Scaling Algorithm Bug In Gimp, Photoshop, Others

[Kirk Gittings]

Kind of a schizophrenic or poorly written article I think Ken. That quote I have above comes directly from one section of the article.

See this thread in Luminous Landscape with contributions by both Mad Man Chan and Jeff Schewe, both of whom do work for PS. This issue is far from clear.

http://luminous-landscape.com/forum/...=\gamma+error\

[Ken Lee]

So... what's the visual effect we're talking about here? Are there any posted examples to look at?

Edit: Ah, this is the old linear vs. gamma space argument that's been hanging around since the beginning of computer image editing. Not that big a deal. Maybe we'll eventually end up with enough standards that we can make a reasonable fix for this some day. Until then, almost everyone will continue to ignore it -- SOP.

I'm not familiar with the issue. Could you elaborate a bit ? No big words please

Is this a Tempest in a Teacup - and a degredated one at that ?

[Kirk Gittings]

Ken, apparently this has been discussed for a couple of years and not gotten much traction-see my link to LL above.

[harrykauf]

Tried it on an image and don`t like the result. Creates dark halos around bright areas.
photoshop cs2

[Ken Lee]

Ken, apparently this has been discussed for a couple of years and not gotten much traction-see my link to LL above.

Excellent - Thanks !

Next time, I'll look around first.

[Francisco Disilvestro]

I'll second Kirk comments. There is another thread at the LL forum that discussed rencently the issues with downsamplig http://luminous-landscape.com/forum/...&hl=downsizing

This is a quote from Jeff Schewe:

I was involved in the original resampling testing back when Photoshop incorporated multiple flavors of Bicubic with Photoshop CS. The engineer working on the algorithms, Chis Cox, went through literally _ALL_ of the various resampling schemes out there. In real world testing with photographic images (both scans and digital captures), we all settled on Bicubic Sharper for downsampling, Bicubuc Smoother for upsampling and regular Bicubuc for general use or as a substitute if Bicubuc Sharper produced artifacts on an image by image basis

You can always find fabricated images that will make the algorithms to fail, just look at each color channel of the Dalai Lama image and I think nobody will say that's a real world case.

Anyway, an interesting point is that in digital image processing, downsampling an image poses a challenge even greater than upsampling. In the analog world, when you optically reduced the size of an image you got a better one. This is not always the case with digital.

[Kirk Gittings]

Next time, I'll look around first.

It was a good question- made me do some thinking....I never knew there was issues like that with down sampling!

[Bruce Watson]

I'm not familiar with the issue. Could you elaborate a bit ? No big words please

I'll just screw it up -- I don't remember enough. What I remember is that it has to do with gamma corrections. That without an embedded profile of some type, the resampling algorithms don't know how to, say, combine four pixels into one, because they don't know the gamma of the image, so they just do a straight (linear) average. Which is right if the image is a gamma 1, but can be wrong if the image is a gamma >1 because the algorithm doesn't take the non-linearity of the curve into account.

This all dates back to the first attempts to do this -- back before there were such things as ICC profiles or any kind of color management. IOW, back when the algorithm had no way of knowing what the gamma correction should be.

This is IIRC a fairly subtle effect. Considerably more subtle than the softening of detail you get from that same down sampling of the image, for example. You'll see the softening long before you see any of the tonal shifts the original story is about, again IIRC.

Not very clear I know. Surely someone can do a better job of explaining this than I.

[Ken Lee]

- Thanks Bruce -

I think we've put this one to rest.

[Patrick Dixon]

Not very clear I know. Surely someone can do a better job of explaining this than I.

I'd say that's pretty good - clear enough for me.