Ad
Ad
Ad
Pages: « 1 2 [3] 4 »   Bottom of Page
Print
Author Topic: Post processing IQ180 files for resolution/camera test  (Read 10938 times)
theguywitha645d
Sr. Member
****
Offline Offline

Posts: 970


« Reply #40 on: October 27, 2011, 09:20:49 PM »
ReplyReply

Here is a crazy idea. Print the both images at say 20x30 inches and then scan the same area of the prints. That way each image will be normalized to a real condition and you do not need to up rez or down rez the files. That way you can have a direct comparison in an actual condition these files would be viewed without the need to touch the file data.

Jack is right though. Whoever believes that they have lost will point to an insignificant detail to "prove" the test is "false." But it would be nice to see.
Logged
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7908


WWW
« Reply #41 on: October 27, 2011, 09:48:03 PM »
ReplyReply

Hi,

There are two advantages of scaling down.

1) Scaling moves around pixels, so if we scale one image but not the other the image that is not scaled would be at advantage.

2) Scaling up the digital image 5X linear will introduce artifacts so the image will be ugly

The way I used to think is: Let's assume we make a print of size so and so at that PPI and rescale the images to achieve that PPI. Differences on screen will be much larger than on print.

Using Imatest on the slanted edge is in my view a good idea. One problem with Imatest is that it is sensitive to sharpening, and sharpening is an essential part of the digital workflow.

Best regards
Erik

Hi Bart - we've got an Imatest Master license and have shot a slanted edge target as well as a couple of general resolution 'trumpets' :-)

Out of interest, I'm not sure what downsizing the higher resolution result to the lower resolution result is supposed to acheive - it sounds like throwing away information? I can understand that you end up comparing contrast at the highest MTF of the lower resolution result - is this all?
Logged

Fine_Art
Sr. Member
****
Offline Offline

Posts: 1152


« Reply #42 on: October 27, 2011, 10:08:41 PM »
ReplyReply

Here is a crazy idea. Print the both images at say 20x30 inches and then scan the same area of the prints. That way each image will be normalized to a real condition and you do not need to up rez or down rez the files. That way you can have a direct comparison in an actual condition these files would be viewed without the need to touch the file data.


I like that.


There is no point in scaling just to degrade the pixels. Given the formats a large print is required. Nobody would go through the trouble or the expense of these systems for a web shot. Like Schewe said, the quality of the print is what matters. If a 4000dpi scan gives you enough pixels that you don't need to upres then so be it, that is why 8x10 was chosen. Otherwise 4x5 or less would have been used.

These systems would be used for a gallery print. Lets see a comparison of a gallery print resolution.

Logged
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7908


WWW
« Reply #43 on: October 27, 2011, 11:14:54 PM »
ReplyReply

Hi,

You can do an excellent 30x40" print from CCD scanned Velvia 67, or a 40 MP digital back. You need to print larger than that for 8x10" or 80MP to really come into play.

I have done something similar with images downloaded from Imaging Resource, here: http://echophoto.dnsalias.net/ekr/images/Pentax645D/A0_print_center.jpg

These prints correspond to A0 (ca 33.1x46.8"), the crops are scanned at 300PPI. Left D3X, LR "scenic scarpening" preset), mid D3X (LR, deconvolution), right Pentax 645D (LR, scenic).

Article is here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/51-a-closer-look-at-pentax-645d-image-quality

I'm fully aware that the things we are discussing now are in a different division than the Pentax or the Nikon, but physics and math still apply, and pixels are pixels.

Best regards
Erik

I like that.


There is no point in scaling just to degrade the pixels. Given the formats a large print is required. Nobody would go through the trouble or the expense of these systems for a web shot. Like Schewe said, the quality of the print is what matters. If a 4000dpi scan gives you enough pixels that you don't need to upres then so be it, that is why 8x10 was chosen. Otherwise 4x5 or less would have been used.

These systems would be used for a gallery print. Lets see a comparison of a gallery print resolution.


« Last Edit: October 27, 2011, 11:30:47 PM by ErikKaffehr » Logged

Fine_Art
Sr. Member
****
Offline Offline

Posts: 1152


« Reply #44 on: October 27, 2011, 11:50:30 PM »
ReplyReply

Print DPI should be 600 for cannon type or 720 epson, whatever an even scaling of the printer's default optimum.

Pixels be damned, let's not lose the plot. This is not about jumping through hoops for a test. This is about tools for impressive gallery prints. Both methods should be done with that in mind. Squeeze as much as possible from the method.
Logged
Stefan.Steib
Sr. Member
****
Offline Offline

Posts: 421



WWW
« Reply #45 on: October 28, 2011, 05:58:16 AM »
ReplyReply

Tim

what Jack probably means is something that is called systemic noise in Chaos theory.
The more instances you need to get to a comparison of 2 values the more you will induce errors and variations caused by these additional steps.
A system (left alone 2 systems !) cannot be calibrated by some normalization as the measurement device itself induces new errors.
In this case there are 2 totally differing media which cannot be compared by numbers in their primary output (File against slide).
To get these both informations into the same system of comparable data you will definitely induce additional variation - this is "per definitionem".
Thus the Test is only "valid" if everybody will agree about the method, which as you can already see is "difficult".........

I agree that it is an interesting subject, but the approach is probably only possible if judged on a final product (maybe Print).
But this does not necessarily tell anything about the original material or its content of information which is the title here.
Maybe you should rename the task in : "What produces better prints at the actual status of technology Analogue or digital ?"
This would be a definition that is much clearer and can be judged in a definite and objective way.

Greetings from Munich
Stefan
Logged

Because Photography is more than Technology and "as we have done this all the time"
www.hartblei.de     www.hcam.de    www.spectralize.com
hjulenissen
Sr. Member
****
Online Online

Posts: 1712


« Reply #46 on: October 28, 2011, 06:04:52 AM »
ReplyReply

There are IMHO 2 aspects to a good film scan. The first is to extract all resolution, and second to minimize the distraction from graininess, and there is a trade-off between those parameters. In practice it means having to scan the film with something like 6000+ PPI by a competent operator who can balance the scanning aperture size with the grain/dye cloud structure. Smaller scans will perhaps not extract all detail, and increase the grain-aliasing effect.
I dont do scans. But why would you want to do grain-suppression in the scanning process? Why not scan the image as accurately as possible (grain and all), then do grain removal in software?
Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3876


« Reply #47 on: October 28, 2011, 07:13:56 AM »
ReplyReply

I dont do scans. But why would you want to do grain-suppression in the scanning process? Why not scan the image as accurately as possible (grain and all), then do grain removal in software?

Hi h,

It's not something like noise reduction, which we can additionally apply to the resulting file with software. It's more about not 'enhancing' the grain-structure, or slightly reducing the micro contrast at the grain/dye cloud level. It takes many grains/dye clouds in 3D to form visible granularity. A drumscanner has (on top of the high resolution which also reduces grain-aliasing) the added benefit of being able to vary the scanning aperture, which can be tuned to optimize detail contrast versus grain/dye cloud contrast.

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

Posts: 1712


« Reply #48 on: October 28, 2011, 07:48:01 AM »
ReplyReply

Hi h,

It's not something like noise reduction, which we can additionally apply to the resulting file with software. It's more about not 'enhancing' the grain-structure, or slightly reducing the micro contrast at the grain/dye cloud level. It takes many grains/dye clouds in 3D to form visible granularity. A drumscanner has (on top of the high resolution which also reduces grain-aliasing) the added benefit of being able to vary the scanning aperture, which can be tuned to optimize detail contrast versus grain/dye cloud contrast.

Cheers,
Bart
But if you could obtain a Nyquistian filtered, sharp scanning at _any_ dpi, would you still prefer to filter out grain in the scanning, or doing it with some software?

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

Posts: 3876


« Reply #49 on: October 28, 2011, 07:51:30 AM »
ReplyReply

Hi Bart - we've got an Imatest Master license and have shot a slanted edge target as well as a couple of general resolution 'trumpets' :-)

Hi Tim, there you go. I knew you guys had a grip on what you are doing ...

Quote
Out of interest, I'm not sure what downsizing the higher resolution result to the lower resolution result is supposed to acheive - it sounds like throwing away information? I can understand that you end up comparing contrast at the highest MTF of the lower resolution result - is this all?

Indeed, comparing 2 MTF curves at 2 different points is one. The outcome is not that easy to predict due to the differences in shape of the curves, especially due to the size difference of the 'sensors'.

I was also thinking about the psychological point of view of a MFDB shooter, who is accustomed to his/her reference, but less so to an upsampled version (only to match a deliberately oversampled filmscan). So by downsampling the filmscan, the familiar reference remains constant, and the unfamiliar rendering can get changed to match the size. A similar reasoning applies to those who are accustomed to large format oversampled scans, they will feel less comfortable with a significantly downsampled version (with risk of enhancing graininess) so the original scan should be the unchanged reference.

As for other suggestions along the line of printing to an identical size and scanning the result, it has it's use, but also adds another variable to the mix which might muddy the waters. Not everybody is equally good at producing large format output, and where e.g. a MFDB might benefit from adding noise, a filmscan might benefit from removing some graininess. It will be very difficult to satisfy all takers. Besides, what is large format for some, is a tad on the small side for others, so what would the output size of choice have to be?

Hence my suggestion for offering both up- and downsampling, and then people can do their own output test, based on their skills and available output modalities from that base material.

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

Posts: 3876


« Reply #50 on: October 28, 2011, 07:55:07 AM »
ReplyReply

But if you could obtain a Nyquistian filtered, sharp scanning at _any_ dpi, would you still prefer to filter out grain in the scanning, or doing it with some software?

Probably at the scanning stage, although it depends on the grain/dye cloud structure. Software noise reduction is good at filtering out periodic noise of a few fixed spatial frequencies, not random aggregates of different size and color.

Cheers,
Bart
« Last Edit: October 28, 2011, 08:22:09 AM by BartvanderWolf » Logged
marcmccalmont
Sr. Member
****
Offline Offline

Posts: 1729



« Reply #51 on: October 28, 2011, 08:24:27 AM »
ReplyReply

Yes I agree that a large print is necessary
Marc
Logged

Marc McCalmont
buckshot
Full Member
***
Offline Offline

Posts: 187


« Reply #52 on: October 28, 2011, 08:37:38 AM »
ReplyReply

Tim - look forward to the test. I'm not a pixel peeper per se, but these sort of things always seem to appeal to the technical voyeur in me. Plus, it's fun to see how hot around the collar some folks get at the thought their $40k digital back is about to get trumped by a piece of gelatin covered polyester.

One thing its worth thinking about is just the general 'look' of the images from an aesthetic point of view (subjective? hell, yes). Just step back from the computer and look at the printed images. To me the beauty of 8x10 and up is that certain 'look' of the images - dictated by the large size of the film, DoF, the glass used etc. etc. Medium format digital is just that - medium format - locked in a world of 6 x 4.5 cm for the foreseeable future.
Logged
hjulenissen
Sr. Member
****
Online Online

Posts: 1712


« Reply #53 on: October 28, 2011, 11:46:43 AM »
ReplyReply

Probably at the scanning stage, although it depends on the grain/dye cloud structure. Software noise reduction is good at filtering out periodic noise of a few fixed spatial frequencies, not random aggregates of different size and color.

Cheers,
Bart
What triggers my curiosity is: what kind of linear/nonlinear desirable operation is it that a scanner does when its resolution is decreased that me or you cannot emulate in an Excel spreadsheet that can be written in an hour?

Of course, there are practical considerations. Scanning at 10000 dpi will take longer time and demand more intermediate storage than 2000 dpi, and if the image contains practically no usable information in those spatial frequencies, it can be hard to defend. Same as taking pictures in raw does not add that much information to shooting in jpeg or sRaw/mRaw, but it does add flexibility in processing. If one is to do home-brew image processing using Excel it is going to be slow.

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

Posts: 3876


« Reply #54 on: October 28, 2011, 12:16:14 PM »
ReplyReply

What triggers my curiosity is: what kind of linear/nonlinear desirable operation is it that a scanner does when its resolution is decreased that me or you cannot emulate in an Excel spreadsheet that can be written in an hour?

Depends on one's Excel skills, but aliasing and sampling with an aperture is similar to what happens in a CCD or CMOS photovoltaic sensor (which has fixed apertures roughly the size of a sensel for each sensel). The main difference is that a drumscanner can vary the scanning aperture (and therefore samples the amount of a point sample all the way to an aperture driven moving averaging of larger areas, even overlapping) and the sampling position is controlled in an almost analog fashion, the film rotates in front of the analog photomultiplier.

Quote
Of course, there are practical considerations. Scanning at 10000 dpi will take longer time and demand more intermediate storage than 2000 dpi, and if the image contains practically no usable information in those spatial frequencies, it can be hard to defend.

Oversampling a signal does have benefits for signal reconstruction, especially when the output needs to be blown up to a large size, but indeed file size is one of the drawbacks. Everything that doesn't need to be invented/interpolated for reconstruction, but is sampled directly from the original signal, is likely to be more accurate (within the limitations of the sensor accuracy, in this case film). Finding out whether the cost (time, storage, equipment, labor) is balanced with the benefits, is part of the exercise.

Here's a nice document to ponder about, and it also mentions scanning through an aperture, and the aperture's size and shape having an influence on the recorded signal.

But then, this thread is intended to be more about the presentation of a resolution/camera test, than discussing the benefits/drawbacks of one of the components.

Cheers,
Bart
« Last Edit: October 28, 2011, 12:32:12 PM by BartvanderWolf » Logged
hjulenissen
Sr. Member
****
Online Online

Posts: 1712


« Reply #55 on: October 28, 2011, 12:42:21 PM »
ReplyReply

Here's a nice document to ponder about, and it also mentions scanning through an aperture, and the aperture's size and shape having an influence on the recorded signal.
A long document, where the issue is discussed in two sentences. From what I could read there, it would seem that the behaviour of a scanner operating at an intermediate resolution can be precisely emulated by using a scanner at very high resolution, and then doing a simple 2-d gaussian filtered downsampler.

Let me know if I am hijacking this thread. I just thougth that optimally scanning/processing film would be relevant.

-h
Logged
theguywitha645d
Sr. Member
****
Offline Offline

Posts: 970


« Reply #56 on: October 28, 2011, 12:43:08 PM »
ReplyReply

You can actually print on 8x10 paper and simply scale the images to any size you want. The paper size is not a limiting factor to a print test--you don't need to print the whole image, just a section of it.
Logged
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7908


WWW
« Reply #57 on: October 29, 2011, 12:21:58 AM »
ReplyReply

Hi,

I don't think you are hijacking the thread. But I don't feel that this thread is about emulating scanning at lower resolution.

I have recently doing some comparisons between 6096 PPI scans and lower res CCD scans. One of the observations I have made is that CCD scans (say 3200 PPI) have more grain than the 6096 PPI drum scan. Also, the grain in the drum scan is quite nice. Downscaling and upscaling to say 2000 PPI and back will increase grain and make it more ugly.

One of the interesting observations was that there was a lot of grain in the original (Markus Zuber) test. The visibility of grain was interpreted as indication that all information has been resolved.

In my view, Bart makes a very good point. Digital photographers are used to see images with very well defined edges and essentially no noise. Hi res scans tend to be quite soft. Blowing up an MFDB image from 8000 pixels to 40000 pixels will introduce a lot of artificial detail.

The proposal to make prints is actually interesting, and crops can be printed and rescanned. This has the advantage that the whole processing pipeline is involved, including output sharpening, rescaling in printer driver, dithering. By choosing a correct scanning resolution we can even emulate eyesight.

My experience is that differences will be smaller on scanned prints than on screen at actual pixels.

This image is "actual pixels" view of detail on a comparison I made on 2003 vs 2010 version of LR:s processing pipeline:
http://echophoto.dnsalias.net/ekr/images/PVCompare/PV_compare_2003_vs_2010_sharpening.jpg

And this one is scanned at 300PPI from a 44x66 cm print:
http://echophoto.dnsalias.net/ekr/images/PVCompare/PV_2003_vs_2010_scan.jpg

Best regards
Erik


A long document, where the issue is discussed in two sentences. From what I could read there, it would seem that the behaviour of a scanner operating at an intermediate resolution can be precisely emulated by using a scanner at very high resolution, and then doing a simple 2-d gaussian filtered downsampler.

Let me know if I am hijacking this thread. I just thougth that optimally scanning/processing film would be relevant.

-h
Logged

harlemshooter
Guest
« Reply #58 on: October 29, 2011, 03:11:21 AM »
ReplyReply

A few comments:

1. Only artists are using large (and ultra) film formats in this day and age - and they aren't doing it to get maximal sharpness or to peep at crops. They are doing it for the artifacts introduced by the process (shallow depth of field, movements, etc) and for psychological or philosophical aspects of the slowed process itself. And most are not scanning the film to print other than for VERY large prints (try 8x10 feet with the IQ 180). Ain't gonna deliver, friend. Ask Thomas Struth. LOOK AT THE PRINT NOT THE 100% CROP if you want to get anything worthwhile whatsoever out of this so called "comparison".

2. Downsampling or upsampling either the IQ or film file is a good way to introduce undesirable junk. Why not just compare prints on actual paper at a size of 40x50 inches or larger? Are not LARGE PRINTS the entire point of high resolution image capture?

3. Look at a contact print from 8x10 (or larger) film (or a 2x optical enlargement). You can't obtain that smoothness digitally, never will. It is a result of millions of unique grain clumps scattered unpredictably within the depth of the emulsion.

4. This is a test for the pixel peepers of the universe, period.

5. This test is the equivalent of trying to use a banana peel as a baseball mitt. Films remarkable positive attributes mostly dissipate with digitization - even in huge prints in my opinion, although Jeff Wall (massive prints comprised of many stitched 4x5 transparencies) will disagree with me here: http://www.moma.org/interactives/exhibitions/2007/jeffwall/

Logged
torger
Sr. Member
****
Offline Offline

Posts: 1590


« Reply #59 on: October 30, 2011, 08:07:14 AM »
ReplyReply

What is best will depend on the intended application.

I think it is important that a print looks good at *any* viewing distance, even if your nose is touching the print. It is however less important that it looks *sharp* at any viewing distance, that is it may not be important that the print outresolves the resolution of the eye.

Analog prints up close is a bit like paintings. When you come close you will on a painting see individual brush strokes, so you see it is a painting and although you don't see valid image detail it is a nice look. For an analog print you see grain so you see it is a photograph made on film, again you may not see much valid detail but the look is pleasing.

A digital image where you see pixelation, jaggies and aliasing when you come close is *not* nice. There's no romance to digital artifacts. Say if your digital file lets you keep the print resolution at 200 ppi or higher there will (generally) be no problem. But say if you will make a *huge* print so you get down to 10 ppi or so it will become more important that the file enlarges well than that it has the highest resolution.

Film enlarges well to any size. With digital, it is more complex. I have no experience from making huge prints, but I have made some experiments which indicate that if you want to make huge prints you should not have a super-sharp image at pixel peep level, that is AA filter or a bit of diffraction or outresolving the lens a bit is good. If you can see jaggies on pixel peep level you will get problems when enlarging. I've tested enlargement software but those spline-based scaling etc does not give a pleasing look, just artifical. The only thing that seems to enlarge well to huge sizes is a fairly soft original file (no jaggies) which is scaled up with simple scaling algorithms such as bicubic-soft.

I could imagine that IQ180 (lacks AA filter) with a very sharp lens will not look pleasing up close when enlarged to low ppi levels, and large format film would be much nicer, even if IQ180 might contain more actual detail. However if you don't need to print lower than say 200 ppi I think IQ180 will fare very well. I could also imagine that if you prepare the digital file well for huge enlargement you can make it look good up close, but you will have to prepare it differently (i e make sure it looks soft up close rather than maximizing detail for "normal" viewing distances so it becomes filled with digital artifacts up close) than you would for more moderate upsizing.
Logged
Pages: « 1 2 [3] 4 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad