One of the most important issues in working with images—and, unfortunately, one which few people seem to understand—is how the resolution can change relative to (or independently of) the size of your image.
There are two ways that you can change resolution: scaling and resampling. Scaling doesn't change the number of pixels, just the resolution. Resampling changes the pixel dimensions. If you take a 2-by-2-inch, 300-ppi image and change the size to 1 inch square in QuarkXPress or InDesign, you're scaling: The pixels get smaller and the resolution gets higher (it increases to 600 ppi).
In Photoshop, you have a choice whether to scale or resample. If you scale that image down without changing the resolution, Photoshop has to throw away a bunch of pixels; that's called downsampling. If you double the size to four by four inches by upsampling, the program has to add more pixels by interpolating between the other pixels in the image.
Upsampling vs. Downsampling
We used to avoid upsampling when our images mostly came from scanned film, but in the digital age, it no longer makes sense to have a hard-and-fast rule. The lack of film grain in digital captures makes them much more amenable to upsampling than film scans ever were. We often upsample digital camera captures by 200 percent, sometimes even more. Upsampling still doesn't add details that weren't there in the capture, but sometimes it does an uncannily good impersonation! Nevertheless, a 4.1-megapixel capture rarely makes a good magazine cover!
Downsampling is much less problematic, because it's just throwing away data in a more or less intelligent manner. In fact, it's a common and necessary practice: We often scan at a higher resolution than is strictly necessary, to allow for cropping and for unanticipated changes in output size or method. We downsample to the required resolution before printing to save time and storage space.
Photoshop offers five resampling methods: Nearest Neighbor, Bilinear, Bicubic, Bicubic Smoother, and Bicubic Sharper, the last two of which were introduced in Photoshop CS. You choose which you want in the General panel of the Preferences dialog box or in the Image Size dialog box (see Figures 3-9 and 3-10). Each has its strengths and weaknesses, and we use them all in different situations.
Figure 3-9 Photoshop's resampling (interpolation) methods
Figure 3-10 Image Size dialog box
- Nearest Neighbor is the most basic, and it's very fast: To create a new pixel, Photoshop simply looks at the pixel next to it and copies its value. Unfortunately, the results are usually lousy unless the image is made of colored lines or shapes (like an image from Illustrator or FreeHand), but it's often useful for preserving the readability of screen shots.
- Bilinear is slightly more complex and produces somewhat better quality: the program sets the color or gray value of each pixel according to the pixels surrounding it. Some pictures can be upsampled pretty well with bilinear interpolation. But we usually use one of the bicubic options instead.
- Bicubic interpolation creates better effects than Nearest Neighbor or Bilinear, but takes longer. Like Bilinear, it looks at surrounding pixels, but the equation it uses is much more complex and calculation intensive, producing smoother tonal gradations.
- Bicubic Smoother is a new interpolation method specifically designed for upsampling. As its name suggests, it gives a smoother result that handles subsequent sharpening better than Bicubic sampling.
- Bicubic Sharper is another new interpolation method, only this time designed for downsampling. It does a better job of preserving detail than does Bicubic.
The differences are often subtle, and since we always recommend sharpening after resampling, the resampling itself is only half of the story. The new simple rule: Use Bicubic Smoother for upsampling (but don't expect miracles, particularly with film scans) and Bicubic Sharper for downsampling.