixel levels. I moved the minimum from its default of 0 to 20 but didn't change the maximum-- I used the Curve for that adjustment. To adjust the curve you need to select the Curve tab, as illustrated in Fig. 6. The curve I chose is quite similar to the Azo curve-- recall that polarities are reversed, so the lower left of the Azo curve corresponds to the upper right of the C1 LE curve. There's nothing rigid about the Azo curve; I choose whatever curve (or straight line) works best with each individual image.

2. Curve applied in the image editor

---

Despite the weaknesses in its interface, the Canon File Viewer Utility does a decent job in converting the RAW image. And if you convert to 48-bit color you can edit with no loss of tonal levels. Fig. 9 illustrates how to edit the image with the Color curves transformation in Picture Window Pro.

The original image-- the anemic output of the Canon File Viewer Utility-- is on the lower left. The preview image is on the top. The Color Curves control box is on the right. I especially like the way it combines the curves adjustment with a histogram-- it gives exceptional control. The curves behave differently from Capture One LE, but the overall result is similar. I used the HSL color representation, which tends to increase saturation as lightness (L) is decreased. I also boosted the Saturation (S) (not visible here). There are gaps in the lower (result) histogram because the preview uses 24-bit color. The gaps aren't present in the histogram of the 48-bit transformed image (Fig. 10, right). But they are present when the same Color Curves transformation is performed on an identical 24-bit color image (Fig. 10, left) More information on using the Color Curves transformation can be found in Digital Light and Color's tutorial, Using Curves and Histograms (a 1.34 MB PDF download).
.

The loss of tonal detail is obvious in the histogram on the left, though it's probably not severe enough to cause serious banding. But if you apply multiple edits-- if you do much selective tonal control (dodging and burning), the degradations will accumulate. The tonal gradation of the final image won't be as smooth as it could be. And of course, this image, like most landscapes, can benefit from additional work. It’s by no means complete at this stage.	24-bit color (bit depth = 8)	48-bit color (bit depth = 16)
	Figure 10. Histograms after 24 and 48-bit Color Curves transformations.

Black & white

---

All digital cameras and most photographic printers are designed for color: it takes effort and imagination to turn images into Black & White. What a contrast to my youth, when color was almost impossibly difficult in the home darkroom! B&W is a more abstract medium; it's not as closely tied to reality. B&W prints typically require and tolerate more manipulation-- dodging and burning. Read Ansel Adams' "The Print" if you're skeptical. The results can be striking-- very different from color. And the tones can be gorgeous. To convert the image into B&W I used Picture Window Pro's Monochrome transformation, which allows you to select and preview a filter of any color. I describe its use in my page on Black & White. I used a red-orange filter because pure red was a little harsh. I also did some tonal manipulation and contrast masking. In some ways I prefer the B&W image.

Figure 11. B&W with red-orange filter

Newer printers such as the Epson 2200, which has both black and gray ink cartridges, do an excellent job with B&W, though glossy surfaces still can't quite match the finest silver prints. (However, matte surfaces whip their silver counterparts.) They do even better if you use a piece of software called a Raster Image Processor (RIP) to control the ink mixture. With an RIP you can minimize metamerism-- changes in the appearance of the print under different light sources, which are much more noticeable in B&W prints than in color.

Summary

---

• If possible, store digital camera images in RAW format and convert them on a computer.
• If possible, use a RAW conversion program that allows a high degree control over image tones and color. A Curves control is particularly desirable.
• Conversion to 48-bit color (with a bit depth of 16) is strongly recommended. It's a must if the RAW conversion program doesn't have flexible controls.
• Take advantage of the ability of Curves controls to simulate the "knees" of traditional photographic papers-- especially in the highlight regions.
• This page focuses on tonal levels in image capture and editing, but several equally important areas are discussed on other pages: Your monitor should be well-calibrated and you should have a high quality photographic printer. If possible, use a color-managed workflow with high quality profiles.

IImages and text copyright (C) 2000-2013 by Norman Koren. Norman Koren lives in Boulder, Colorado, where he worked in developing magnetic recording technology for high capacity data storage systems until 2001. Since 2003 most of his time has been devoted to the development of Imatest. He has been involved with photography since 1964.