Ad
Ad
Ad
Pages: « 1 [2]   Bottom of Page
Print
Author Topic: which enlargement method gives a better picture  (Read 9695 times)
bwana
Full Member
***
Offline Offline

Posts: 183


« Reply #20 on: February 19, 2012, 12:39:15 AM »
ReplyReply

Best thing to do is test this on your own printer because YRMV...


yes. a little like touring the greek islands...much better to do it on your own and explore rather than going on a cruise ship and following the herd...were those pictures last summer 2010? timeless places those villages...
Logged
Schewe
Sr. Member
****
Offline Offline

Posts: 5415


WWW
« Reply #21 on: February 19, 2012, 12:53:20 AM »
ReplyReply

Sry, I was still in a fog about M. Skurski who had complications and was thinking about him.

Yep..we lost Mike first, then Bruce...was a tough time for PixelGenius...

The upside is that we brought Mac Holbet on board for PKS 2. We released PKS2 in early 2011 followed by PK2 in mid 2011.

We did work out a deal with Adobe to include PhotoKit Sharpener into LR and ACR...a tribute to Bruce and the PG team.
Logged
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3412


« Reply #22 on: February 19, 2012, 08:27:16 PM »
ReplyReply

thank you for your lively debate. my research shows that greater enlargement and subsequent greater reduction results in softening of boundaries. so an image scaled up 10 fold with bicubic resampling and then resized to 360 dpi from 72 shows softer edges than an image scaled up 7.5 fold and resized to 270 dpi.

Only if no output sharpening/filtering at the final output size was applied. Also, the subsequent mandatory upsampling from 270 PPI to 360 PPI by the printer driver will not add any detail either.

Quote
Intuitively that also makes sense. But this is only based on what I see on a monitor. When the images are sprayed onto inkjet paper, I cant really see a difference. I think this may be because the epson printer is resampling the 270 dpi image to 360 dpi and thereby softening the edges. or maybe my old eyes just cannot see.

No, it is indeed the printer driver trying to achieve at least 360 PPI, because it's dithering algorithms are optimised for that.

Quote
whether or not an image looks like crap from this kind of extreme rescaling and resizing is not the question. i am trying to determine which rescaling algorithm would produce the least crappy image when you need to upsize your image. And rather than split hairs over subtle differences, I am using an extreme example so I can magnify the effects to come to a clear conclusion.

As I've indicated, a 4x upsampling is only modestly stressing the system. Double that, and we're talking about a challenge.

Quote
Since my original post, I have also been reading about others' experience with epson printers. It seems the most critical variable is to maintain the dpi of the output image as a factor of 360.

Indeed, or 720 PPI with 'Finest Details' checked. When there is enough detail, upsampling to 720 PPI on Epson printers is warranted. Some upsampling methods will add high spatial frequency detail beyond that what the file offers natively. It would be a waste not to use that.
 
Quote
This eliminates the printer from the rescaling/resizing loop and removes another cycle of image degradation. So the subtle difference I was trying to measure in my initial post, is actually overshadowed by something I never took into account-the printer nozzle density which is fixed at a specific dpi. This renders the initial question moot .

Not really, but you won't hear the so-called authorities mentioning it. There are additional gains to be achieved with better software methods. There's 'some' more info to read at:
http://www.luminous-landscape.com/forum/index.php?topic=54798.msg447163#msg447163

Cheers,
Bart
Logged
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3412


« Reply #23 on: February 20, 2012, 10:03:11 AM »
ReplyReply

Here are a couple of examples of what to really expect from upsampling.

I started with a crop of 450x450 pixels from a 1Ds Mark III camera file (Capture One conversion, no sharpening):


Then I've made 2 examples of 400% enlargements, in line with what the OP gave as a theoretical example (they should print as 5 inch square at 360 PPI). Here is one with Photoshop Bicubic Smoother, a quality which according to Jeff "prints out like crap, lots of ringing and artifacts from the upsample". And here is one example of a 400% upsampling with Photozoom Pro (S-Spline Max method).

Both are not sharpened yet, so if you want to print them, you can use your trusted method of sharpening. For viewing them on screen, they are best viewed at approx. 25% zoom for a size that resembles the printed output, but the quality should really be judged from a printed version, properly sharpened for the media used.

And then I've made 2 examples of an 800% enlargement, which would resemble the quality of a 3.17 x 2.11 metres output from a single 1Ds3 file at 360 PPI, or can be directly compared to the above prints when you use 720 PPI ("finest detail" option selected) as output resolution. Some authorities say there is no benefit to printing at 720 PPI when the original file has a native resolution of below 360 PPI. I say, try it yourself, and like me draw your own conclusions.

Here is the Bicubic Smoother version, and here is the Photozoom Pro version. Both are only upsampled, not sharpened yet.

Of course there is no substitute for real pixels, but I'm looking forward to hearing what the findings are ..., does upsampling (sometimes as a last resort) produce crap or what?

Cheers,
Bart
« Last Edit: February 20, 2012, 10:06:20 AM by BartvanderWolf » Logged
Schewe
Sr. Member
****
Offline Offline

Posts: 5415


WWW
« Reply #24 on: February 20, 2012, 11:03:23 AM »
ReplyReply

Of course there is no substitute for real pixels, but I'm looking forward to hearing what the findings are ..., does upsampling (sometimes as a last resort) produce crap or what?

Well, I looked at the images (didn't print them mind you) and they are definitely on the crap side of the scale. The Photozoom looked a tad better but I think S-Spline has a component of sharpening built in, doesn't it?

Look, if you trying to see a license plate for forensic purposes, this stuff is all very interesting. But fuzzy, upsampled low rez images are not something I want to print.
Logged
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3412


« Reply #25 on: February 20, 2012, 11:19:10 AM »
ReplyReply

Well, I looked at the images (didn't print them mind you) and they are definitely on the crap side of the scale.

Hi Jeff,

Well at least you are consistent... Wink

Quote
The Photozoom looked a tad better but I think S-Spline has a component of sharpening built in, doesn't it?

It allows to optionally add USM, but I had it's checkbox switched off. More detailed tests prove that there is indeed no sharpening applied, other than the interpolation method (resembles continuous vector/spline edges, rather than rasterized edges) which does add resolution, not just fuzzy pixels, but edge detail that's smaller than the upsampling percentage. The other software I've tried, loses a bit of resolution because it has to avoid blocking, ringing, and other artifacts.

Cheers,
Bart
Logged
joofa
Sr. Member
****
Offline Offline

Posts: 485



« Reply #26 on: February 20, 2012, 04:18:34 PM »
ReplyReply

Here are a couple of examples of what to really expect from upsampling. ... Of course there is no substitute for real pixels, but I'm looking forward to hearing what the findings are ..., does upsampling (sometimes as a last resort) produce crap or what?

According to JIDM, going from the original crop to 400% enlargement the Photozoom Pro version reduces sharpness by a factor of 0.6, where as the BiCubic Smoother version by a factor of 0.125.

Sincerely,

Joofa
Logged

Joofa
http://www.djjoofa.com
Download Photoshop and After Effects plugins
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3412


« Reply #27 on: February 20, 2012, 05:35:09 PM »
ReplyReply

According to JIDM, going from the original crop to 400% enlargement the Photozoom Pro version reduces sharpness by a factor of 0.6, where as the BiCubic Smoother version by a factor of 0.125.

Hi,

Although I don't know if your metric scales linearly (or another correlation) with resolution or with perception, a difference in favor of the Photozoom Pro version is not a surprise to me. A short while ago I did an analysis of the potential benefit of upscaling from 360 PPI to 720 PPI, and it showed similar findings, only a benefit for Photozoom Pro and not for other more traditional upsampling methods.

Attached is a graphic that shows the MTF curves of perfect 360 PPI and 720 PPI pixel detail, and the upsampled results, in this case from 360 to 720 PPI for Lightroom 3.6 with medium output sharpening, and Photozoom Pro without sharpening. It shows that Lightroom doesn't add resolution, while Photozoom Pro upscales the edge detail almost perfectly, effectively adding fake, but convincing fake, resolution proportional to the upscaling.

Cheers,
Bart
« Last Edit: February 20, 2012, 06:01:48 PM by BartvanderWolf » Logged
joofa
Sr. Member
****
Offline Offline

Posts: 485



« Reply #28 on: February 20, 2012, 11:30:22 PM »
ReplyReply

Although I don't know if your metric scales linearly (or another correlation) with resolution or with perception,

Well, JIDM is the talk of the town, so you better know ....  Grin

Thanks for the interesting MTF plot.

Here is the FFt of the 100% crop:
http://djjoofa.com/data/images/7640_Crop1_100pct_fft.jpg

FFT of the Photozoom Pro 400% enlargement:
http://djjoofa.com/data/images/7640_Crop1_400pct_SSmax_fft.jpg

FFT of the BiCubic Smoother 400%:
http://djjoofa.com/data/images/7640_Crop1_400pct_BiCubSmth_fft.jpg

Here is an interesting thing. I overlaid the FFTs of the 100% crop over the respective 400% enlargements of Photozoom Pro and BiCubic Smoother. Photozoom Pro is shown below. Interestingly, the common portions of the 100% crop and 400% Photozoom Pro enlargement are quite identical, and the spikes in the 100% crop seem to continue in the extended space of the 400% Photozoom Pro.


Below is the 100% crop overlaid BiCubic Smoother 400%. Here the common portion is not as similar, and the extension of spikes is non-existent.


Sincerely,

Joofa
Logged

Joofa
http://www.djjoofa.com
Download Photoshop and After Effects plugins
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3412


« Reply #29 on: February 21, 2012, 10:23:44 AM »
ReplyReply

Here is an interesting thing. I overlaid the FFTs of the 100% crop over the respective 400% enlargements of Photozoom Pro and BiCubic Smoother. Photozoom Pro is shown below. Interestingly, the common portions of the 100% crop and 400% Photozoom Pro enlargement are quite identical, and the spikes in the 100% crop seem to continue in the extended space of the 400% Photozoom Pro.

Hi,

Indeed, it shows that signal is added at higher spatial frequencies, not just more blurry pixels to fill the gaps. Of course this can start to look unnatural when pushed much too far, because it is mostly already high frequency detail that gets enhanced. When we look at uncorrelated noise, then the gain is much lower or non-existing. That does have the effect of suppressing noise, but noise can be reintroduced with the noise control of Photozoom Pro, which reduces the artificial look a bit.

Most of the other resampling methods do not add any resolution to speak of, but do try to keep looking natural without introduction of too many artfacts. I've added some MTF curves of the more successful methods in the chart below. The usual suspects are ImageMagick with it's new "-distort Resize" algorithm with a Mitchell filter in the lead, followed by Qimage's Hybrid SE algorithm and Lightroom 3.6 with Standard output sharpening, and finally Photoshop's Bicubic Smoother. All of them required some additional sharpening to improve the resolution that was lost when resampling, but it was never enough to come even close to Photozoom Pro (which was not sharpened). Lightroom used it's built in Standard output sharpening, and the others were deconvolution sharpened with FocusMagic. To avoid excessive sharpening artifacts with the Bicubic Smoother method, I had to apply a little Low-pass filtering (blur) on the highest spatial frequencies.

Cheers,
Bart
« Last Edit: April 19, 2012, 08:14:41 PM by BartvanderWolf » Logged
joofa
Sr. Member
****
Offline Offline

Posts: 485



« Reply #30 on: February 21, 2012, 10:00:11 PM »
ReplyReply


Thanks for the interesting graph again.

Indeed, it shows that signal is added at higher spatial frequencies, not just more blurry pixels to fill the gaps.

A thing that I have noticed is that it appears Photozoom Pro is "thining" out the edges by defining sharper transitions.

Sincerely,

Joofa
Logged

Joofa
http://www.djjoofa.com
Download Photoshop and After Effects plugins
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3412


« Reply #31 on: February 22, 2012, 06:41:32 AM »
ReplyReply

A thing that I have noticed is that it appears Photozoom Pro is "thining" out the edges by defining sharper transitions.

Hi,

That's correct, which is good news for 600 PPI and 720 PPI scenarios. Those resolutions are beyond normal visual acuity, but well within Vernier acuity, and as such will enhance the perceived sharpness of printed output, even if the native file's resolution is lower than 360 PPI. And the added acuity does not suffer from halos at all so there are only gains to be had.

However, the degree of edge sharpening/thinning can be influenced by reducing the Sharpness slider control, which I do recommend for extreme enlargements. It will help to avoid the mental disconnect beween edge sharpness and surface structure detail. The effect is dependent on the particular spatial frequency as can be seen when using e.g. a "star" target (like this one for 600 PPI or 720 PPI). Of course Photozoom Pro picks up aliasing artifacts (such as stairstepping) that are already in the original image but it also allows to reduce the severeness of those artifacts to a degree with its Artifacts control slider, which is good news for OLPF-less sensors, and/or users of very sharp lenses.

Cheers,
Bart
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2756



« Reply #32 on: February 22, 2012, 08:08:01 AM »
ReplyReply

Because your question indicates a severe lack of understanding and little or no effort on your part? Maybe that's why I answered the way I did. Do some research on your own and get back to us.

The Right Resolution

And The Art Of The Up-Res

Read those and get back to us...

While those articles give much useful information, they do not use more advanced methods of upsampling and sharpening that Bart discusses in his post here in this thread and a previous thread.

The critical question is, if one does a lot of upsamplilng, is it worth the trouble and expense of using these additional methodologies over what is available in the standard Adobe applications?

Regards,

Bill

Logged
hjulenissen
Sr. Member
****
Offline Offline

Posts: 1666


« Reply #33 on: November 13, 2012, 08:10:40 AM »
ReplyReply

Then I've made 2 examples of 400% enlargements, in line with what the OP gave as a theoretical example (they should print as 5 inch square at 360 PPI). Here is one with Photoshop Bicubic Smoother, a quality which according to Jeff "prints out like crap, lots of ringing and artifacts from the upsample". And here is one example of a 400% upsampling with Photozoom Pro (S-Spline Max method).
The photoshop one obviously is poor, also with what appears to be blocking artifacts?

Photozoom pro appear very sharp, but has a "water-colory" quality that I dislike. I actually prefer a more smeared, "linear filtering-like" look.

-h
Logged
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3412


« Reply #34 on: November 13, 2012, 05:41:10 PM »
ReplyReply

The photoshop one obviously is poor, also with what appears to be blocking artifacts?

Photozoom pro appear very sharp, but has a "water-colory" quality that I dislike. I actually prefer a more smeared, "linear filtering-like" look.

Hi,

I agree that the Photozoom Pro upsample looks a bit strange this close, but try printing these (@ 600 or 720 PPI) side by side and the quality difference will become quite obvious. And of course one can reduce the edge sharpening, e.g. for display at lower resolution, and more noise can be added if things need some structure.

Cheers,
Bart
Logged
Pages: « 1 [2]   Top of Page
Print
Jump to:  

Ad
Ad
Ad