I'm seeing two problems with the Bicubic Automatic setting in the Image Size dialog.
1. On reduction, this produces oversharpening. Skin particularly comes out awful. Or maybe it's just that the sharpening is being applied equally throughout the image, instead of intelligently as with the Smart Sharpen command.
2. If I change this setting to something else, it reverts back to Bicubic Automatic the next time I enter the dialogue. In the past, this setting has always been sticky. It's very annoying to have to be careful to change it each time.
To sum up, you've made it hard to change from a default setting that is not working well!
Aaron
''Bicubic Sharper'' just does that. Adobe's ideas of ''best for downsampling'' and mine just differ. Have you used 'Bicubic Sharper'' before?
''Bicubic'' is what Photoshop has always defaulted to up to now.
Make your default ''Bicubic Smoother'' to completely avoid any sharpening (which by the way there seems to be just a little of in just plain ''Bicubic'' as well).
-Noel
I believe I've tried all three choices, and never had a problem with any of them, but usually had to sharpen additionally anyway, regardless of the choice. That's why I've stuck with plain Bicubic. So, it seems to me that Bicubic Automatic is sharpening more heavily than any of those choices. But I could be misremembering, or maybe I didn't test thoroughly enough. In any case, an automatic function of this kind shouldn't be making bad choices.
Okay, I did some comparisons of the various forms of downsampling. What I did was find a representative image with some good contrasts, then downsample to 25% using the various methods in Photoshop CS5 and CS6, then re-upsample to original size using Nearest Neighbor, so we can clearly see the pixels. Here are the results:
Photoshop CS5:
Photoshop CS6:
Based on these, I'd say there's actually some USM sharpening in Bicubic Smoother as well, more than just plain Bicubic!
Seems pretty obvious to me that just plain Bicubic is still the best, but to each his own!
-Noel
Noel Carboni wrote:
Based on these, I'd say there's actually some USM sharpening in Bicubic Smoother as well, more than just plain Bicubic!
Seems pretty obvious to me that just plain Bicubic is still the best, but to each his own!
Pretty sure it ain't USM...but there IS a sharpening component to BiCubic Sharper (as the name implies). I actually think the sharpening is a bit more advanced than USM but I couldn't tell you exactly what algorithm it's using but I think it's a hybrid–Chis Cox would know…
What I'm not sure of is whether or not CS6's Auto is picking up on LR's Adaptive Bicubic. In LR there's an interpolation between Bicubic Smoother/Sharer based on the amount that the resampling is doing. Close to the original size it's pretty much 100% regular Bicubic, the more you enlarge or reduce the more it moves tower Smoother (for up) and Sharper (for down). Again Chis would have to fill in the blanks…
I "think" the CS6 Auto is simply picking and using Smoother for upsample, Sharper for down sample and Bicubic if not changing the overall pixel dimensions. But again, I don't know for an absolute fact…this is a "Chris question".
:~)
In any event Image Size always follows the default resampling set in the PS prefs...it allows you override in the Dlog but is not sticky.
I find that the results from using CS6 Auto is a bit too "crunchy" when downsampling to perhaps only 25% of the original size (as you might do for Web display purposes).
Also there is usually some considerable darkening/clogging of the mid-tones when you downsample and it would be great if a correction for that could be built-in to the resampling algorithm.
Noel, I have to agree with you. Based on these samples, it seems wiser to stick to standard Bicubic and sharpen separately -- which, luckily for me, was what I was mostly doing anyway. It never occurred to me that "Bicubic Smoother" would sharpen MORE than standard Bicubic. I assumed that Bicubic was somehow positioned midway between Smoother and Sharper, in terms of sharpening. The names are thoroughly misleading!
Also, looking at these samples, I'm completely at a loss as to why Bicubic Sharper is said to be for reductions and Bicubic Smoother is said to be for enlargements, since a higher-resolution image needs and tolerates MORE sharpening than a low-resolution one. (At 300 ppi, I typically Smart Sharpen at 100%, but at 72 ppi, typically only about 25%.) Now I understand why Scott Kelby suggested reversing the settings and using Bicubic Sharper for enlargements. And now I know why my test reductions with Bicubic Automatic looked so bad.
Sharper == reducing in size, because reduction leads to a lot of blurring and loss of detail and needs some sharpening to make it look right.
Smoother == enlarging in size, because enlarging needs that extra smoothing to keep edges from showing artifacts, and a little sharpening to keep edges crisp.
Using them backward really won't help you (and would only look even halfway decent on *perfect* test images).
The names are quite correct, even if the actual parameters aren't as one dimensional as you assumed.
Chris Cox wrote:
It's not the same math as LR, but similar in concept.
Thanks for the confirmation...
As far as Sharper/Smoother, these variants were created specifically to address the problems with resampling up or down. Yes, Sharper has a sharpening component, but it's not really useful for upsampling as Bicubic Sharper will introduce sharpening artifacts that preclude other post upsampling sharpening–which must be done separately from the upsampling to get optimal results. Yes, on the "surface' some might see useful results with upsampling but it's not as good as applying post upsample image sharpening. This isn't hard to prove...see this article The Art of the Up-Res from Digital Photo Pro magazine (and yes, I wrote that article a while ago, but it still is applicable now)
Well, Jeff, you too seem to be agreeing with Noel, in saying that standard Bicubic is the way to go, at least for enlargements. I understand now from what Chris said that the name Bicubic Smoother refers to blurring being applied, not to a weaker sharpening. Yet he does confirm that some sharpening is applied as well, which according to you, Jeff, is not a good idea at this stage.
As for Bicubic Sharper, I completely understand that reduction requires sharpening, but the amount being applied here seems just too much, at least for the Web. So, I'll keep sharpening as a separate step, after reduction with standard Bicubic.
I'll definitely read your article, since I've already read two of your books!
I agree with Noel.
Unfortunately both recommended down-sampling methods in CS6 (Bicubic Sharper and the new Auto) do over-sharpen for most kinds of subject matter.
It really would be better if the degree of sharpening in both algorithms could be toned down a bit so that the User can then add further Output Sharpening or local sharpening at their own discretion.
CameraAnn wrote:
Unfortunately both recommended down-sampling methods in CS6 (Bicubic Sharper and the new Auto) do over-sharpen for most kinds of subject matter.
Yes, that's my impression, too. I do understand that digital images need some re-sharpening after (or while) downsizing ... but Photoshop's Bicubic Sharper now grossly exaggerates the sharpening effect in CS6. Strange I never noticed this in CS5 or earlier.
One thing I did not realize (noticed this when testing out Bicubic Automatic) is if you use one interpolation method when resizing a Smart Object. Then return to another. Then Edit Contents of that SO, Photoshop will re-render the scaled SO with the current interpolation method, even if you did not additional scaling to the SO.
I suppose that makes sense, Charles, insofar as it has to resample the changed SO data to make it the proper size.
But what you're really saying is that the interpolation method is not stored in the file itself; it uses the ambient resampling, and I can see how that could be unexpected with a file edited on multiple different machines (or at different times) with different settings.
-Noel
North America
Europe, Middle East and Africa
Asia Pacific