Link This | Email this | Blog This | Comments (0)

Resolution in general
November 13, 2007

The time-tested “rule” about resolution is that you must have two pixels in your file (linear) for every halftone dot in the output. The 300 ppi file is still considered a minimum, but experimentation over the years has shown that you can get-by with much less.


It’s important to remember that the “rule” is based on Harry Nyquist’s research in radio physics in the 1930s. Nyquist’s Theorem, known to all engineering students, indicates that you need to transmit twice what you expect to be received (the short version) in order to have a reliable signal. RF noise, interference, and sun spots all contribute to signal loss in radio transmissions, but those forms of interference are not present in digital halftones.

Years ago I had a Crosfield drum scanner – a 646IM – a Mother-of-All-Scanners that could make a 30 x 40 inch film separation set. It used “turret” settings A, B, and C, which defined the sampling resolution. In reality, these changed the feed-screw speed of the scanning head, resulting in three different “resolutions” of scans. Since nothing was stored for longer than a single rotation of the scanning drum, we never knew the actual pixel count of the file we weren’t creating.

The subject became highly controversial when Photoshop became popular. I had heated discussions with many a hard-line “purist” who insisted that we have “always” worked with 2:1 ratios. But, in the era of analog photography we had no resolution concept at all – just grain. In the quasi-digital Crosfield era we had a hybrid system which stored small amounts of tonal information just long enough to record it to film. In the all-digital era we have boat-loads of pixels, and we store all of them. In fact, the Crosfield only scanned above 300 ppi on the top two of its imaging turret settings. The lowest resolution turret resulted in a file of about 225 ppi. Yet images still looked great at this resolution.

Photos of people, especially, can be managed at resolutions much lower than 300 ppi, as the soft-edges of the human face are not obviously harmed by a low sampling ratio. Machines, cloth, furniture, paintings (sometimes) all benefit from higher resolution. And, anything sampled at greater than 2.5:1 (375 ppi for 150 lpi reproduction) will be sub-sampled by a PostScript RIP anyway, so there is no particular reason to sample at rates higher than necessary.

I tested this once in an empirical test (trial-and-error). I ran film on a Hurkules imagesetter at 3,000+ resolution. The tests ran from ratios of .85:1 up to 4:1. For people photos, resolutions from slightly less than 1:1 all the way up to 4:1 looked about the same. Only eyeglass frames, with their sharp edges, were a give-away that the resolution was better. For those images sampled at a lower rate I also tried applying a little bit of Unsharp Mask, and achieved about the same visual result.

Photos of machines and gears showed signs of low-sampling at rates lower than 1.5:1. Again, a bit of USM and the appearance was improved.

With a consumer digital camera with 6 Mp resolution, you have just short of enough resolution to bleed a letter-size page at 300 ppi. But, it’s OK to scale the photo up a little bit, bleed the page, and “sacrifice” image resolution a bit. In fact, you would be very close to 2:1, but not close enough to please the purists.

I am a purist when it comes to image quality. Really. But, I have also learned from experience to apply my  rules where necessary, and not worry about things that don’t matter. Being too much of a purist on image resolution will only give you ulcers – not better images.

Posted by Brian Lawler on November 13, 2007 | Comments (0)