Toronto Photography Meetup Group

http://www.4p8.com/eric.brasseur/gamma.html

from the page:

There is an important error in most photography scaling algorithms. All software tested have the problem: The Gimp, Adobe Photoshop, CinePaint, Nip2, ImageMagick, GQview, Eye of Gnome, Paint and Krita. Also three different operating systems were used: Linux, Mac OS X and Windows. These exceptions have subsequently been reported: the Netpbm toolkit for graphic manipulations, the developping GEGL toolkit, 32 bit encoded images in Photoshop CS3, the latest version of Image Analyzer, the image exporters in Aperture 1.5.6, the latest version of Rendera, Adobe Lightroom 1.4.1, Pixelmator for Mac OS X, Paint Shop Pro X2 and the Preview app in Mac OS X starting from version 10.6.

I suspect few photographers (except maybe the pixel peepers on dpreview ) have noticed the problem, and even fewer care enough to understand the reasons behind the problem. I noticed it in the past when scaling down line art and text with anti-aliased edges, but it wasn't a big enough of a problem to explore alternative scaling methods.

Incidentally, it's the same reason why "middle grey" is 18% grey and not 50% grey...

I never knew or understood the WHY but I have known for a while that Adobe's implementation of downsizing was less than optimal. I have read many criticisms of Adobe's implementation of downsizing. Neither Adobe's bicubic nor bicubic sharper is the best. It's a problem that Adobe engineers have neglected. Certain optimal strategies have been proposed, NONE of which are worth the problem to implement for me - a mere hobbyist. Apparently there is better downsizing technology around - developed in astronomy photography - that involves pre-blurring of the photo and then...well, I don't remember the rest. Once the technique involved degaussian blurring on another layer in order to optimize sharpness, I figured I'd avoid trying to wrap my head around it. Some heuristics are just beyond my desire to understand. I don't imagine painters get all fussed up in the technology of their brushes and paints so then why the hell should I. I don't imagine all musicians get too wrapped up in the physics of acoustics and vibrations of different matter so why am I trying to srap my small brain around that. Just gimme a workaround.

So, I created a PS action that downsizes in small 10% increments. Many of the downsizing artifacts do not occur when downsizing at that magnitude. So I just downsize using this action until I get close to my desired target size. The result is much cleaner than doing the downsize in one step.

Do it yourself. Open a large file. Downsize it several times to a desired target size. Pick something small like 500 pixels at length. Then take the same original file and downsize it once. Compare the 2.

A lot of commercial landscape photogs are aware of this.....people printing downsized versions with lots of detail.

I think most photographers who use photoshop have noticed this type of effect.
Isn't that why we're all told to apply a bit of sharpening after downsizing for web?

What they have done is artificial synthesize a test image that has every second line a constant grey scale this causes re-sampling aliasing errors with most resizing algorithm Interlaced broadcast TV has developed filters that eliminate these problems. Incremental by 10% +- a random different amount for each cycle gets around many of the aliasing problems. For downsizing only a less then a pixel Gaussian blur for each cycle with one final sharpening helps with any jagged line problems.

For the example used try expanding the narrow contrast range to the whole range then downsize and then restore contrast and colour saturation and you should be able to get an acceptable image except for the outside edges where you have to use a different trick.

Actually the sharpening is to make edges more visible not directly because of downsizing effect. Even full size prints usually need some sharpening.

Resampling has been pointed out to cause problems. I've left this box unticked. Though this was written when CS3 was the latest, Convert and Assign Profile are different and might cause confusion.