A closer look at two images from the S2 and the D3X

[ErikKaffehr]

Hi,

Updated 2010-07-22: Reprocessed images and sharpening details now given in the article.

I just published a short article where I look at two images from Lloyd Chambers DAP site.

http://echophoto.dnsalias.net/ekr/index.ph...a-s2-raw-images

This is a first release of the article I'm planning a few updates in the coming days.


Best regards
Erik

[John R Smith]

Well, thanks for that link and the article, Erik. I may be missing something here, but I find that a very odd choice of subject for a test shot. I suppose it is sort of technically demanding in some ways, but I can't really tell if I'm looking at a digital artefact or just the actual appearance of the mosaic. And I don't even like the look of the damn thing, let alone the photograph of it. I don't find it at all useful. If I was doing it I would shoot a detailed landscape, a nice architectural shot, and a studio portrait. And I wouldn't be looking for silly little artefacts, but compelling pictorial quality.

Not your fault, of course, you didn't choose the subject.

John

[ErikKaffehr]

Hi,

I actually don't agree with you. It is very difficult to do a proper lens/sensor test. Lloyd has found this subject and he shoots it with all equipment he tests. Also Lloyd is very careful in his testing.

I find this to be an excellent subject for testing.

Best regards
Erik

Well, thanks for that link and the article, Erik. I may be missing something here, but I find that a very odd choice of subject for a test shot. I suppose it is sort of technically demanding in some ways, but I can't really tell if I'm looking at a digital artefact or just the actual appearance of the mosaic. And I don't even like the look of the damn thing, let alone the photograph of it. I don't find it at all useful. If I was doing it I would shoot a detailed landscape, a nice architectural shot, and a studio portrait. And I wouldn't be looking for silly little artefacts, but compelling pictorial quality.

Not your fault, of course, you didn't choose the subject.

John

[bjanes]

Hi,

I just published a short article where I look at two images from Lloyd Chambers DAP site.

http://echophoto.dnsalias.net/ekr/index.ph...a-s2-raw-images

This is a first release of the article I'm planning a few updates in the coming days.

Best regards
Erik

An excellent first attempt, Erik. The D3x has cleaner shadows and good shadow detail and likely better DR as judged by both the photographic and engineering definition of DR. This is consistent with Digilloyd's conclusions. However, the effects of OLP (optical low pass) filtration with the Nikon are difficult to judge. While Moire is objectionable, it typically does not occur in photographs of most natural subjects and the effects of other forms of aliasing in such images are difficult to judge. Digilloyd seems to think that the Nikon would be better without an OLP, and he states that he has never had much of a problem with Moire with the M9.

The OLP does require extra sharpening and this introduces another variable into the comparison. How much sharpening and what type of sharpening? Some say that deconvolution techniques are particularly good for restoring the sharpness lost by the OLP.

Regards,

Bill

[tho_mas]

The sensor is a Kodak CCD sensor typically used in MF cameras and does not have an Optical Low Pass filter, but untypically for MF it has micro-lenses.

just a note ... not exactly correct. There are some DBs with micro-lenses: P21, P30, H31, H40, eSprit 65 .. and probably more.

since you are at the topic "artifacts": where do these ugly red halos in all of the D3x images come from?

[sbay]

Thanks for posting the images.

How difficult is it to remove moire in software? The demo I saw with Phocus for H4D images (also using kodak 40mp sensor) made it seem very easy, but I did not get a chance to critically review the images.

[tho_mas]

How difficult is it to remove moire in software?

sometimes it's easy and fast (the S2 color artifacts here are extremely easy to correct), sometimes it really takes time. Sometimes it's impossible to be removed completely.
This one took me 2 minutes (note only the color artifacts are removed while the moire structure is still there... though quite okay here | too, this was made with the available JPG...):
[attachment=23249:moire_cc.jpg]

[ErikKaffehr]

Hi,

I lost my sharpening settings, that the reason they are not given. I'll redo the images in a couple of days and include the settings.

Best regards
Erik

An excellent first attempt, Erik. The D3x has cleaner shadows and good shadow detail and likely better DR as judged by both the photographic and engineering definition of DR. This is consistent with Digilloyd's conclusions. However, the effects of OLP (optical low pass) filtration with the Nikon are difficult to judge. While Moire is objectionable, it typically does not occur in photographs of most natural subjects and the effects of other forms of aliasing in such images are difficult to judge. Digilloyd seems to think that the Nikon would be better without an OLP, and he states that he has never had much of a problem with Moire with the M9.

The OLP does require extra sharpening and this introduces another variable into the comparison. How much sharpening and what type of sharpening? Some say that deconvolution techniques are particularly good for restoring the sharpness lost by the OLP.

Regards,

Bill

[ErikKaffehr]

Hi,

I don't know. Some of that may be lateral chromatic aberration, which I have not tried to correct.

Regarding microlenses, I'm aware of some MFDBs having microlenses but I don't think they are very typical. To my understanding microlenses don't work well with lens shifts, and would reduce the flexibility of the back when used on view cameras.

Best regards
Erik

just a note ... not exactly correct. There are some DBs with micro-lenses: P21, P30, H31, H40, eSprit 65 .. and probably more.

since you are at the topic "artifacts": where do these ugly red halos in all of the D3x images come from?

[Guy Mancuso]

FYI the S2 has micro lenses.

[ErikKaffehr]

Guy,

That's what I write: "The camera also does use a mix of MF and DSLR technology. The sensor is a Kodak CCD sensor typically used in MF cameras and does not have an Optical Low Pass filter, but untypically for MF it has micro-lenses."

Thanks for commenting, anyway.

Best regards
Erik

FYI the S2 has micro lenses.

[tho_mas]

I'm aware of some MFDBs having microlenses but I don't think they are very typical. To my understanding microlenses don't work well with lens shifts, and would reduce the flexibility of the back when used on view cameras.

re movements: true. But LF users are probably just a smaller part of the market while the 31MP backs were/are used in particular by fashion shooters...
My impression is the 31MP backs are the most common backs (but I might be wrong)

[Bart_van_der_Wolf]

Hi,

I just published a short article where I look at two images from Lloyd Chambers DAP site.

http://echophoto.dnsalias.net/ekr/index.ph...a-s2-raw-images

This is a first release of the article I'm planning a few updates in the coming days.

Hi Erik,

Thanks for your overview. Just one question, which Raw converter did you use (AFAIK it isn't mentioned on your page, yet)? It will probably be automatically answered when you update the page with the sharpening settings.

Cheers,
Bart

P.S. If you need a sample of a deconvolution sharpened version, I'd be happy to help (I have not subscribed to LLoyd's site, so I'd need a totally unsharpened crop). Since neither all Raw conversions nor upsampling methods provide a good basis for sharpening, we may need to try something else as a basis. Just PM me if I can be of help.

[ErikKaffehr]

Bart,

Thanks a lot for comments!

I use LR3.

Regarding deconvolution I would be interested but I don't feel free to use Lloyd Chambers's files.

Regarding my article I have a problem with it, namely that it doesn't answer any question. Obviously the files need different amount of sharpening and that is highly subjective. Everything I have found was totally expected.

Best regards
Erik

Hi Erik,

Thanks for your overview. Just one question, which Raw converter did you use (AFAIK it isn't mentioned on your page, yet)? It will probably be automatically answered when you update the page with the sharpening settings.

Cheers,
Bart

P.S. If you need a sample of a deconvolution sharpened version, I'd be happy to help (I have not subscribed to LLoyd's site, so I'd need a totally unsharpened crop). Since neither all Raw conversions nor upsampling methods provide a good basis for sharpening, we may need to try something else as a basis. Just PM me if I can be of help.

[Bart_van_der_Wolf]

Bart,

Thanks a lot for comments!

I use LR3.

Regarding deconvolution I would be interested but I don't feel free to use Lloyd Chambers's files.

Regarding my article I have a problem with it, namely that it doesn't answer any question. Obviously the files need different amount of sharpening and that is highly subjective. Everything I have found was totally expected

Hi Erik,

I expected LR3 was used for the entire workflow, but wasn't sure.

Well, you/Lloyd are/is comparing two entirely different capture devices, so it's a bit of an apples and oranges comparison anyway, from shooting experience, to available lens quality, to sensor technology, to Raw conversion quality, to post-processing. There are only a few things that can be done using the same method, but most are different.

Cheers,
Bart

[jduncan]

Hi Erik,

I expected LR3 was used for the entire workflow, but wasn't sure.

Well, you/Lloyd are/is comparing two entirely different capture devices, so it's a bit of an apples and oranges comparison anyway, from shooting experience, to available lens quality, to sensor technology, to Raw conversion quality, to post-processing. There are only a few things that can be done using the same method, but most are different.

Cheers,
Bart

In fact that's a problem, for me, with the review: It is using the best S2 raw converter agains a so, so one for the D3x. That's not saying that the review is invalid. I don't believe it is.  But I guess it has an impact in the results.

[bjanes]

Hi Erik,

I expected LR3 was used for the entire workflow, but wasn't sure.

Well, you/Lloyd are/is comparing two entirely different capture devices, so it's a bit of an apples and oranges comparison anyway, from shooting experience, to available lens quality, to sensor technology, to Raw conversion quality, to post-processing. There are only a few things that can be done using the same method, but most are different.

Cheers,
Bart

Bart,

Since the two cameras are quite different, as you point out, I would submit that proper testing should be to obtain the best possible image from each camera rather than minimizing the variables by using the same test parameters for both cameras. However, this approach adds complexity to the testing, and some will maintain that the optimum parameters were not used (especially if they do not agree with the reported results for their favorite camera). Diglloyd did use his preferred raw converter (Nikon Capture NX2) for the Nikon, but he used the same Smart Sharpen parameters for both cameras. Since the Nikon uses a blur filter, optimum results would require more sharpening, perhaps using a deconvolution technique with an optimum point spread function (PSP). Smart sharpen is a deconvolution algorithm, but the PSP that it uses is relatively limited. The 60mm f/2.8 AFS Nikkor is an excellent lens, but should a US $500 lens be compared to a $5000 lens? The Nikkor is a macro lens that has better performance at distance than its predecessor, but it is presumably optimized for close ups. Photozone.de reported that this lens suffers from a rather disturbing amount of lateral chromatic aberration. CaptureNX2 can correct this automatically, but lens corrections were turned off in Diglloyd's tests. Perhaps Diglloyd should have used the $5,000 Coastal Optics lens on the D3x.

To his credit, Diglloyd did provide raw files, but the debate will go on. In the end, it is not surprising that a camera with a larger sensor and state of the art optics will deliver a better result.

Regards,

Bill

[Bart_van_der_Wolf]

Bart,

Since the two cameras are quite different, as you point out, I would submit that proper testing should be to obtain the best possible image from each camera rather than minimizing the variables by using the same test parameters for both cameras.

Bill,

Fair enough.

However, this approach adds complexity to the testing, and some will maintain that the optimum parameters were not used (especially if they do not agree with the reported results for their favorite camera). Diglloyd did use his preferred raw converter (Nikon Capture NX2) for the Nikon, but he used the same Smart Sharpen parameters for both cameras. Since the Nikon uses a blur filter, optimum results would require more sharpening, perhaps using a deconvolution technique with an optimum point spread function (PSP).

Indeed, hence my offer to at least level the playing field on that aspect.

To his credit, Diglloyd did provide raw files, but the debate will go on. In the end, it is not surprising that a camera with a larger sensor and state of the art optics will deliver a better result.

One would assume, yet the importance of a good raw converter is clearly demonstrated, e.g. by the false color artifacting due to moiré when a proper optical low-pass filter (OLPF) isn't used. The lack of an OLPF may cause serious workflow slow-down in postprocessing for subjects that are likely (though not always predictable) to generate such artifacts.

Cheers,
Bart

[Ray]

Hi Erik,
First let me congratulate you on your dedication to this issue. I can see that you are doing your best to be as objective and as thorough as possible.

However, it's not entirely clear what image sizes you are comparing here. I get the impression that, for the purposes of comparison, you have interpolated both files to a whopping 200MB in 8 bit, and are comparing crops at the pixel level. Is this correct?

The native file size from the D3X in 8 bit mode is about 72MB and that from the S2 about 111MB.

The resolution of my 17" laptop is 1440x900 pixels. The screen is actually 330mm wide. That works out at 4.36 pixels per mm (1440/330) , or 109 ppi.

In order for my laptop to display the entire interpolated image from either the D3X or S2, my laptop screen, at its current resolution of 109 ppi, would need to be 2.3 metres x 1.64 metres, or 7' 6" x 5" 4".

When I examine your small crops of both images on my laptop, I do see fairly subtle differences in resolution. But I've seen greater differences when comparing different models of lenses, and sometimes even when comparing different copies of the same model of lens.

If such resolution differences that you've demonstrated were apparent in an A3+ size print, or even an A2 size print, or even a 24" x 36" print (of the entire image), then I might consider such differences as being significant.

But what you seem to be demonstrating are differences that would only be apparent in a 64" x 90" print, if one were to examine the detail in such prints from the same distance that one peers at one's computer monitor, ie. about 600mm or a couple of feet.

Assuming one were in a position to compare two 64"x90" prints on the wall, by the time you had walked from one print to the other, you would have forgotten the precise nature of those subtle resolution improvements in the S2 print. For all practical purposes the two prints would appear identical from a viewing distance that was sufficient to enable the viewing of both prints next to each other.

Do not apply for a job as a salesman for Leica   .

[ErikKaffehr]

Hi Ray,

Yes, indeed a large print 50x70 cm at 360 PPI. So if we look at the print on 100 PPI computer screen it would correspond to 180x252 cm print. I made a print of four of my crops on my Epson SP 3800 in approximately intended size and the differences are visible (but not very). The problem is that we need different amount of sharpening and sharpening is highly subjective.

Looking at "simulated" 50x70 prints is my normal way of comparing  I have done that comparing Sonya Alpha 700 and Sony Alpha 900, scans from MF and so on. I was surprised how well the Nikon keeps up, I must say.

The S2 images are very sharp at native resolution but they needed sharpening after scaling up. The Nikon needed much more. There is the OLP filter on the  Nikon but I guess that the Leica lens has significantly better MTF at Nyquvist than the Nikon lens, too.

Actually, I don't know if the Leica S2 makes that much sense to me. As Diglloyd points out Nikon and Canon will arrive in the same territory soon enough, at least regarding resolution. There may be other factors, like the camera being extremely well made. In my view they could go for a larger size, or a sensor having a smaller pitch. It would make sense from Leica to announce that they will upgrade present generation Leica S2 to next sensor generation when that is becoming available.

I'm not personally interested in moving up in resolution. Normally I print at A2, even if I have a couple of 70x100 cm prints on my wall. One of those prints is from a 10 MPixel Sonya Alpha 100 the other from a scanned Velvia. My guess is that I can print at least A1 from my Sony Alpha 900.

An important issue that Diglloyd points out is that it is hard to fully utilize the resolution potential of the S2 because of issues with focusing. In his view the AF is not selective enough, but does also lack the ultimate accuracy. It's interesting that Mark Dubovoy found AF to be dead on, so users obviously have to different views.

No, I will not apply for sales position at Leica (or for that part Nikon ;-).


Best regards
Erik

Hi Erik,
First let me congratulate you on your dedication to this issue. I can see that you are doing your best to be as objective and as thorough as possible.

However, it's not entirely clear what image sizes you are comparing here. I get the impression that, for the purposes of comparison, you have interpolated both files to a whopping 200MB in 8 bit, and are comparing crops at the pixel level. Is this correct?

The native file size from the D3X in 8 bit mode is about 72MB and that from the S2 about 111MB.

The resolution of my 17" laptop is 1440x900 pixels. The screen is actually 330mm wide. That works out at 4.36 pixels per mm (1440/330) , or 109 ppi.

In order for my laptop to display the entire interpolated image from either the D3X or S2, my laptop screen, at its current resolution of 109 ppi, would need to be 2.3 metres x 1.64 metres, or 7' 6" x 5" 4".

When I examine your small crops of both images on my laptop, I do see fairly subtle differences in resolution. But I've seen greater differences when comparing different models of lenses, and sometimes even when comparing different copies of the same model of lens.

If such resolution differences that you've demonstrated were apparent in an A3+ size print, or even an A2 size print, or even a 24" x 36" print (of the entire image), then I might consider such differences as being significant.

But what you seem to be demonstrating are differences that would only be apparent in a 64" x 90" print, if one were to examine the detail in such prints from the same distance that one peers at one's computer monitor, ie. about 600mm or a couple of feet.

Assuming one were in a position to compare two 64"x90" prints on the wall, by the time you had walked from one print to the other, you would have forgotten the precise nature of those subtle resolution improvements in the S2 print. For all practical purposes the two prints would appear identical from a viewing distance that was sufficient to enable the viewing of both prints next to each other.

Do not apply for a job as a salesman for Leica   .