Digital Image Basics

The Barest Essentials of Managing Digital Images

-by ctLow

-on this page:

I would like to present a few essential bits of information that I have garnered about digital imaging. I do not claim any particular expertise, and would like to hear any suggestions or corrections. Please do not copy or use in any commercial capacity without permission. Thank you. ctLow

Video vs. print resolution

Many image file formats specify a resolution - often using dpi, meaning "dots per inch." This means that if you make an image file, say from a scanner or a digital camera, and specify 200 dpi, the image will print at that resolution. Clearly, the larger the dpi the more detail in the image, although of course there are also reasonable limits, and the arithmetic tells us that doubling the dpi makes a file four times larger.

Vancouver from Granville Island (C) ctLow However, dpi refers to printing on paper. Many computer programs disregard the dpi, and display digital images at one image pixel per one screen pixel. Adobe Photoshop calls this "100%" for example. The Windows™ desktop does this; it doesn't matter what resolution is specified in the image file - it comes out "1:1."

For example, if an image is supposed to be 6x4 inches at 75 dpi, it will print at that size and at that resolution; another file of the same image could be 3x2 at 150 dpi ("rescaling"), which would be the same size in bytes, but print smaller and "finer." Both files, however, would look the same on a video screen, if displayed 1:1.

That means that a high-resolution image could occupy huge amounts of video screen space. To fill an 800x600 video screen with a 6x4 inch image would only require about 133 dpi. But at 300 dpi you would see less than a quarter of the picture. If you wanted it to appear 6x4 inches on a video screen, then you would need to know details about the monitor size and resolution which are not always available. The figure of 72 dpi is often used, but is only a very loose approximation. Regardless, when choosing a pixel resolution, you need to know how the output device behaves first; one dpi number will not likely work for video and paper printing.

File formats

Knot (C) ctLow If you've made an important exposure in a digital camera, or scanned an image that you might like to refer back to and manipulate later, then save it in a non-lossy format, like TIFF. (And 24 bit colour is plenty - the human eye can't see that much anyway.) True, it will occupy much, much more storage memory than a lossy format such as JPEG.

JPEG is good for web pages and emailing - only. Even then, usually you want nearly the highest quality possible - it's still going to be a much smaller file than a TIFF.

Every time you modify and then save an image, adjusting all of the many things you might want to adjust incrementally, a lossy format loses something, and a non-lossy format doesn't. So:

  1. always save a final copy in non-lossy format unless you're absolutely sure you won't ever need it again, and then
  2. convert to "lossy" only once you've finished manipulating the image and are ready to post or email it.

(Using compression like LZW on a TIFF is just fine: it's also non-lossy.)

Resampling vs. rescaling

Always assume that changing an image's size by resampling (Photoshop terminology) discards all of the original pixels and replaces them with recalculated ones. In so doing, you may lose faithfulness to the original image. Changing size by rescaling does not discard pixels - it just changes the printing resolution, but retains all of the "data."

Rescaling is like in the example above: converting a 6x4 at 75 dpi to a 3x2 at 150 dpi. The data stored in the image file does not change - it just prints smaller but more densely (or larger but more spread out). Because you're not altering any of the data, you can rescale endlessly and the image quality will not degrade.

Seed Filaments (C) ctLow Resampling, however, changes the data. A digital image's data is stored as a grid of pixels, lots of individual points of colour and tone. When you resample, you make a new grid, and the new individual pixels have to be invented from the old ones by a best-guess method.

That would be like going from 6x4 to 3x2 but leaving it at 75 dpi: three-quarters of the data will get discarded. And if you go back to 6x4, three quarters of the data will have to "guessed" back into existence, and it absolutely will not be just the way it was originally. Or, leaving the print size the same but changing the resolution ("dpi") works just the same.

Many programs resample very well - but if you keep repeating this process on the same image, you lose some faithfulness each time. So, if you have to resample, try to do it only once as you work from your original to your final image version.

I try to resample, where possible, using integer divisors or multipliers (2, 3, 4). I haven't seen visually that it makes a big difference, but the theory is this: if you resample down by a linear factor of 2 (as above), then although you discard three quarters of the data, the other quarter could remain intact. (I'm not sure that it does - it just could.) The new pixel grid lines up exactly with (a fraction of) the old one. But if you use of factor of 2.08 (and I choose this number because that's the factor for going from 150 dpi to 72 - see below), then none of the pixel grid points will line up, and all of the new pixels will have to calculated approximations of the old ones. Resampling up, let's say from 75 dpi to 150 dpi, at least one quarter of the new data could be identical to the old; but from 72 to 150, this no longer applies.

Anyway, as you're working with your image, save the file once in a while (only if you're very sure you're happy with it in its current condition!). Then, if you make a change you don't like, you can close it without saving, and then re-open it the way you liked it. (This is safer than using the Undo function, although I do use that a lot too.) In this context, if you resample but find it isn't quite the right size, don't resample again - go back to your original image size, and from there go directly to the next size you want to try. This keeps the number of times that your final image data has been resampled to "one."

In Photoshop, to rescale, choose Image -> Size, and uncheck the Resample Image box. To resample, I always use Bicubic, because it's listed in the help files as the best algorithm - just a little slower to execute.

Histogram Histogram

When scanning, or otherwise first manipulating a digital image, use the histogram feature to get a full tonal range - brightness, in other words. Clip off the unused ends so that the darkest darks are fully black and the lightest whites are fully white. Most of the time, this will make your image better, and if it doesn't, you can see it and "Undo" it or modify it. Using the histogram this way gives more control than using an automatic exposure control feature like Adobe's Auto Levels or the Microtek Scan Wizard's Auto Contrast, even though they often are good too.

Unsharp mask

Unsharp mask Photographs almost always benefit from sharpening with the USM. This isn't for trickery or to improve poor digital photographs or scans - it actually improves a digital image's faithfulness to the original. (Other image types, for example graphics as opposed to photographs, follow different rules.) There are three parameters to adjust, and the amount should start somewhere in the middle, and the radius and threshold at very low levels.

Radius has to do with how far the sharpening extends out from areas of tonal contrast, and threshold relates to how much contrast there has to be to apply the effect. If in doubt, underdo the USM a bit - too much amount or radius, or too little threshold will make your digital image look - well - digital! View on the monitor at 1:1 ("100%"), for greater control, even if it will print much smaller. Don't, as a rule, use other sharpening tools with photographs - stick with the USM. With rare exceptions, using the USM should be your last step before your final save.

Saturation

This is new ground to me, but I just came across some advice on the Internet newsgroup, alt.photography, that many people routinely punch up the saturation on their scans, again being careful not to overdo it (kind of until you like it, then back off a little?), but finding that it produces an image more pleasing, and/or more like the original. I've tried it on my last few scans, and so far am pleased, but it's early days yet!

In Adobe Photoshop, I open the Image menu, click Adjust -> Hue/Saturation, and go from there.

Scanning slides or negatives

Autumn Shrubs (C) ctLow This question comes up frequently on the photography newsgroups, and the standard answer is that you need a proper, dedicated film/slide scanner to do this well. Many flatbed scanners come with a little attachment for slides or negatives, and I have learned a few techniques which have returned much better results with slides than formerly (always scanning at maximum resolution being the main one), but in general, although you can get an image with these, it just won't turn out nearly as well as with a dedicated slide scanner. There is no magic way around this. Scanning negatives is more problematic, and involves things like learning about removing orange masks, and I haven't gotten into this in any serious way.

How much resolution to use?

This is a bit tricky, and certainly no one answer will suit all circumstances. Please feel free to correct me if I don't have all of this quite right.

Twig (C) ctLow You have to consider what resolution your scanner or camera can provide, and then how much the output device can handle. If in doubt, of course, too much resolution is better than too little. It works much better to resample down than up. So, if you have the disk space (and time, and computer RAM...), you might want to scan something at, let's say, 600 dpi, even if you think you might eventually save your final copy at 300 dpi. Especially if you're going to put your image through several digital manipulations, some people think that it's better to work on a higher-resolution file, and then resample down near the end of the process.

(Note the integer divisor: 600 divided by 2 makes 300; for that reason, I scan at a fraction of my scanner's maximum of 1200, usually using 600, 300, 200, or 150, and then resample for web/email JPEG's to 75 dpi, not the "conventional" [but arbitrary] 72.)

But if you're for sure only to going to display an image on a video screen, then something in the 70 to 100 range will suffice, and I think you'll be hard pressed to see any difference by resampling down from a higher resolution. (But beware: I think some digital cameras make 75[-ish] dpi files - but at a huge size; these can later safely be rescaled down to a more usual photograph size with higher resolution.)

Apparently, a print photograph just doesn't have more in it than 300 dpi. Even if it did, higher resolution than this is not visible - or I wonder if some would say barely visible - to the naked eye. Film (negatives and slides) contain more than 300 dpi - but not prints. That's why some film scanners go up into the 2400 dpi range - it allows large digital images to be created from the small film (see Rescaling and Slides above) - the same as creating conventional print enlargements.

Many publishers use 150 dpi for photographs in books, and say that higher resolutions just use up disk space without contributing more in picture quality. I think that fine art publications using top quality paper might disagree.

Many household inkjet printers claim resolutions in the 600 dpi range, but the technology they use for "dithering" tones and colours often limits their equivalent pixel resolution to much less. I find that higher-grade paper makes a big difference to photo printing quality, but of course there are still limits.

You can experiment yourself with all these settings, and see at what limits your own equipment stops producing better results. Zoom in and enlarge portions of digital images, or examine the output with a loupe, and compare versions using a lower original resolution to those using a higher, both on a computer monitor and on paper.

Photograph scanning recipe

Stump (C) ctLow So, translating theory into practice, here are the steps I go through when scanning a print photograph (not a slide, not a "document," not a "graphic" - only a print photograph). There are some variations, but on my (old) scanner, this seems to work well most of the time and with a minimum of fuss. Manipulating a digital image from another source, such as a digital camera, is similar except of course I skip the first few steps. This may seem like quite a production, but with experience it doesn't take long to get through it all.

  1. Clean the scanner glass and put the photograph in the scanner.
  2. Choose the scanning size. For example, I don't use the whole scanner surface for a 4"x6" photograph, but then will later need to enlarge the scanning area if scanning an 11"x14" document.
  3. Press the "Reset" button to return all parameters to baseline.
  4. Do a "Preview" scan.
  5. Crop the area I want to scan (and sometimes then enlarge that area to make it fill more of the monitor screen).
  6. Expand the brightness range with the Histogram tool.
  7. Choose a resolution. If this is a high-quality image that I might want to keep and use further in future, I will choose 600 dpi. If I know it's going to be a temporary image for computer monitor use only, I might use 75 dpi. For casual snapshots that I'm sure no one will ever want to print or at least to enlarge, I will select 150 or 300 dpi.
  8. Make sure "Color Correction" is off (personal preference).
  9. Scan!
  10. Close the scanning window, which returns me to my image editing program.
  11. Save the image as a TIFF. If it's a really precious photograph, I may then find the file on my computer with a file management program (like the Windows® Explorer), and make it "read-only" so it cannot be further altered. Then, make a copy of it and work on the copy - usually from the image program's "Save As" menu.
  12. Then, as I go through the following steps, I will occasionally "Save" the image as I go, if I'm happy with my results so far, making it easier and safer to go back a step should that seem appropriate. At any time, I can repeat that cycle, of saving a "read-only" copy and then renaming it before doing further work. That let's me try several variations of it to compare with each other later. Multiple "saves" are not usually necessary - but available when desired.
  13. Do "auto-levels." This is another automated histogram function, but it simply makes many photographs look better and/or more faithful to the original, so I try it. If I don't like it, I "undo" it.
  14. Bump up the "Saturation" a little - generally not more than 8-12%. If the colours actually start to look noticeably vivid, I back it off a bit.
  15. Lower the resolution. Generally, 150-300 dpi is plenty for a print photograph. Working on the higher resolution until this step is thought by some experts to give you a better final result.
  16. Do the Unsharp Mask.
  17. Save the image, still as a TIFF.
  18. If I want a JPEG version, then resize the image, if desired, and then save it in JPEG format, using a high-quality setting;
  19. or, if I want the JPEG for computer monitor display only, then reduce the resolution (from the TIFF version) to 75 dpi and save it using a lower-quality setting.

Scanning tips, by Wayne Fulton

Water Grass (C) ctLow Wayne Fulton has packed more information than you could possibly imagine into a monograph called A few scanning tips. You'll learn much more than just about scanning. Although wordy and repetitive in the extreme, in the final analysis the content is there, and that's what matters. Peruse his fascinating web site, then buy the book. I get no consideration for saying this - it's just what I think. I get noticeably better scans since I learned so much beyond just pushing the Go button.

Digital Image Basics is © 2003-2004 ctLow
-first posted: 2003-01-06
-updated: 2007-04-15