Stretching and Squeezing the Bits

Every tonal edit you make causes some data loss. The purpose of this bald statement isn't to scare you, but simply to make you aware that like most things in digital imaging, editing tone and color involves a series of trade-offs. Photoshop's tools let you make the proverbial silk purse from a sow's ear, but they also let you do reverse!

Losing data is a natural and normal component of image editing. The trick is to throw away what you don't need, and to keep what you do need, and teaching you that valuable trick is one of the goals of this book.

When you work with images in Photoshop, they're often made up of one or more 8-bit channels, in which each pixel is represented by a value from 0 (black) to 255 (white). Grayscale images have one such channel, while color images have three (RGB or Lab) or four (CMYK). If you're more adventurous, you may work with high-bit images, where each channel uses 16 bits per pixel to represent a value from 0 (black) to 32,768 (white).

The key point is that when you use Photoshop's tonal controls, you're stretching and squeezing various parts of the tonal range, and in doing so, you inevitably lose some information. You lose a great deal more information in 8-bit-per-channel files than you do in high-bit ones, simply because you have much less data to start with. Here's a worst-case scenario that you can try yourself.


Create a new grayscale file in Photoshop, 7 inches wide by 5 inches tall, at 72 dpi, in 8-bit mode.


Use the Gradient tool to create a horizontal gradient from black to white across the entire width of the image (it's the third gradient in the default Gradients palette in the Options bar).


Choose Levels (from the Adjust submenu under the Image menu), change the gray input slider (the middle Input setting) to 2.2, and click OK (see Figure 6-1). You'll notice that the midtones are much lighter, but you may already be able to see some banding in the shadows.

Figure 6-1. Adjusting the gamma


Choose Levels again, and change the gamma to 0.5. The midtones are back almost to where you started, but you should be able to see that, instead of a smooth gradation, you have some distinct bands in the image (see Figure 6-2).

Figure 6-2. Data loss due to tonal correction

What happened here? With the first midtone adjustment, you lightened the midtonesstretching the shadows and compressing the highlights. With the second midtone adjustment, you darkened the midtonesstretching the highlights and compressing the shadows.

But with all that stretching and squeezing, you lost some levels. Instead of a smooth blend, with pixels occupying every value from 0 to 255, some of those levels became unpopulatedin fact, if we're counting right, some 76 levels are no longer being used.

If you repeat the pair of midtone adjustments, you'll see that each time you make an adjustment, the banding becomes more obvious as you lose more and more tonal information. Repeating the midtone adjustments half a dozen times produces a file that contains only 55 gray levels instead of 256. And once you've lost that information, there's no way to bring it back.

Now repeat the experiment, but create a 16-bit-per-channel file instead of an 8-bit one (in the New dialog box, choose 16-bit from the menu to the right of the Color Mode popup menu). The difference in the result is dramatic, as shown in Figure 6-3.

Figure 6-3. Data loss due to tonal correction

This set of adjustments represents, as we noted earlier, a worst-case scenario. No one with any significant Photoshop experience would make edits like this, with one edit attempting to undo the effect of the previous one. (And now that you've read this, you won't either!) The two immediate lessons to be drawn are:

  • Successive edits that counteract each other should always be avoided.

  • Data loss is much less of an issue with 16-bit/channel files than with 8-bit/channel ones.

    But all edits lose data in one of two ways:

  • When you stretch a tonal range, pixels that formerly had adjacent values may now differ by several levels. When you stretch the data too far, you lose the illusion of a continuous gradation, and you start to see distinct jumps in tone or color.

  • When you squeeze or compress a tonal range, pixels that formerly had different values are now compressed to the same value. If you compress the range too much, you may lose desirable detail.

In short, whenever you edit tone and color, something has to give. See the sidebar, "Difference Is Detail: Tonal-Correction Issues," on the next page.

Data Loss in Perspective

Photography has always been about throwing away everything that can't be reproduced in the photograph. We start with all the tone and color of the real world, with scene contrast ratios that can easily be in the 100,000:1 range. We reduce that to the contrast ratio we can capture on film or silicon, whichif we're exceptionally luckymay approach 10,000:1. By the time we get to reproducing the image in print, we have perhaps a 500:1 contrast ratio to play with.

This loss of image information may seem scarier than it really is. Nonetheless, we should hammer home the following lessons and the ensuing pieces of advice.

  • All tonal manipulations incur some data loss.

  • Once the data is gone, you can't bring it back.

  • You lose much less data making corrections in high-bit mode.

  • Successive tonal manipulations lose data at an increasing rate.

Difference Is Detail: Tonal-Correction Issues

What do we mean when we talk about image information? Very simply, adjacent pixels with different values constitute image detail. If the difference is very slight, you won't be able to see it (especially in shadows); but it's there, waiting to be exploited. You can accentuate those differencesmaking those adjacent pixels more differentto bring out the detail.

The color of noise. Difference isn't always detail, though. Low-cost scanners, and digital cameras shot at high speeds, introduce spurious differences between pixels. Those differences aren't detail, they're just noise (like the static on the radio that drowns out the traffic report), and they're one of our least favorite things. Photoshop can't tell the difference between genuine image information and device-induced noise. You need to decide what is desirable detail and what's noise, accentuating the detail while minimizing the noise.

Posterization. Noise isn't the only problem to deal with when you're doing tonal correction, though. There's also posterizationthe stair-stepping of gray levels in distinct, visible jumps rather than smooth gradations (see Figure 6-4). Unless you're working with a high-bit image, Photoshop gives you only 256 possible values for a gray pixel.

Figure 6-4. The effects of posterization

Posterization manifests itself in two ways. When you start making dark pixels more different, you eventually make them so different that the image looks splotchycovered with patches of distinctly different pixels rather than smooth transitions. But posterization can also wipe out detail and turn smooth gradations into flat blobs.

Lost detail. When you accentuate detail in one part of the tonal range, making slightly different pixels more different (expanding the range), you lose detail in other areas, making slightly different pixels more similar (compressing the range).

For instance, when you stretch the shadow values apart to bring out shadow detail, you inevitably squeeze the highlight values together (see Figure 6-5). If you make two different pixels the same, that detail is gone forever. That's what we mean when we say that information is "lost."

Figure 6-5. Loss of highlights

Get good data to begin with

If you've got a high-bit capture device, (which you almost certainly do), it's better to get the image as close to "right" as you possibly can while the image is in high-bit form. The whole point of high-bit capture is to let you manipulate the high-bit data, so that you get the right 8-bit data when you downsample to 8 bits per channel.

Use adjustment layers

You can avoid some of the penalties incurred by successive corrections by using an adjustment layer instead of applying the changes directly to the image. Adjustment layers use more RAM and create bigger files, but the increased flexibility makes the trade-off worthwhileespecially when you find yourself needing to back off from previous edits. But since the various tools offered in adjustment layers operate identically to the way they work on flat files, we'll discuss how the features (Curves, Levels, Hue/Saturation, and so on) work on flat files first. For a detailed discussion of adjustment layers, see Chapter 7, The Digital Darkroom.

Use high-bit data

If your scanner or camera allows it, you can bring the 10-, 12-, 14-, or 16-bit-per-channel data directly into Photoshopyou are, in effect, telling your capture device "just give me all the data you can capture"and then edit it in Photoshop. In the 16-bit-per-channel space, you have much more editing headroom before you run into posterization. Photoshop CS2 lets you do just about anything to a high-bit file that you can do to an 8-bit one, so the main reason for downsampling is to keep file sizes managable. If you bring digital captures into Photoshop using Camera Raw, note that Camera Raw always operates on the high-bit data, and offers the option to bring the edited high-bit data into Photoshop.

Minimize tonal correction

Small tonal moves are much less destructive than big ones. The more you want to change an image, the more compromises you'll have to make to avoid obvious posterization, artifacts due to noise, and loss of highlight and shadow detail.

Be decisive, or use adjustment layers

If you push an edit too far, don't try to correct it with a second editgo back and fix the problem instead. You can use History to step back before the edit, or if you use adjustment layers, you have the freedom to keep changing your mind right up to the point where you flatten the image.

Cover yourself

Since the data you lose is irretrievable, leave yourself a way out by working on a copy of the file, by saving your tonal adjustments in progress separately (without applying them to the image by applying your edits using adjustment layers), or by using any combination of the above. Or you can use the History feature to leave yourself an escape routejust remember that History only remembers the number of states you specify in the Preferences dialog box, and it can consume mind-boggling amounts of scratch disk space.

Data loss can be good

Sometimes you want to throw away information. For example, none of these restrictions applies when you're working on masks or alpha channelsin fact, you usually want to throw away data on those, since you're often trying to exaggerate a feature or isolate it from its background. (See "Step-by-Step Silhouettes" in Chapter 8, Making Selections.)

With all these caveats in mind, let's take a look at Photoshop's tonal-manipulation tools.

Real World Adobe Photoshop CS2(c) Industrial-Strength Production Techniques
Real World Adobe Photoshop CS2: Industrial-strength Production Techniques
ISBN: B000N7B9T6
Year: 2006
Pages: 220
Authors: Bruce Fraser © 2008-2017.
If you may any questions please contact us: