Just as the process of turning photons into pixels introduces softness, so does the process of turning pixels into marks on a substrate (typically but not always ink on paper). There are three basic types of printed output.
We can't control the way pixels get turned into dots. What we can control is the resolutionthe number of pixels per inchwe send to the print device, and the way we sharpen those pixels. Moreover, the resolutions we usually send to the various different print technologies aren't just picked out of thin air. They're based on the physiology of the eye itself, which imposes limits on the finest detail we can see.
Resolution and the Eye
It should be obvious, but I'll state it anyway, that it takes two photoreceptors to detect a difference in luminance or color, because to detect a difference, you need two signals. It follows that the smallest difference we could possibly see is one that is projected onto two photoreceptors on the retina.
Human Visual Acuity
The generally accepted definition of normal (20/20) visual acuity is the ability to resolve a spatial pattern whose features are separated by one minute of arc, or 1/60 of a degree. This number comes directly from the retina. The lens in the eye projects one degree of the scene across 288 micrometers (or microns) of the retina.
In the area of the retina where the photoreceptors are most tightly packed, the fovea, a linear 288 micrometers contains about 120 photoreceptors. So, if more than 120 alternating black and white lines, or 60 cycles, are projected onto this area, someone with normal visual acuity will see a solid gray mass. Most printing processes exploit this fact to create the illusions of continuous tone from a bunch of discrete dots.
The actual size of the features that fall at the limit of visual acuity depend, of course, on viewing distance. With a little trigonometry, we can calculate the threshold of normal visual acuity for a given viewing distance. One minute of arc (1/60 of a degree) is 0.00029089 radians. We can calculate the limit L of visual acuity at distance D by the formula:
Table 2-2 shows the limit of visual acuity at different viewing distances, and the minimum print resolution, in dots per inch, needed to provide the illusion of continuous tone to an observer with 20/20 vision.
However, these numbers, although useful, aren't set in stone for the following important reasons.
So treat the numbers in Table 2-2 as useful guidelines rather than as absolutes. One reason for mentioning these numbers is to provide insight into the reasons we print at the resolutions we typically use, but another equally important reason is to help us understand what we need to do to keep our sharpening haloes near the threshold of visual acuity. By doing so, we can produce prints that appear sharp, yet lack the obvious and disturbing sharpening haloes that mar so much of the work we see in print.
I've observed the phenomenon so many times that I can't resist commenting on it. When some otherwise-sane photographers learn that a print was produced digitally, the concept of "normal viewing distance" suddenly changes from the distance at which they can see the image to one that is largely determined by the length of their noses. If you examine a 30- by 40-inch traditional darkroom print at that distance, you'll likely see grain artifacts that become invisible when you move back far enough to see the image. It's the same with digital!
Figure 2-23 contains 75 line pairs per inch, with increasing contrast from solid midtone gray at the top to solid black and white at the bottom. As you move the image further or closer to your eye, the point at which you can no longer discern the individual line pairs changes. As you move closer, you can discern the line pairs higher up the page, and if you move back far enough, the whole figure becomes one solid gray mass.
Figure 2-23. Line pairs with increasing contrast
Notice that the right side of the figure is different from the left side. The difference is that the left side of the figure has had no sharpening applied, while the right side has been sharpened. As a result, you can discern the line pairs on the right side higher up than you can do so on the left side at any given viewing distance.
This is exactly what good sharpening does. It increases localized contrast to reveal detail. But each printing process imposes its own requirements and limitations.
When we print, we need to decide how many pixels per inch we send to the printing process. In part, the decision depends on how many pixels we captured in the first place. We can create more pixels by interpolation, but interpolation can't create detailit simply takes the original pixels, spreads them apart, and creates new pixels with intermediate values in the spaces between the original ones.
When interpolation is combined with careful sharpening, it can improve the print results very slightly on some images, but it tends to be a great deal of extra work for questionable return. I recommend attempting only when printing at the uninterpolated resolution has clearly failed.
Interpolating downward to achieve the required output resolution is, however, necessary and normal. It's possible to drown printers in data: If you send far more data than the printer can resolve, it will either throw the excess data away (in which case you've just wasted some time), or, worse, it will attempt to use the extra data, with the result that detail gets blocked up instead of being resolved.
In either case, final sharpening for print must be done after any interpolation. If you downsample, the sharpening haloes can simply disappear as they're downsampled out of existence, and if you upsample, the sharpening haloes become too large and hence visually obvious. The print sharpening must always be done at the final print resolution.
The general rule of thumb for halftone output is to send a number of pixels per inch that corresponds to between 1.5 and 2 times the screen frequency in lines per inch. If the platesetter or the imagesetter is driven by PostScript (which they invariably are), anything over 2.5 times the screen frequency is automatically discarded, so it's an absolute certainty that there's no reason to send more than 2.5 times the screen frequency.
In the real world, I've yet to encounter an image that showed any visible difference when printed from 2.5 the screen frequency instead of 2 times the screen frequency. But images with fine detail generally reproduce better when 2 times the screen frequency is used rather than 1.5 times the screen frequency, particularly at lower screen rulings. At higher screen frequencies of 175 lpi and greater, the difference becomes more subtle and is apparent on fewer images.
Figure 2-24 shows the same image printed from a 300 ppi file (native resolution, 2x the screen frequency) and from a 225 ppi file (downsampled from 300 to 225 ppi, 1.5x the screen frequency). Can you see the difference? (There are differencesI had to sharpen the 225 ppi version more aggressively for output than I did the 300 ppi versionbut they're quite subtle!)
Figure 2-24. Resolution for halftone
A great deal depends on the software used to drive the continuous-tone printer, but a useful rule of thumb is to send pixels at the printer's native resolution. Some older dye-sublimation printers may still insist that you do so, but most modern continuous-tone printers have sophisticated controllers that will perform the necessary interpolation to the printer's native resolution.
Some advocate sticking to resolutions that are even multiples of the printer's native resolution. Doing so certainly simplifies the interpolation tasks that the software controller needs to conduct, but I've only found the practice to be advantageous with some older printers and with some entry-level dye-sub printers.
It's almost certainly a bad idea to send significantly more than the printer's native resolution, not because the interpolation algorithms will do a bad job, but simply because in doing you lose some control over the sharpening since the sharpening haloes will be downsampled along with the rest of the image. But it's not worth downsampling an image that had been prepared at 305 ppi for a Fuji Frontier to 300 ppi before printing it on a 300 ppi Océ LightJetthe difference is just too small to worry about on normal images. (You may see a detectable difference on synthetic targets made up of black and white line pairs, but that difference disappears with even slightly lowered contrast.) Most minilabs and online services who offer continuous-tone output specify 300 ppi for high-quality printing. If you aren't sure about the optimal ppi for your preferred service, ask!
Error Diffusion Dither Output
Most inkjet and color laser printers use an error diffusion dither of some kind, though the details are usually proprietary. Most color laser printers quote a resolution of 600 or 1200 dots per inch, while most inkjet printers specify some multiple of these resolutions. Canon's photo inkjet printers range from 1200 x 1200 dpi to 9600 x 2400 dpi, Hewlett-Packard's photo inkjets generally use 4800 x 1200 dpi, while Epson's inkjet printers typically specify 720 x 2880 or 1440 x 5760 dpi.
What these numbers represent is the addressable resolution of the printerthe accuracy with which it attempts to lay down dots of ink. Their relationship to the pixel resolution of the image is fairly indirect, but the optimal ppi value is lower than these numbers may suggest. The conventional wisdom is that the "native" resolution of Epson inkjets is 360 ppi, while that of the Canon and Hewlett-Packard inkjets is 300 ppi.
What my own testing indicates is that there's definitely no point in upsampling images to achieve a resolution higher than the native ones stated above, and little point in upsampling even to these resolutions. For large prints that are likely to be viewed from 20 inches or more, 180 ppi is probably plenty.
However, if you're making small prints, and you have real captured data (that is, with no interpolation) in excess of the native resolution, you may not want to downsample it to the native resolution. With Epson printers, there seems to be a small but useful advantage to sending 480 ppi of real data, suitably sharpened, to the printer. At higher resolutions the advantage diminishes, and sending more than 720 ppi actually seems to degrade the image.
Some pundits claim that you'll always get better results if you print at even multiples of the native resolution. This holds true for line pair targets, but on real-world images the benefit is much less certain, and may often be outweighed by the damage done by resampling the image. I almost always print at the native capture resolution, and if that turns out to be 342 ppi rather than 360, or 191 ppi rather than 180, I simply don't worry about it.
Note, however, that this is simply a statement of what I do, albeit based on considerable testing and experience. Each generation of new printers brings new capabilities and, probably, new challenges, so beware of any statements on optimal resolution for inkjet printers that claim to be definitive, and don't be afraid to put conventional wisdom to the test!
Sharpening for Output
Whether you adopt the multipass workflow I'll advocate in Chapter 3, Sharpening Strategies, or the more traditional single-pass sharpening, output sharpening is where the strongest sharpening comes into play. The key point to bear in mind when sharpening for output is that it's resolution dependentit's all about the size of the pixels.
Rememberyou have no control over how the pixels get turned into dots, so all you can do is to sharpen the pixels themselves. So for any given size of output, you need to sharpen higher-resolution images more aggressively than lower-resolution ones to achieve the same perceived sharpness. More aggressive sharpening can mean wider haloes, higher contrast between dark and light contours, or a combination of both.
The goal with output sharpening is produce a satisfactorily sharp image without introducing visually obvious sharpening haloes. To do so, the secret is to keep the size of the haloes below the threshold of visual acuity at the anticipated viewing distancethis is where the size of the pixels on output becomes a critical factor.
Bear in mind that the numbers given in Table 2-2 for the limits of visual acuity refer to very high-contrast edges indeed, since they're based on black and white line pairs. In practice, you can take some liberties with these numbers. The following rules of thumb have served me well.
Paper stock also has an influence. Inks bleed more on uncoated papers than on coated ones, so sharpening for uncoated papers has to be stronger than that for coated papers to achieve the same degree of sharpness.
Figure 2-25 shows the same native-resolution pixels sharpened for different print sizes and processes, zoomed in to approximately 400% view to make the differences obvious. The sharpening for the 4 by 6 inkjet print appears the gentlest of the three, because the inkjet can actually resolve the fine detail that wider sharpening haloes would obscure.
Figure 2-25. Size-Sensitive Sharpening
The haloes for the halftone sharpening are wider than for the inkjet, because it takes several pixels to make up one halftone dot. Hence rendering single-pixel details isn't possible (you could render single-pixel details by sending a lower-resolution image, but then you'd have obvious single pixels, which looks worse than bad sharpening). The haloes for the 85-line halftone need to be higher-contrast than those for the 175-line halftone to achieve the same apparent sharpness.
There's one last thing to note about output sharpening. Unlike the other sharpening factors I've discussed, the relationship between pixels and printer dots, for any given print process at any given resolution, is fixed, and doesn't depend on image content or source. I'll discuss the implications of this in detail in the next chapter, Sharpening Strategies.