Medium Format Digital Vs. Large Sensor DSLR

[Guillermo Luijk]

No one says that MFDB don't provide superior IQ. If they didn't, all people spending thousands of euros on them would be stupid. As I said this was a DR test.

[ErikKaffehr]

Hi Anders,

I suspect that Guillermo would gladly analyze any raw image from any MFDB. The problem is that adequate comparison images from DSLRs and MFDBs are very hard to come by.

To begin with we need raw images and we also need some comparison images shot under identical conditions to be able to do any comparison. Also we would preferably have a sample from a modern sensor, like Nikon D3X or Pentax K5 and not from Canon EOS D which has high readout noise.

Getting that kind of comparison images is not easy.

I have checked out four pairs of images (where I found equivalent raw images from MFDB and DSLR)

http://www.luminous-landscape.com/forum/index.php?topic=50977.0 

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

http://echophoto.dnsalias.net/ekr/index.php/photoarticles/38-observations-on-leica-s2-raw-images

http://www.luminous-landscape.com/forum/index.php?topic=50895.msg421886#msg421886

In addition I have also looked at some sample images by Peter Eastway from the Phase One site and two images from Capture Integration.

I did not publish crops from the Phase One images but here is a short description what I saw: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/41-phase-one-images-for-download

By the way, I don't have any bias against MFDBs. I may have some bias against film, based on the testing I have done (involving 6000 PPI drum scans).

On the other hand I have objections to statements that lack physically feasible explanation, unless proper evidence is presented. In light of proper evidence the search feasible explanation starts.

Best regards
Erik

Hmpf... right?... wrong? Lets touch on the last one first, yes DSLRs offer more versatility because they are more general photographic tools, as compared to MFDB who are for LOW ISO or LOWER ISO and OPTIMUM IMAGE QUALITY at decent light. DSLR and MFDB are different TOOLS for different PURPOSES.

Now to touch on the first.... namely "specs of digital backs that would have been produced 5 years ago", we have a very good example test referenced here;


But uhh... is it not that P25+ has same Kodak sensor that the old P25 had, thus that (although it is a + back) it has a sensor that is around TEN YEARS OLD ?

To comment more about that comparison, was the DSLR using f/11 and P25+ f/22 ? Hmpf.

Come on gents, if you are really content and happy with latest DSLRs give it rest and enjoy them and please keep posting your great images instead. Yet for those few who are really looking for a higher image quality for lower ISOs it is worth to check out the 20, 28 and 33MP digital backs or even the newer higher spec backs. Postings of false information is misleading.

 

[fotometria gr]

I posted on another forum a comparision between the RAW files produced by another user shooting the same scene with:

- Nikon D7000
Iso: 100
nºf: 11
obturación: 1''
Objetivo: Nikkor 24-70mm f/2.8 @ 36mm

- Phase One P25+
Cuerpo: Mamiya 645DF
Iso: 50
nºf: 22
obturación: 10''
Objetivo: Phase One 80mm f/2.8



I was only interested in the Dynamic Range test (i.e. noise test). As a side effect the level of detail could also be compared.
But don't look here for any conclusions about colour accuracy, or any other typical subjective perception ('my MFDB provides fantastic tonal gradation' and etc...).

To find out which sensor had more DR I obtained images representative of the RAW files so that no trick under the hood performed by the RAW developer could fool the results. This was achieved by a RAW development using DCRAW without any white balance (that's why the images look greenish), and no output colour profile conversion (that is way individual RGB clipping becomes representative of the individual RAW channels clipping):



Another highlight area, the MFDB was clipped in the G channel while the Nikon RAW data was intact in all three channels:


The conclusions are:

• The Nikon shows less noise in the darkest areas and has no clipped information, while the MFDB is a bit noisier in the shadows and got its highlights clipped in the G channel. The verdict is that the Nikon D7000 has more dynamic range. Note the difference in resolution would play in favour of the MDFB when comparing the same output resolution, because noise would reduce statistically (not much however, 16Mpx vs 22Mpx).
• Despite the noise, the level of detail achieved by the MFDB is higher. Note this depends not only on the sensor, but also on the lens used, aperture and manual focusing performed.

Regards

It's not higher, it's in a different level for Christ sake! When will some people understand that DSLRs ARE "TWEAKED" even at their "Raw" files? Some people are so busy with theory that can't see anymore... THE DSLR IMAGE IS THERE TO SERVE THE "AVERAGE" OBSERVER and that includes Pentax! Regards, Theodoros.

[ErikKaffehr]

Hi Theodoros,

Do you have any proof for that?

BTW, I'm aware of some tweaking in some cases, as this shows up in autocorrelation. My understanding is that Nikon has no bias on their digital data which is hiding half of the noise (as far as I understand). All CMOS cameras probably use correlated double sampling, but I wouldn't exactly call that tweaking.

I would presume that MFDBs also do some things that could be called tweaking, but it may possibly be done in raw conversion and to a lesser extent in the ASIC.

Can you be a bit more specific on the issue?

Best regards
Erik

It's not higher, it's in a different level for Christ sake! When will some people understand that DSLRs ARE "TWEAKED" even at their "Raw" files? Some people are so busy with theory that can't see anymore... THE DSLR IMAGE IS THERE TO SERVE THE "AVERAGE" OBSERVER and that includes Pentax! Regards, Theodoros.

[roskav]

I don't stand a chance against you guys...!  My main point was about usability and speed etc... It's easy to get distracted into the finer points of image quality.  So I'm happy to be corrected about the equivalent dates of sensor designs.  We all tend to look at image quality on calibrated monitors and fine art prints.... but the main way that my work is experienced is via websites and magazines... and I'm just saying that perhaps we care more about image quality than some of our clients do.. and if we are trying to make money at this game then perhaps we should be open minded about how we capture our images.

R

[jeremypayne]

It's not higher, it's in a different level for Christ sake! When will some people understand that DSLRs ARE "TWEAKED" even at their "Raw" files?

Rather than swearing and insulting people ... offer some evidence for once. 

Other posters have offered plenty of evidence showing a simple, linear relationship between 'photonic' input and the resulting RAW data from a variety of different digital cameras.  You have yet to show a single piece of evidence supporting your extraordinary claims.

Time to put up or shut up.

[fotometria gr]

Hi Theodoros,

Do you have any proof for that?

BTW, I'm aware of some tweaking in some cases, as this shows up in autocorrelation. My understanding is that Nikon has no bias on their digital data which is hiding half of the noise (as far as I understand). All CMOS cameras probably use correlated double sampling, but I wouldn't exactly call that tweaking.

I would presume that MFDBs also do some things that could be called tweaking, but it may possibly be done in raw conversion and to a lesser extent in the ASIC.

Can you be a bit more specific on the issue?

Best regards
Erik

No Erik, I can't be more specific than I have been last time we were on a conversation for the same matter, just do the test I proposed then..., take your wedge, scan it, print it on transparency, tune the profile so that it agrees with your monitor at more than 95%, reprint it, put it in front of a backlit (use a fluorescent studio light) light, reshoot it, reprint it, re-do the calibration, put the final one in front of the light and do the test I suggested for linearity! (I don't have to repeat the procedure, do I?) Is that good enough? If it's not "what the hey"! And of course you are right that MF makers have started doing the same...
 Who is the one who can't see that on the images that "Guillermo" posted, the DSLR has its noise reduction active even on the RAWs? Can't you see the softness? JESUS! Regards, Theodoros.
 P.S. That is EXACTLY why DXO DR measures are "wrong for photography" since they have to do with "test noise acceptance" and it's also the reason why ALL BACKS (even the oldest) will beat any DSLR in highlight DR by at least a stop.

[fotometria gr]

I don't stand a chance against you guys...!  My main point was about usability and speed etc... It's easy to get distracted into the finer points of image quality.  So I'm happy to be corrected about the equivalent dates of sensor designs.  We all tend to look at image quality on calibrated monitors and fine art prints.... but the main way that my work is experienced is via websites and magazines... and I'm just saying that perhaps we care more about image quality than some of our clients do.. and if we are trying to make money at this game then perhaps we should be open minded about how we capture our images.

R

+1. Plus better photographers if I should add... Regards, Theodoros.

[jeremypayne]

just do the test I proposed then

Your test involves examining a rendered and printed image.  That is not a testing method for determining the characteristics of a raw file.

Try again.

[eleanorbrown]

Since i find this topic interesting I did a test yesterday between my 60 megapixel P65+ and Leica M9 at 18 megapixels.  The shot was done in my garage of one wall..P65 on a Hassy H2 with hassy prime 80mm at f8 (iso 100) on tripod with timer set.  Leica with 50/ 1.4 mm prime at f8 (iso 160) also on tripod using timer. Raw processed in LR4 beta and both printed with short side at 16" to make 16X20 prints.  THIS IS NOT A SCIENTIFIC TEST...just my personal test to satisfy my curiosity but though I would share.  Phase file printed at 720 dpi and Leica file at 360 dpi on Canson Baryta paper on Epson 7900 printer.  Forget about looking at "viewing distance".  I view at reading distance and my eyesight is corrected through contacts for 20/20.  On first very careful study the prints look very very equal...really.  it is only when I study in bright light that I can begin to see some advantage in the Phase One print but only in a few specific areas like the weave in the blue dog leash...or the edge detail in the far right pair of skis.  Really..this is incredibly close match even at 16X20.  In this print size the Phase One file has a tiny advantage but you have to know what you are looking for and study it close at reading distance.  Both files have deep shadows and the only area where detail is totally blocked (on both files) is inside the boot in the lower left corner.  Highlights are held on all metal surfaces.  Let me be clear:  These are MY visual impressions only, and others looking at the same files and prints may certainly have different views.


Eleanor

[fotometria gr]

Rather than swearing and insulting people ... offer some evidence for once. 



Time to put up or shut up.

Is that up there a "shut up" that refers to me?  By who? A self declared "accidental artist"?
P.S.Promise I will never answer back your stupidity (as I have always done during your last provocations), mind you that its about the 20th quote that you are politely asked not to refer to me again.

[hjulenissen]

that its about the 20th quote that you are politely asked not to refer to me again.

It does not work that way. You don't get to say whatever you want on a place like this and request that no-one critizise your unique opinions.

-h

[jeremypayne]

[more off-topic nonsense]

Personal insults don't win arguments.  Neither does running away.

If you assertions are grounded in fact, you should have no trouble citing references or providing real evidence.  At a minimum, you should at least propose a test that is relevant.

So far all you have done is make outlandish claims, insult people, invoke the christian savior, propose inappropriate tests and then try and change the subject.

Try again.

[fotometria gr]

No Erik, I can't be more specific than I have been last time we were on a conversation for the same matter, just do the test I proposed then..., take your wedge, scan it, print it on transparency, tune the profile so that it agrees with your monitor at more than 95%, reprint it, put it in front of a backlit (use a fluorescent studio light) light, reshoot it, reprint it, re-do the calibration, put the final one in front of the light and do the test I suggested for linearity! (I don't have to repeat the procedure, do I?) Is that good enough? If it's not "what the hey"! And of course you are right that MF makers have started doing the same...
 Who is the one who can't see that on the images that "Guillermo" posted, the DSLR has its noise reduction active even on the RAWs? Can't you see the softness? JESUS! Regards, Theodoros.
 P.S. That is EXACTLY why DXO DR measures are "wrong for photography" since they have to do with "test noise acceptance" and it's also the reason why ALL BACKS (even the oldest) will beat any DSLR in highlight DR by at least a stop.

Erik, "reshoot" and "reprint" is for making sure that the result is relevant with what you see on your screen... ie: to make sure that the profile is as close as possible to the wedge (just to add, its got to be done in "color" although the wedge is B&W). Regards, Theodoros.
 P.S. If you are not sure (or if I am not understood), send me a PM.

[John R Smith]

It seems to me that a "test that is relevant" has just been made by Eleanor, but nobody seems to be taking any notice of it. A nice bit of work, Eleanor, and what strikes me as relevant here is that the quality of the lenses used was probably more of an influence than the sensor, at least at 16x20 ins. Two cracking lenses, and that's why the prints look so good and why they are very hard to tell apart.

So will you be selling the Hassy and the P65 now then, Eleanour?  

John

[eleanorbrown]

Thanks John......Ha!  nooo, won't be setting the Hassy and P65 any time soon :-)... and yes you're right about the lenses...I like high quality lenses...I get spoiled with primes. Eleanor

It seems to me that a "test that is relevant" has just been made by Eleanor, but nobody seems to be taking any notice of it. A nice bit of work, Eleanor, and what strikes me as relevant here is that the quality of the lenses used was probably more of an influence than the sensor, at least at 16x20 ins. Two cracking lenses, and that's why the prints look so good and why they are very hard to tell apart.

So will you be selling the Hassy and the P65 now then, Eleanour?  

John

[hjulenissen]

On first very careful study the prints look very very equal...really.

I was struck by the similarity in the attached jpegs. Even color match seems excellent.

it is only when I study in bright light that I can begin to see some advantage in the Phase One print but only in a few specific areas like the weave in the blue dog leash...or the edge detail in the far right pair of skis.  

Do you think that you could crop out those section and attach them in full resolution?

Thank you for your contribution

-h

[fotometria gr]

It seems to me that a "test that is relevant" has just been made by Eleanor, but nobody seems to be taking any notice of it. A nice bit of work, Eleanor, and what strikes me as relevant here is that the quality of the lenses used was probably more of an influence than the sensor, at least at 16x20 ins. Two cracking lenses, and that's why the prints look so good and why they are very hard to tell apart.

So will you be selling the Hassy and the P65 now then, Eleanour?  

John

Eleanors, test is a trusted and well done one because its based on "photographer trust your eyes and sense", which is the only trusted method for a photographer, ...yet, the problem is that it is a MFDB against "MFDB not DSLR" testing, this is because M9's sensor is really a MFDB cropped... Regards, Theodoros.
 P.S. DSLRs (in the way that market considers them), have a ...long way before they can even "face" the M9, ...not to mention "compete" with it and its not only because of the lenses..., lenses is the least part of the fact (rather mysticism).

[fredjeang]

Tell me if I'm wrong, but the Phase is the left one no?

Because I'm watching those images in my computer at about 1,20m distance in this small size and differences are there clearly.

And according to what I see, the Phase for me is the left image. If I'm wrong no prob, my eyes have been fooled then.

[eleanorbrown]

Yes Phase is on the left, but I had to go back into Lightroom to check as I couldn't tell the difference.  Here are screen shots from Lightroom: left Phase at 100 percent, Phase at 50 percent and Leica at 100 percent.  Really, this is not done scientifically...certainly there are people who could process better than me too but after all the talk on this topic I wanted to see what differences, if any I could see for myself. I could have probably spent more time in Lightroom to get the images to match more carefully so please keep that in mind.   eleanor