Luminous Landscape Forum
Equipment & Techniques => Cameras, Lenses and Shooting gear => Topic started by: ErikKaffehr on April 28, 2012, 08:37:28 am
-
Hi,
Michael Reichmann produced two very good test shots from the Nikon D800 and the Nikon D800E using Bart van der Wolf's test target. I have analyzed these two images using Norman Koren's Imatest program, using two different sharpenings. The two sharpenings used were Lightrooms "Narrow Edges (Scenic)" called "LS" and my own standard sharpening called EKRNES. See the results below.
What I see is that D800 with my own sharpening is pretty close to D800E with "Landscape" sharpening. If we check the MTF50 values without "standard sharpeing" they are
D800 - LS 3465 LW/PH
D800 - EKRNES 4106 LW/PH
D800E - LS 4025 LW/PH
So the D800E images with Lightrooms landscape sharpening preset is very close to the D800 image with my own sharpening preset. I also enclose screen dumps of the star image. The text in the lower left corner may give a realistic view of microcontrast. The third image is the same as the second one, but converted to monochrome.
More to come...
Best regards
Erik
-
So, how was the D800E with your own sharpening preset? What you can do to one image, you can do to the other...
-
Hi,
Here are the figures:
D800E (left) : 4666 LW/PH at 50%MTF
D800 (right): 4130 LW/PH at 50%MTF
But I don't see this a sharpening race. Sooner or later the image will be oversharpened and the sharpening will induce artifacts of it's own. So there is a decision needed on the amount of sharpening.
An OLP filtered image needs more sharpening, this is a well known and widely accepted fact. How much is to much may be in the eye of the beholder.
Best regards
Erik
So, how was the D800E with your own sharpening preset? What you can do to one image, you can do to the other...
-
There is something I don't really understand about how the Imatest presents it's values. How can there be an MTF 18 with massively high values? There are about 4900 pixels vertically on the chip so anything over that is meaningless. Or what does it mean? Please explain.
When I draw a 100 pixel circle around the centre it looks pretty good. The D800 really is presenting good data at the pixel level. With 36MP I would not quibble about trying to make the circle down to nyquist.
Now the question is, is there a presentation medium that can keep up? A screen has higher contrast without the resolution. A print has resolution without the massive DR, color gamut. Why pay more for MFDB? When can I get my $2k Sony camera version of this chip?
This is a wow camera.
-
Here are the figures:
D800E (left) : 4666 LW/PH at 50%MTF
D800 (right): 4130 LW/PH at 50%MTF
But I don't see this a sharpening race.
True, it's not a sharpening race, but by omitting the results of the D800E sharpened by your method seemed to imply that the D800E is only as good as the D800 with optimal sharpening applied.
Thanks for posting these results!
-
The difference between 4666 and 4130 is certainly non-trivial. At a bit over 10% it closely tracks what I am seeing in real-world comparisons.
The difference is clearly visible at 100% onscreen, and in large prints. But, the caveat is one which I've made before...only if the best lenses and the best technique are used. For normal hand-held use the D800E's resolution advantage over the D800 is mostly lost. For this reason I have set up my system so that the D800 is ready to be used for mobile shooting while the D800E will be treated like a medium format camera and used in that manner.
Michael
-
Michael as an owner of the IQ180 and the D800e I would be interested in hearing your comparison of the two cameras with the phase file cropped down to 36 mp size. I have the P65+ and have the D800 e on order so would be interested to hear your impressions of IQ of the Phase and Nikon if time permits. Eleanor
The difference between 4666 and 4130 is certainly non-trivial. At a bit over 10% it closely tracks what I am seeing in real-world comparisons.
The difference is clearly visible at 100% onscreen, and in large prints. But, the caveat is one which I've made before...only if the best lenses and the best technique are used. For normal hand-held use the D800E's resolution advantage over the D800 is mostly lost. For this reason I have set up my system so that the D800 is ready to be used for mobile shooting while the D800E will be treated like a medium format camera and used in that manner.
Michael
-
OK!
So I posted both results. On the other hand my personal opinion is that my standard sharpening would be a bit excessive on the D800E.
Best regards
Erik
True, it's not a sharpening race, but by omitting the results of the D800E sharpened by your method seemed to imply that the D800E is only as good as the D800 with optimal sharpening applied.
Thanks for posting these results!
-
Hi,
I hope you will enjoy your D800E. I also hope that you share your experience.
Best regards
Erik
Michael as an owner of the IQ180 and the D800e I would be interested in hearing your comparison of the two cameras with the phase file cropped down to 36 mp size. I have the P65+ and have the D800 e on order so would be interested to hear your impressions of IQ of the Phase and Nikon if time permits. Eleanor
-
So, did the RAW files become available yet?
-
There is something I don't really understand about how the Imatest presents it's values. How can there be an MTF 18 with massively high values? There are about 4900 pixels vertically on the chip so anything over that is meaningless. Or what does it mean? Please explain.
When I draw a 100 pixel circle around the centre it looks pretty good. The D800 really is presenting good data at the pixel level. With 36MP I would not quibble about trying to make the circle down to nyquist.
Spuriously high "resolution" above the theoretical limit imposed by the Nyquist theorem is a well known artifact of measurements using the ISO 12233 method using a slanted edge target to measure the SFR (spatial frequency response) around the edge of the target. Sharpening detects edges and enhances them and the measurement is also sensitive to noise and can be affected by noise reduction. Such a processed image can have a very sharp edge but little detail (texture) in other areas. This is explained in an article (http://www.dxo.com/var/dxo/storage/fckeditor/File/embedded/2009_EI_TextureSharpness.pdfhttp://) by DXO. Norman Koren has a table (MTF measurement matrix (http://www.imatest.com/docs/sharpness/)) with links to the various methods of measuring MTF and the targets that are used. He is implementing methods using the Siemens star and a random scale invariant target (similar in principle to the dead leaves method), but I do not know if these are available in the Studio version of Imatest, which is what most of us enthusiasts use. The Professional version is very expensive and is targeted to large corporations.
Bart's sinusoidal Siemen's star is an elegant method, but it measures resolution at low MTF (near the Rayleigh limit), similar to what one obtains with the USAF 1951 target. MTF at lower frequencies is often more important for perceived image quality. When measuring MTF one must state the frequency as well as the contrast as is typically done with MTF charts published by lens manufacturers, which typically show contrast at 10 lp/mm and 30 or 40 lp/mm. The 10 lp/mm is often regarded as the more important figure. Measurement of image sharpness is difficult, and one must correlate the results with actual test images.
Regards,
Bill
-
But I don't see this a sharpening race. Sooner or later the image will be oversharpened and the sharpening will induce artifacts of it's own. So there is a decision needed on the amount of sharpening.
Hi Erik,
That's right. It's easy enough to boost the numbers, but what really matters is how the image looks. From the point of how much is enough, I always look at the Edge Spread function (ESF). When it shows a halo (overshoot) that exceeds half a pixelwidth on both sides, then the capture sharpening is going too far. Were creating artifacts in the pixels next to the ones we want to address, not sharpness. Another indicator is the 10-90 percent metric on the ESF (called Edge profile in the Imatest charts). When it goes below 1.0 pixels, we're usually pushing things too far (edges shouldn't be smaller than 1 pixel, that will lead to severe jaggies).
We can use that overshoot to compensate for losses further down the line, e.g. in print due to ink diffusion, but output sharpening is a different subject that has little to do with restoring inherent image sharpness, it's a remedy only to compensate for the losses.
An OLP filtered image needs more sharpening, this is a well known and widely accepted fact. How much is to much may be in the eye of the beholder.
Well, actually one can use the same principle as above, it's just going to require a lower amount of sharpening to reach the same levels. Again, going further is going too far.
Cheers,
Bart
-
Hi,
The way it works we set a certain MTF, like 50% and look for the frequency that yields 50%. Sorry for leaving the 18% figure in place. I intended for diffraction studies. Just care about whatever is left of Nyquist.
Best regards
Erik
Spuriously high "resolution" above the theoretical limit imposed by the Nyquist theorem is a well known artifact of measurements using the ISO 12233 method using a slanted edge target to measure the SFR (spatial frequency response) around the edge of the target. Sharpening detects edges and enhances them and the measurement is also sensitive to noise and can be affected by noise reduction. Such a processed image can have a very sharp edge but little detail (texture) in other areas. This is explained in an article (http://www.dxo.com/var/dxo/storage/fckeditor/File/embedded/2009_EI_TextureSharpness.pdfhttp://) by DXO. Norman Koren has a table (MTF measurement matrix (http://www.imatest.com/docs/sharpness/)) with links to the various methods of measuring MTF and the targets that are used. He is implementing methods using the Siemens star and a random scale invariant target (similar in principle to the dead leaves method), but I do not know if these are available in the Studio version of Imatest, which is what most of us enthusiasts use. The Professional version is very expensive and is targeted to large corporations.
Bart's sinusoidal Siemen's star is an elegant method, but it measures resolution at low MTF (near the Rayleigh limit), similar to what one obtains with the USAF 1951 target. MTF at lower frequencies is often more important for perceived image quality. When measuring MTF one must state the frequency as well as the contrast as is typically done with MTF charts published by lens manufacturers, which typically show contrast at 10 lp/mm and 30 or 40 lp/mm. The 10 lp/mm is often regarded as the more important figure. Measurement of image sharpness is difficult, and one must correlate the results with actual test images.
Regards,
Bill
-
Hi,
I compared the D800E with Landscape Sharpening to the Pentax 645D using one of the Iamging Resource test images, also with LS sharpening.
In my view the Pentax 645D clearly comes out on top, at least regarding LW/PH at 50% MTF. I'd guess the Pentax is much more Moiré prone than the D800E from the MTFs.
Pentax 645: 5348 LW/PH
Nikon D800E: 4035 LW/PH
So my take is that D800E may be mighty, but at least for resolution MF still has a handsome lead. Real world performance may be different, though. I have seen a comparison between Leica S2 and Nikon D800 by Diglloyd where I would give the nod to the D800. But the numbers speak for the Pentax 645D, unless you object to aliasing.
On the other hand, the Pentax 645D is low end in MF, cameras like the IQ180 would have smaller sensor pitch and be less prone to Moiré.
Best regards
Erik
-
Spuriously high "resolution" above the theoretical limit imposed by the Nyquist theorem is a well known artifact of measurements using the ISO 12233 method using a slanted edge target to measure the SFR (spatial frequency response) around the edge of the target.
Hi Bill,
I wouldn't call it an artifact of the slanted edge method used in several ISO standards for resolution testing of digital equipment. That method just allows to detect signals at higher spatial resolution levels than Nyquist. But we need to wrap our thinking around that, since it is not a signal we can use anymore, it's aliasing which will be mirrored back into lower spatial frequencies.
Because that separation of signal and aliasing is so alien to our everyday interpretation of image sharpness, I find the realism of the starchart resolution decline much easier to grasp. The difficulty is that we can no longer exactly separate signal and aliasing because they got fused together.
So both methods kind of complement each other, and that's why my test chart offers both.
Cheers,
Bart
-
I'm sceptical a subjective/judgement thing like appropriate image sharpening can be done on a target like this and not real world images. It's impossible to know what is generally appropriate for each camera's files without real world shooting with both. Still interesting though.
-
Hi,
Just as a comment. I used to sharpening settings- One is a preset coming from Lightroom and is probably developed by Jeff Schewe and is a general preset for "Landscape images". I found a different preset using deconvolution that I feel works well with my Sony Alpha cameras, that have an OLP filter. So I was using these two sharpenings.
Of course, any sharpening can be applied to any image. Overdoing deconvolution sharpening can cause ringing artifacts.
Best regards
Erik
I'm sceptical a subjective/judgement thing like appropriate image sharpening can be done on a target like this and not real world images. It's impossible to know what is generally appropriate for each camera's files without real world shooting with both. Still interesting though.
-
I'm sceptical a subjective/judgement thing like appropriate image sharpening can be done on a target like this and not real world images. It's impossible to know what is generally appropriate for each camera's files without real world shooting with both. Still interesting though.
Hi Mark,
There are two different things that need to be kept separate for a better understanding. The first is capture sharpening. This will only restore the original image quality by removing the blur that was introduced as a result of the capture process and demosaicing. That capture process and the effects of imperfect demosaicing can be very well analysed and quantified.
Ther second thing, and I think that you are referring to that, is 'creative' sharpening to bring out certain scene qualities that you want to (de-)emphasize. The Clarity control is one of the ways to tackle that, and its use is indeed subject dependent. There are no general rules there, because output size also determines how that works out.
Cheers,
Bart
-
I sure would love to see an evaluation of a real world print from both cameras with optimum sharpening appropriate for each (.9 800, .75 E) to see if in print, it's a toss up or is the E with less sharpening more natural?
Marc
-
Yes, to some extent the original question remains: what happens when optimal processing, including sharpening, is applied to each file? (Since it is clearly an error to apply the same settings to both.) We already have learned that the resolution difference is minimal. And now we know that a D800 file, sharpened carefully, can equal a D800e file with standard processing. As for what's left...well, if I had the original files, I would know for sure in a few minutes. At least to satisfy myself.
-
I find the whole conversation somewhat humorous based on my historical context. Having worked with the first Leaf back the cameras today are light years ahead but we still did great work with those antiques in the day. I am quite sure the MF cameras exceed any practical film I used.
The question is not if MF is better then the Nikon for me. It is if it is 3 times better, since the Pentax cost that much more. For me no. I guess its the business guy and not the scientist that is making the decision. From a cost analysis point of view the Nikon is a better decision. Two bodies plus extra cash instead of one. Then again I never have been paid to shoot a test target so my reasoning maybe flawed.
-
Paul: exactly. I don't think it could be said better.
-
Photography can be a business. It can also be an art. Based on simply cost, then the new 24MP APS senors would make the most sense businesswise. You could buy three to six of those compared to the D800. And then it might be a consideration of how much your business makes.
-
But the numbers speak for the Pentax 645D, unless you object to aliasing.
Interesting. In the last week or so I have seen more examples of moire in the D800e than I ever have in the 645D. I have short more than 5k images over the last year with the 645D and have never seen moire--but I have done few images with fabric or manmade features. I am not saying you are wrong, but the 645D has been moire free all year.
Naturally, since I have written that, I will find moire in every other image I take.
-
Hi,
It may have to do with the Nikon D800E having live view. Exact focus at large apertures would make the images more prone to moiré.
Best regards
Erik
Interesting. In the last week or so I have seen more examples of moire in the D800e than I ever have in the 645D. I have short more than 5k images over the last year with the 645D and have never seen moire--but I have done few images with fabric or manmade features. I am not saying you are wrong, but the 645D has been moire free all year.
Naturally, since I have written that, I will find moire in every other image I take.