Pages: 1 [2]   Go Down

Author Topic: image up-sizing  (Read 8636 times)

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8914
Re: image up-sizing
« Reply #20 on: August 08, 2016, 11:50:19 am »

That's very insightful Bart, thanks very much.

When enlarging grainy images, I see that Perfect Resize does a much much better job than Photozoom Pro, which tends to generate weird artifacts in the upsampled grain structure.

These programs obviously do not know what is noise and what is detail. I would either try and (locally) reduce the noise of the source image before upsampling or, if that is not possible, e.g. use a layer mask to reveal a Bilinear upsampled blurred version of the image for the sky regions. If the original image is pre-treated with Topaz Detail, then one can use settings to reduce noise amplification (with a negative 'boost' setting).

Quote
I'm seeing that Photozoom Pro's "perfect edges" are not as perfect as I used to think they were. I've attached a magnified portion of your first example, and the blue arrows point to some of the offending regions where Photozoom Pro doesn't seem to do as well.


The question (besides the settings that were used) is which is more accurate though. Maybe the edge is not as sharp and that is just replicated by a less abrupt transition. In actual print, I do not think it would be clearly visible or distracting, if at all.

Quote
In a test image of mine, I found that PZP tends to emphasize any zipper artifacts from demosaicing which Perfect Resize avoids. I agree that when the edge is rendered sharply, PZP is sharper. That is quite clear in the new owl eye crop. It's interesting to study where one does better than the other. Indeed, a lot can be invisible at print sizes.

Artifacts will be magnified, so it's best to avoid them in the source image. My experience with Capture One showed that since version 7, the stairstepping or mazing or zipper kind of artifacts are significantly reduced. However, with AA-filterless cameras, aliasing artifacts can cause all sorts of issues which will only be magnified when creating large output. So good tools with proper filtration can help a lot. So if you cannot avoid zipper demosaicing artifacts and they become visible in the actual output, then perhaps Perfect Resize is the better tool for those.

Quote
I see that Perfect Resize generates some weird color fringing artifacts on the high contrast edges. What's up with that?


Maybe different control settings will suppress that? Perfect Resize has quite a number of controls that need to be balanced against each other. And maybe it has to do with the amount of noise/grain addition.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

rogerxnz

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 527
    • Hayman Lawyers
Re: image up-sizing
« Reply #21 on: August 09, 2016, 11:04:01 pm »

Thank you, Bart for your examples. I notice you have provided some comparison examples from LR but not PS. Do you consider both LR and PS produce the same results? Or is there another reason for not showing the results of using PS to resize images?

I have seen some old reviews of image resizing apps but no reviews from the last two years. I understand PS CC now has improved resizing with a preserve details option and noise control. Have you compared PS's current resizing to the other dedicated resizing apps?

I don't want to put you to any bother . . .

Lastly, do you think the apps differ in how they handle B+W images compared to colour ones? What would be the best app to resize a B+W photo by 137%?
Roger
Logged
Roger Hayman
Wellington, New Zealand

samueljohnchia

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 498
Re: image up-sizing
« Reply #22 on: August 10, 2016, 02:53:41 am »

Maybe different control settings will suppress that? Perfect Resize has quite a number of controls that need to be balanced against each other. And maybe it has to do with the amount of noise/grain addition.

Unfortunately at all texture settings, with zero addition of noise or grain and zero USM applied, there are still color fringing artifacts. It's in both the examples you've shown too.
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8914
Re: image up-sizing
« Reply #23 on: August 10, 2016, 11:21:54 am »

Thank you, Bart for your examples. I notice you have provided some comparison examples from LR but not PS. Do you consider both LR and PS produce the same results? Or is there another reason for not showing the results of using PS to resize images?

Hi Roger,

The default resizing quality of PS is not as good as from LR. I also suspect that there are more people using LR for (large) output than from PS.

Quote
I have seen some old reviews of image resizing apps but no reviews from the last two years. I understand PS CC now has improved resizing with a preserve details option and noise control. Have you compared PS's current resizing to the other dedicated resizing apps?

No, I haven't, because I do not support the CC subscription model for a mature software product (unless I would get paid instead for the beta testing that current paying subscribers seem to be involved in lately, but that's a different topic), and trialing it will almost certainly break my rock solid 'CS-6 Extended' installation. But then I also haven't seen many examples produced by PS CC users either, which is strange if it were that much better.

Quote
I don't want to put you to any bother . . .

No problem, if there is no gain for anybody then I wouldn't do it anyway. In this case PS CC would also disrupt my system more than I would personally gain from it, and at a higher running cost. The examples that I provided should allow folks to print for themselves and see if the differences are worth the effort for them without having to invest yet.

Quote
Lastly, do you think the apps differ in how they handle B+W images compared to colour ones? What would be the best app to resize a B+W photo by 137%?

If the images are monochrome RGB (R=G=B), then there should be no issues. I'm not sure that a single channel Grayscale image can be handled by all applications/plugins out there, but it's easy enough in Photoshop to change the mode to RGB.

A change to 137% is not that much (even a change of 137% to 237% isn't), and most algorithms will do a decent job. Good output sharpening at that final size will make more of a difference. I'm a fan of Topaz Detail for that (and more, like opening up the shadows and boosting detail there for matte media), although I might also use FocusMagic for the output sharpening only. It depends on the image content.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

rogerxnz

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 527
    • Hayman Lawyers
Re: image up-sizing
« Reply #24 on: August 15, 2016, 05:48:59 am »

Thank you, Bart, for your reply.

I said earlier that I could not find any recent reviews of the various resizing products but I just found one in a local (NZ) magazine by a local expert.

I have attached it so you can read his verdict even if you cannot judge the image quality for yourselves.
Roger
Logged
Roger Hayman
Wellington, New Zealand

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8914
Re: image up-sizing
« Reply #25 on: August 15, 2016, 08:21:45 am »

Thank you, Bart, for your reply.

I said earlier that I could not find any recent reviews of the various resizing products but I just found one in a local (NZ) magazine by a local expert.

I have attached it so you can read his verdict even if you cannot judge the image quality for yourselves.
Roger

Thanks Roger,

He comes to similar conclusions, for his 300% upsampling test. I consider 300% approximately the tipping point from where it gets easier to see the better output quality as one goes larger. Below the 300% there is a difference, but depending on subject matter it may be less obvious.

Apparently, the Photoshop 'Preserve details' option does help a bit on edges, so that could be useful for modest magnifications. I do not know if it introduces any artifacts, more samples would be needed to test that.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==
Pages: 1 [2]   Go Up