Archive for June, 2008

Scanning vs Upsampling to Enlarge

June 21, 2008

mea culpa

I have to eat some crow and apologize to Steve. In the HAL-PC digital photography SIG on Saturday June 21,2008, I gave some advice and made statements about this subject that were in one case probably wrong and in another uncertain.

Technology had passed me by. A few years ago home flatbed scanners could not really (optically) scan at resolutions above 300-600 ppi. As I understood the example given in the meeting, the question was, “If you are going to scan a 2″ x 2″ print and enlarge it to 8×10 inches should you scan at 300 ppi and upsample to get the required resolution or should you let the scanner produce a scan of the required resolution.”

To be clear at the outset, if you scan and enlarge a print more than a modest amount the result will be a degraded picture. This article assumes you want make such an enlargement anyway and are prepared to accept a poorer result.

First, except in rare circumstances, no more image content/detail can be obtained by scanning a print at greater than 300 ppi. So, to make a bigger print, in one way or another pixels must be generated that do not increase image detail, but instead simply “pad-out” pixel count.

To scan a 2″ print and enlarge it to 8″ requires a 4x increase in resolution. If you want to print the image at 300 ppi and 8″, four times 300 ppi is 1200 ppi. One can either scan the small print at 1200 ppi or scan it at 300 and upsample 4 times in your photo editor. Until recent years if you scanned a print at more than about 600 ppi the scanner interpolated/upsampled (faked in pixels) to create the higher resolution. I said in the meeting the upsampling algorithm in your photo editor will probably be as good as, and maybe better than the one used by the scanner, thus if I must enlarge a print I scan at 300 and upsample in Photoshop. In saying the scanner must upsample the image to get the higher resolution, I was probably wrong. That’s no longer necessarily true.

I rarely have occasion to upsize a print, and the world changed while I wasn’t paying attention. Today it’s common to have home scanners that will produce real scans at 2000+ ppi. So, my statement that to make a marked enlargement the scanner must upsample is no longer true. If you have a modern scanner it can probably scan at 1200 ppi or more without upsampling. Thus, today the choice is between increasing the scanner resolution beyond the useful picture content it can produce in order to pad-out the extra pixels, or scan at the highest meaningful resolution (300 dpi) and upsample in your photo editor.

After doing some some Googling, I haven’t found an understandable discussion of this choice. Some theoretical considerations suggest if there’s any difference you may get smoother edges in high contrast areas, but my tests didn’t bear this out. And, in making these illustrations I found it was much faster to scan at 300 dpi and upsample, so one downside to using higher resolution scans is, it takes longer. I asked a authority on scanning and scanners whose opinion I respect and he said,”My opinion is neither method will be very satisfactory but the larger scan is your best shot.” For reasons that are unclear to me, my tests did not bear this out.

When doing my tests I found that using the bicubic “smoother” algorithm, which Adobe recommends for upsampling, produced a loss of image definition compared to the higher resolution scans. Regular bicubic did not cause this loss. The illustrations I link to below use regular bicubic.

Unfortunately, I wasn’t able to put my hand on an ideal image to illustrate the difference. To my eye these results of using the two methods seem to give better definition using upsampling. Perhaps there’s a sharpening effect from the bicubic algorithm.

The images in my tests were scanned from a sharp 4×6 print.

I couldn’t figure out how to make “mouseover” comparisons in this blog editor, so Click Here to see the results of my trials.

The images are fuzzy because they have been magnified 2 1/2 & 4 times the maximum useful resolution of these prints as would be required to print 4 x 6 scans as 16 x 24 — 4x, or to print them at 10 x 15. This is the same as enlarging a 2 x 2 print to 8 x 8 or 5 x 5 respectively. Thus such a large size change is going result in a big blurry image.

Below is part of the image from which the samples were taken. It’s shown at a reasonable resolution — about 250 ppi. This crop a small portion of the original picture.

.

As I said earlier, these pictures are not ideal for illustrating the difference between the two methods. To best illustrate the difference the picture needs to have some high contrast shapes, like a clock face or the text on a sign. If I run across a better picture for this I’ll swap these out.

Advertisements