Bicubic limit. of resizing

TT
Posted By
Tom_Thompsonw
May 18, 2005
Views
516
Replies
15
Status
Closed
Serif Photoplus 8, another photo editor I use, has an option for resizing images at Lanczos, a step beyond Bicubic.

Laczos as explored some online has been talked about as preferred of greater quality than bicubic.

Other than resizing in Serif Photo Plus 8, saving it, opening it into Adobe Photoshop 7, is there a plugin for Photoshop that provides for Lanczos resize within Photoshop? Serif program takes a long time to open.

Master Retouching Hair

Learn how to rescue details, remove flyaways, add volume, and enhance the definition of hair in any photo. We break down every tool and technique in Photoshop to get picture-perfect hair, every time.

CC
Chris_Cox
May 19, 2005
Lanczos (really Lanczos weighted sinc filter) is different from Bicubic — but definitely not "a step beyond" (unless you mean "a step in the wrong direction"). In most cases it gives visually worse results than the Bicubic resampling kernels in Photoshop due to excessive ringing artifacts.
TT
Tom_Thompsonw
May 19, 2005
Thanks for the provocating direction to look closer at what is actually happening with what others promote as much preferred.

I generally do that to severe degree. But that slipped passed.

Since questioning it has got me wanting to look at the other methods within Photoshop other than Bicubic to see if there might be reason to want to use them for anything. Since the options are there.

Thanks again.

If familiar with any of the commercial resizers and wanting to offer thought about, "Genuine Fractals" I’ve heard hyped pretty big, including in a poll of preferred resizing methods, I would be a lot interested.

Tom
F
Friend
May 19, 2005
I was very pleased with Genuine Fractals 4 under CS. I installed CS2 last night and GF4 does not function. I am awaiting advice from technical support at LizardTech.

N. Schwartz

If familiar with any of the commercial resizers and wanting to offer thought about, "Genuine Fractals" I’ve heard hyped pretty big, including in a poll of preferred resizing methods, I would be a lot interested.


Tom

If familiar with any of the commercial resizers and wanting to offer thought about, "Genuine Fractals" I’ve heard hyped pretty big, including in a poll of preferred resizing methods, I would be a lot interested.
Tom
JH
Jake_Hannam
May 19, 2005
Tom,

Rather than spending money on Genuine Fractals or some other program just for resizing purposes, I would use the money to upgrade to CS2 because the built in sharpening options have been expanded since version 7. Genuine Fractals has been reported to have problems with the new CS2 PLUS they tend to focus their efforts to the Mac side and seldom publish fixes or updates to the Windows versions.

You can download a trial version of CS2 from Adobe that is good for 30-days. If you like it, you can upgrade for $149 which is not a whole lot different than buying GF or some other resizing program.

Jake
GH
Gernot_Hoffmann
May 19, 2005
Tom

some comments about resampling (instead of resizing):

1. Resampling for scalefactors about 1, rotation, morphing, perspective:
Bicubic or Bicubic B-Spline+Sharpening (not available by name in PhS, but perhaps somewhere hidden).
Lanczos is indeed bad, like all windowed-sinc methods.
2. Downsampling (less pixels):
In several steps by factor 0.5, final step with by e.g. 0.865 Everything Bicubic. Or Bicubic B-Spline as above.
3. Downsampling for thumbnails: this needs other methods, like Box Averager or First Blur – Then Nearest Neighbour.
4. Upsampling (more pixels):
It’s a widespread misunderstanding that this can improve the printing quality. It doesn’t, not for offset and not for large format printing (tested by three RIPs).
Upsampling shouldn’t be used for serious purposes. For posters, where the viewing distance is larger than normal, the inter- polation is done by RIPs. No visible blocks.
5. For resampling (e.g. rotation) many scientific investigations were published. Some methods are far too time consuming. Sometimes they were not tested for real-life images. Inter- pretations in the frequency domain are often misleading.

This post shouldn’t be understood as a comment on any software, especially I’m not criticising PhS. Some issues are anyway left as a matter of opinion (different results for different image contents).
Illustrated examples and explanations are here (1100kB): <http://www.fho-emden.de/~hoffmann/bicubic03042002.pdf>

Best regards –Gernot Hoffmann
D
deebs
May 19, 2005
Lanczos is free with IrfanView which is also free
CC
Chris_Cox
May 19, 2005
In CS and CS2, you shouldn’t need to do downsampling in steps. (we should have the full prefiltering done, and downsampling in steps will lead to artifacts)
GH
Gernot_Hoffmann
May 20, 2005
Is it a sequence like
Blur – Bicubic – Sharpen ?

Best regards –Gernot Hoffmann
D
deebs
May 20, 2005
Hmmm more of a composition rather than a sequence.

Sequences usually have u(i-1), u(i) and u(i+1) with some specifics and relational details.

On the other hand a composition is just that – a string of things composed from some options.

If it is a sequence it will be groovy to know the details though
CC
Chris_Cox
May 21, 2005
Gernot – nope, it spreads the bicubic kernel to accomplish the blur. And the bicubic kernels have sharpening built-in.
DM
dave_milbut
May 22, 2005
thanks for a clear explination gernot (i usually find it difficult to follow you, as your knowlege of this kind of stuff is way over my head! <g>). and thanks chris for the new resampling methods in cs & cs2. love them – but STILL trying to get my mind around which is better in which circumstances. most cases it’s trial and error for me! 🙂

is the resampling some of your work chris? very nice, and suprisingly quick.

cheers to the both of you
dave
GH
Gernot_Hoffmann
May 22, 2005
Dave, thanks for the feedback.

As you can see I’m reluctant suggesting newer and better methods.

For PhS users this doc by Bart van der Wolf might be interesting: < http://www.xs4all.nl/~bvdwolf/main/foto/down_sample/down_sam ple.htm>

For those who are interested in theory: Google search by Thevenaz
Meijering

My layman opinion: for rotation, moderate scaling, perspective correction and morphing, standard bicubic is fine.
I’m refusing to UPSAMPLE any image for serious applications.

Best regards –Gernot Hoffmann
DM
dave_milbut
May 22, 2005
you still on 7 gernot? so you haven’t seen the new methods? have you tried them in the cs2 (or cs) 30 day trial?
D
deebs
May 22, 2005
Interesting views and thanks for the google tips but – and it’s a great big deebal B-U-T – I think context has to be introduced at some point (IMHO)

It’s probably easier to summarize this thread as: all tests are context sensitive.

The images of disks and concentric rings will play, I am sure, a very important role in establishing and developing fidelity in algorithms but sometimes in plain old pragmatic terms processing fidelity is not the end test. I propose/posit that the end image is the best outcome and that will always be subject to human perception and the quality of displays, printers and maybe even aesthetic qualities of observers (explicit or implicit).

For example, levels adjustments layers (as others or filters) are quite wonderful in helping overcome some limitations in dsp with respect to images but I guess they are tools to provide alternate ‘realities’ (steady now, this is a serious point)

In other words, seeking the reality of a 2D image is for most purposes constrained to human perception rather than techical or abstract definitions of reality.

This is not to rubbish research or findings – quite the opposite really. It is to posit that the context of assessment or outcomes is itself subject to the governing assumptions behind those views.
JJ
John Joslin
May 22, 2005
I couldn’t have put it better myself.

Come to think of it…

Oh, never mind.

Master Retouching Hair

Learn how to rescue details, remove flyaways, add volume, and enhance the definition of hair in any photo. We break down every tool and technique in Photoshop to get picture-perfect hair, every time.

Related Discussion Topics

Nice and short text about related topics in discussion sections