-
Thank you Michael for getting this first look published so quickly!
I'm in line for an 800E and this is helpful. Looking forward to hearing more - especially any assesment of how the 800E's files may relate to prints of various sizes with a variety of papers.
Update: April 23 - took delivery of my D800E.
S.
-
It's always rewarding to say "told you so" ;) The "E" should be the regular model and the 'regular' 800 the specialty item for portrait/fashion and wedding shooters.
....kind of reminds me of 'unleaded' gas, where they charged you more not to put in something you never wanted in the first place....
- N.
-
It does indeed seem clear from the 100% crops that the D800E has a resolution advantage. If I were comparing the resolution of two lenses of equivalent focal lengths, at their sharpest apertures, and got this degree of difference, I know which lens I would choose, if the price were right..
I'm sure it's true that such differences can be narrowed with additional sharpening applied to the D800 image, but sharpening always seems to introduce additional noise. I imagine if one were to succeed in getting the D800 image as sharp as the D800E, with careful sharpening and additional sharpening, it would be a noisier image.
-
One request: I'm sure comparing two cameras for a highly demanding crowd isn't easy, but is it possible to add a video comparison as well? The big problem with DSLR video is that they line-skip, and the resolution is isn't on par with what one would expect from their sensors. But maybe the higher resolution of the E might make the resulting images sharper - that would be stellar.
On the other hand I would also love to find out if the 36MP E trumps a 645D or S2 as far as large prints are concerned.
Thanks!
-
Just played around with the unsharpened crops from the two cameras. If I sharpen each crop independently "to taste" to make it look as good as possible, the resolution difference remains.
D800E clearly wins the day IMHO, but you'd need to print big to see that difference.
-
Actually when printed it's not the increased sharpness that stands out, subjects look more three dimensional and "Palpable" when viewed at a distance
Marc
-
The D800E ... well it's a no brainer!
Kinda like to sell all my Canon gear including the 5D3.
-
I hope that other camera manufacturers are taking note of this.
I suspect that quite a few of us are willing to trade in the threat of moire for less blurry images.
I wonder if there's a "camera-mod" business that camera manufacturers could get into here, rather than forcing owners to look for 3rd parties to remove the AA filter.
-
In my opinion a more correct way to compare would be to use proper (capture) sharpening. Ironically Michael and Jeff Schewe go to great lengths to explain how best capture sharpening is done (and how much difference it makes in 1:1 view) in their very recommendable Lightroom tutorials. The differences would be less and actually both would be sharp which they aren't now. I think Nikon made the smart move here to provide both alternatives. The bottom line, though, is what difference it really makes in real life for the photographer. I'm sure Michael will more adequately deal with that in the coming articles on the subject. A comparison with medium format would in my view be more interesting. I have no doubt that if we go for very large prints that a full frame medium format camera would win as easily over the D800E as the Canon 5D mkII did over the 40D :) The sensor area ratio is about the same between Canon APS-C and full frame as there is between 35mm full frame to full frame medium format sensors. But again what are the realities about when these differences count and can be seen? That's much more interesting than the fairly minute differences between the D800 and the D800E (in my view).
-
In my opinion a more correct way to compare would be to use proper (capture) sharpening.
Hi Hans,
I agree, the Low-pass filter is not there to create a sharp high resolution image, it is there to avoid problems downstream in the postprocessing (false color artifacting when demosaicing Bayer CFAs), and to reduce the tendency for all sorts of aliasing artifacts, not only moiré.
Ironically Michael and Jeff Schewe go to great lengths to explain how best capture sharpening is done (and how much difference it makes in 1:1 view) in their very recommendable Lightroom tutorials. The differences would be less and actually both would be sharp which they aren't now.
Correct. It's like comparing images before and after color balancing. It's not a fair comparison, in fact it's misleading. One doesn't judge quality of a final product based on one of it's early stages of development. The only justification of the comparison shots as they are presented now would be to show how much difference the higher price of the 'E' produces, and then one should demonstrate if there is a benefit to paying the bonus for the final output after sharpening. And it is not the case that sharpening improves the quality of both image types, as Michael suggests in his preliminary conclusion. The images of the D800E will produce more artifacts (e.g. jaggies) at the pixel level when sharpening is added.
I think Nikon made the smart move here to provide both alternatives. The bottom line, though, is what difference it really makes in real life for the photographer. I'm sure Michael will more adequately deal with that in the coming articles on the subject. A comparison with medium format would in my view be more interesting. I have no doubt that if we go for very large prints that a full frame medium format camera would win as easily over the D800E as the Canon 5D mkII did over the 40D :) The sensor area ratio is about the same between Canon APS-C and full frame as there is between 35mm full frame to full frame medium format sensors. But again what are the realities about when these differences count and can be seen? That's much more interesting that the fairly minute differences between the D800 and the D800E (in my view).
A agree completely.
Cheers,
Bart
-
I wonder if there's a "camera-mod" business that camera manufacturers could get into here, rather than forcing owners to look for 3rd parties to remove the AA filter.
Try googling "maxmax".
-h
-
At last a good comparison ! from this first batch it is easy to choose for the E version. The difference is more than i thought.
I have seen d800 photos with moiré so I wonder how much it will get worse with the d800e
the moire i saw could be cured with the moiré tool in capture NX
Actually when printed it's not the increased sharpness that stands out, subjects look more three dimensional and "Palpable" when viewed at a distance
Marc
This more three dimensional look is what interests me very much, but i think you have to have two examples and print them first to find out? or can you see it on screen?
-
Thanks Michael. If it fits into your schedule, would you mind checking the AF focus point accuracy across the sensors? I have seen several people commenting that their cameras are off on the left side. Also it would be interesting to see comparisons of properly sharpened images from both cameras. I have a E on order, but just spent ten days using the D800. The detail I saw amazed me, and made me really anxious for the E’s arrival.
I photographed the shuttle Discovery on the 747 the day after I had to return the D800. Boy did I ever wish I had that camera one more day....
Bill
-
The images of the D800E will produce more artifacts (e.g. jaggies) at the pixel level when sharpening is added.
Noob question: Why is this the case for the 800E but not the 800?
-
One request: I'm sure comparing two cameras for a highly demanding crowd isn't easy, but is it possible to add a video comparison as well? The big problem with DSLR video is that they line-skip, and the resolution is isn't on par with what one would expect from their sensors. But maybe the higher resolution of the E might make the resulting images sharper - that would be stellar.
On the other hand I would also love to find out if the 36MP E trumps a 645D or S2 as far as large prints are concerned.
Thanks!
I plan on doing video comparisons, but I'm working the stills side of things for the moment. It might be a couple of weeks yet. Some quick and dirty video clips that I've shoot look very nice.
Llyod Chambers has done a comparison between the D800 and the Leica S2 and finds that it is somewhere between and raw and a slight win for the Nikon. The D800e will therefore likely win in almost all IQ counts.
The D800e is medium format's worst nightmare. Resolution matching cameras and backs up to 40MP, superior high ISO, equal colour depth, lighter weight, availability of hundreds of lenses, availability of stabilization, live view, video, much higher frame rates, and all for an integrated body costing less than US $3,300.
Michael
BTW – Anyone in the Toronto area who owns a Leica S2 and would like to do an S2 / D800e comparison with me one day, drop me a note.
-
As to the issue of sharpening, there is no simple answer. Because sharpening is done "by the numbers" but instead is highly subjective, it seemed to me that at this early stage unsharpened was the way to display the comparison.
I will say that in prints made with my usual input and then print sharpening, the difference seen and described relatively remain the same, so it's likely an academic issue at best. But I will do some further comparisons with appropriate sharpening.
Michael
-
Since the price differential between the two bodies is pretty much negligible (compared to that of a MF or Leica S2), the choice boils down to what the individual wants. I think Bernard opted for the 800 and it will be interesting to read his continuing thoughts on this as well as Michael's. It's good to see Nikon come up with a winner and those of us who are contemplating upgrades to our Nikon systems have a real good choice here (I'm sure that my daughter would love to have a free gift of a D300).
-
Noob question: Why is this the case for the 800E but not the 800?
The 800E (effectively) does not have an AA filter. The AA filter is a mechanism to reduce details that the sampling grid of the sensor can not record "reliably" - presence of such details will to some degree introduce "false" data (stair-stepping, moire) into the raw file that can not reliably and automatically be removed. The subjective importance of this is highly debated.
The 800 does have an AA filter, and should therefore have less such artifacts, at the price of a blurier raw file.
The question that many people have right now is: "what image will look the best (if any difference) when printed reasonably large and using ideal processing for each camera". Presumably "ideal" processing means more sharpening for the D800 than the D800E, and (in some cases) moire-removal for the D800E. If you do sharpening there is a risk for more visible noise, and as the noise characteristics are basically the same, "optimally processed" D800 images might be more noisy than their D800E counterpart, and less sharp (compromise between sharpening and noise visibility), but perhaps only for images that are prone to visible noise in the first place.
Since an image contains a finite number of bits/pixels, there is a naiive possibility that an expert photoshopper might develop any cameras images to be equal or better than any other camera. Although this seldom seems to happen, it is an argument that "ideal processing" should be somewhat restricted: test results where moire removal or noise/sharpness tradeoffs that can only be achieved using a number of expensive software tools and many hours of skilled manual labour is really only relevant to those that have the resources to do just that.
-h
-
First of all it's good news that the E has finally found it's way into some experienced hands. OTOH, my dealer originally told me I would get mine in the first shipment, but now says it is more likely to be one to two months ???. While a bit bummed about that, I now have ample time to upgrade my hard drives!
-
Noob question: Why is this the case for the 800E but not the 800?
Any regularly sampled discrete image data that is not properly low-pass filtered will exhibit aliasing. Whether that aliasing manifests itself in a recognizable and disturbing manner depends on the image content. Since the 'E' version has reduced AA-filtering, the aliasing will be stronger. Again, visibility depends on the circumstances, although Murphy's law dictates that it will show when we can least use it, e.g. when under time constraint to a publishing deadline.
Sharpening will 'enhance' artifacts as well as image detail, since it cannot discriminate between the two anymore, aliasing and image detail have merged.
Cheers,
Bart
-
As to the issue of sharpening, there is no simple answer. Because sharpening is done "by the numbers" but instead is highly subjective, it seemed to me that at this early stage unsharpened was the way to display the comparison.
Agreed, but both the AA filter and the Bayer filter cause blur, so some sharpening will be beneficial on both cases (I guess preaching to the choir :)). One approach could be to show some examples both without capture sharpening and with the (subjective) capture sharpening that is found optimal in each case. Then the reader can judge the cases. Yet another approach like on dpreview (that I like) is that it is possible to double check by the reader by downloading the RAW files for individual check in the personal favorite RAW processor. When that has been said, I think people read the articles here on LuLa as personal judgement of equipment in use in contrast to the camera reviews on the major review sites where personal opinions by the reviewer takes a back seat. I think the personal opinions are both respected and valued when they are stated as such and not as absolutes. No matter how it is done there will be long threads here and there about how wrong it was done ;)
Another issue raised is the false detail from a sensor without an AA filter. It seems to me that one approach to really investigate this rather than debate the theoretical aspects of it, would be to use a much higher resolution image to check to which degree there is false detail. This could be done by zooming in on high frequency detail by going much closer to the subject and stitch the pictures to a (very) high resolution picture where the false detail will be revealed in comparison. There may be other ways to do it. I know it would be quite work intensive to do this, but it may put an end to some of the discussion and make some conclusions about this issue. And see how the AA filter equipped D800 will fare in this comparison as well.
-
Hopefully this pre-review will not make the feeding frenzy ,frenzier!!
Michael , could you please drop a line or two on the ergonomics ?
Thank you
-
Hopefully this pre-review will not make the feeding frenzy ,frenzier!!
Michael , could you please drop a line or two on the ergonomics ?
Thank you
Coming up in my full report. In a word - typical Nikon. Grip is fin for me, with small hands, but those with large hands may find it a bit "short".
Michael
-
Coming up in my full report. In a word - typical Nikon. Grip is fin for me, with small hands, but those with large hands may find it a bit "short".
Michael
Great, thank you
-
I will say that in prints made with my usual input and then print sharpening, the difference seen and described relatively remain the same, so it's likely an academic issue at best. But I will do some further comparisons with appropriate sharpening.
Hi Michael,
Appreciated. What you'll see is that Capture sharpening the D800 images will boost the MTF of the lower spatial frequencies as well. That should bring the results much closer together. The only thing the D800E will do better is very low contrast (and high frequency) micro-detail.
Cheers,
Bart
-
Another issue raised is the false detail from a sensor without an AA filter. It seems to me that one approach to really investigate this rather than debate the theoretical aspects of it, would be to use a much higher resolution image to check to which degree there is false detail. This could be done by zooming in on high frequency detail by going much closer to the subject and stitch the pictures to a (very) high resolution picture where the false detail will be revealed in comparison. There may be other ways to do it. I know it would be quite work intensive to do this, but it may put an end to some of the discussion and make some conclusions about this issue. And see how the AA filter equipped D800 will fare in this comparison as well.
The simple way to demonstrate if there is any real resolution benefit, is to do an objective comparison of a star target like the free one I proposed (http://www.openphotographyforums.com/forums/showthread.php?t=13217). In its center it will demonstrate if the limiting resolution differs or (probably) not by the diameter of the center blur, and how the sensitivity for aliasing differs at various angles by diverging patterns (in color if the demosaicing fails), all in a single shot. That single shot can be an in-camera JPEG and a Raw file if you want to simultaneously verify the quality of the in-camera JPEGs.
Objectivity will help the readers, more than subjectivity/opinions (which will always be contested, this is the internet...) does.
Cheers,
Bart
-
This more three dimensional look is what interests me very much, but i think you have to have two examples and print them first to find out? or can you see it on screen?
In my experience, maxmax 5D and 3 MFDB's you can see it in all media: screen, printed or projected. I hope Michael posts a still life comparing the two 800's. I found it most noticeable on round and shinny objects. Perhaps a restored classic car with chrome bumpers would be a good demonstration? Unfortunately I sent the 5d out had it modified and didnot think to take before and after shots. Attached is an example of the 3 dimensional appearance I took this shot shortly after the modification.
Marc
-
I second the request for D800 vs D800E video performance information. The only reason I've ordered a D800 over an "E" is my fear of video artifacting.
The construction site comparison photos are most instructive. For the first time on viewing such images, I was unable to locate the source of the 100% crop in the full image. : )
-
Perhaps a restored classic car with chrome bumpers would be a good demonstration?
I know that Mark Dubovoy posted a bunch of images of classic cars on this site that would be well suited for this test but he doesn't favor the 35mm format. :) Would be interesting to get his take on the new Nikon.
-
In my experience, maxmax 5D and 3 MFDB's you can see it in all media: screen, printed or projected.
Hi Marc,
The so-called 3D-effect is IMHO caused by the shape of the MTF curve, and can be applied to other images as well. One just needs to boost the medium frequencies a bit (which also happens automatically when the highest spatial frequencies are Capture sharpened). That's why it indeed shows at all different magnifications.
Cheers,
Bart
-
I'd be interested to canvas views on this.
I own a couple of Sigmas and a NEX which have either no or very light AA filtering. I also have a Kodak 14n with no filter. Over the years of using these cameras, I have gradually become aware of something with respect to filter-less cameras: aliasing. Not the colour moire everyone likes to go on about, but luminance aliasing. I think it looks very characteristic between all the cameras, Foveon and Bayer CFA alike. I dislike it immensely, it renders like a strange noise pattern to my eye.
What I'm talking about is the way that some thin diagonal lines (that are actually tubes) render as if they were made of lego bricks or separate planks of wood, like a laminate. I appreciate this is endemic in the digital picture structure and this is what you see if you zoom in 2000% or so but I'm not talking about simple pixelation. I'm talking about something that is easily visible at 100% as "jaggies" and "strange patterns" that turn smooth tubes into a rope-like weave and make straight edges ripple - if you know what you are looking for. At 200% it is blatantly obvious, even if this is for forensic rather than practical purposes.
I find at realistic prints sizes the jaggies are rarely visible but it makes prints look both sharp and grainy/noisy at the same time in a way that doesn't quite look like noise. I like my little NEX, it should be a perfect lightweight landscape camera but I find the aliasing so obtrusive I've stopped using it.
The huge rez of the D800e should make it much less of an issue but take a look at MR's Moire test shot of the tree. Look VERY carefully at the thin branches, especially those at different angles.
Do you see the jaggies and ragged looking blockiness I see? Look using the full size version you get if you click. I need reading glasses to be sure.
I'd be very interested to find out who sees it and who doesn't, and whether many are bothered by it.
Cheers
Dave
-
What's most noticeable to me about this shot is the aliasing - and in a shot that is all curves and no patterns. The lip of the bowl at top left hand side just in front of the avocado where it is pure highlight is clearly suffering luminance aliasing on the curve.
In my experience, maxmax 5D and 3 MFDB's you can see it in all media: screen, printed or projected. I hope Michael posts a still life comparing the two 800's. I found it most noticeable on round and shinny objects. Perhaps a restored classic car with chrome bumpers would be a good demonstration? Unfortunately I sent the 5d out had it modified and didnot think to take before and after shots. Attached is an example of the 3 dimensional appearance I took this shot shortly after the modification.
Marc
-
I plan on doing video comparisons, but I'm working the stills side of things for the moment. It might be a couple of weeks yet. Some quick and dirty video clips that I've shoot look very nice.
Lloyd Chambers has done a comparison between the D800 and the Leica S2 and finds that it is somewhere between and raw and a slight win for the Nikon. The D800e will therefore likely win in almost all IQ counts.
The D800e is medium format's worst nightmare. Resolution matching cameras and backs up to 40MP, superior high ISO, equal colour depth, lighter weight, availability of hundreds of lenses, availability of stabilization, live view, video, much higher frame rates, and all for an integrated body costing less than US $3,300.
Michael
Hey Michael,
Since you've tested the Sigma SD1 as well do you feel that at low iso compared to the D800e, the Sigma is dead in the water from the aspect of pure image quality alone in prints 16x20 and under?
-
The so-called 3D-effect is IMHO caused by the shape of the MTF curve
Bart, I have been wondering about this, and asked in another thread somewhere. Can you explain what if anything, is the connect between the 3D effect and absence of an OLPF? My naive conjecture is that aliasing effects make the transitions at the edges of objects sharper, and they human visual system responds to this in its interpretation of foreground/background differences.
-
Hi,
Just reducing an image would also produce aliasing and this may be the case the tree image.
I'm not really sure about the NEX-7 lacking OLP filter, as I have measured MTF from test images on Imaging Resource and all Sony camera I tested came in about 20%MTF at Nyquist.
On the other hand I have seen moiré on my Sony Alpha 55SLT, but not yet on the Alpha 77SLT which has the same sensor as the NEX-7.
As a side note, Lloyd Chambers (http://diglloyd.com ) has tested the Nikon D800 (with OLP filter) against the Leica S2, with 120 /2.5 Macro on the Leica and Zeiss Macro Planar 100/2.0 on the Nikon D800. Both cameras used deconvulition sharpening, somewhat more on the D800 than on the Leica. The cameras were close in the center, with the Leica having some more contrast but the Nikon was definitively better in the corner. The subject was a building facade with lots of mosaiv and stuccos. The Leica S2 had excessive amount of Moiré.
I have some images from a corresponding test by Lloyd Chambers, involving the Leica S2 and the Nikon D3X. I published an analysis of those images with the kind permission of Mr. Chmbers, here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/38-observations-on-leica-s2-raw-images
Best regards
Erik Kaffehr
I'd be interested to canvas views on this.
I own a couple of Sigmas and a NEX which have either no or very light AA filtering. I also have a Kodak 14n with no filter. Over the years of using these cameras, I have gradually become aware of something with respect to filter-less cameras: aliasing. Not the colour moire everyone likes to go on about, but luminance aliasing. I think it looks very characteristic between all the cameras, Foveon and Bayer CFA alike. I dislike it immensely, it renders like a strange noise pattern to my eye.
What I'm talking about is the way that some thin diagonal lines (that are actually tubes) render as if they were made of lego bricks or separate planks of wood, like a laminate. I appreciate this is endemic in the digital picture structure and this is what you see if you zoom in 2000% or so but I'm not talking about simple pixelation. I'm talking about something that is easily visible at 100% as "jaggies" and "strange patterns" that turn smooth tubes into a rope-like weave and make straight edges ripple - if you know what you are looking for. At 200% it is blatantly obvious, even if this is for forensic rather than practical purposes.
I find at realistic prints sizes the jaggies are rarely visible but it makes prints look both sharp and grainy/noisy at the same time in a way that doesn't quite look like noise. I like my little NEX, it should be a perfect lightweight landscape camera but I find the aliasing so obtrusive I've stopped using it.
The huge rez of the D800e should make it much less of an issue but take a look at MR's Moire test shot of the tree. Look VERY carefully at the thin branches, especially those at different angles.
Do you see the jaggies and ragged looking blockiness I see? Look using the full size version you get if you click. I need reading glasses to be sure.
I'd be very interested to find out who sees it and who doesn't, and whether many are bothered by it.
Cheers
Dave
-
I plan on doing video comparisons, but I'm working the stills side of things for the moment. It might be a couple of weeks yet. Some quick and dirty video clips that I've shoot look very nice.
The simple way to demonstrate if there is any real resolution benefit, is to do an objective comparison of a star target like the free one I proposed (http://www.openphotographyforums.com/forums/showthread.php?t=13217). In its center it will demonstrate if the limiting resolution differs or (probably) not by the diameter of the center blur, and how the sensitivity for aliasing differs at various angles by diverging patterns (in color if the demosaicing fails), all in a single shot. That single shot can be an in-camera JPEG and a Raw file if you want to simultaneously verify the quality of the in-camera JPEGs.
Bart,
I have the 800E on order, and would perform the test if I had the camera, but I don't and so I can't at present. Since Michael has both cameras, he could do us a great service by taking shots of the target and posting the images. One could gain a great deal of information with minimal effort.
Hint, hint, Michael
-
Bart,
I have the 800E on order, and would perform the test if I had the camera, but I don't and so I can't at present. Since Michael has both cameras, he could do us a great service by taking shots of the target and posting the images. One could gain a great deal of information with minimal effort.
Hint, hint, Michael
I'll be up at my studio this weekend and will try and do it then.
Michael
-
I'll be up at my studio this weekend and will try and do it then.
Michael
Thanks in advance. The results will be interesting.
Regards,
Bill
-
I appreciate this set of comparisons. For me, though, the real issue isn't whether or not the D800e would be sharper with no sharpening, or the same sharpening as the D800. The question for me is whether the 800e would be sharper if files from both cameras were sharpened to their maximum potential, especially using deconvolution. The 800e files probably can take less sharpening before showing artifacts.
I wasn't clear on whether the second set of comparisons was done with identical sharpening or not. But to me it looks like the 800 file could handle more sharpening, especially for printing.
It is also interesting to me that the second set of comparisons was so much less dramatic than the first. So it sure seems that optimal sharpening could be a big player in any real world usage.
-
I appreciate this set of comparisons. For me, though, the real issue isn't whether or not the D800e would be sharper with no sharpening, or the same sharpening as the D800. The question for me is whether the 800e would be sharper if files from both cameras were sharpened to their maximum potential, especially using deconvolution.
And right you are.
The 800e files probably can take less sharpening before showing artifacts.
Yes, that's what's to be expected, QED.
I wasn't clear on whether the second set of comparisons was done with identical sharpening or not. But to me it looks like the 800 file could handle more sharpening, especially for printing.
That's what is often the case for larger output sizes. The (onset of) aliasing is a limiting factor for the amount of output sharpening, although there are applications that do better (http://www.luminous-landscape.com/forum/index.php?topic=62609.msg505337#msg505337) than average.
Cheers,
Bart
-
I wasn't clear on whether the second set of comparisons was done with identical sharpening or not. But to me it looks like the 800 file could handle more sharpening, especially for printing.
Consider the noise implications of greater sharpening.
-
The question for me is whether the 800e would be sharper if files from both cameras were sharpened to their maximum potential, especially using deconvolution. The 800e files probably can take less sharpening before showing artifacts.
Regarding this, I think it's important to distinguish between capture sharpness (accutance), and real detail (information). Sampling theory says that no detail is lost if an ideal AA filter is used followed by ideal impulse sampling. The differences in real detail (not in capture sharpness) between the D800 and D800E should be explained through the non-ideal AA filter of the D800.
Just for demostrative purposes, using Photoshop I resampled down to 256x256 a 2048x2048 image with repetitive patterns prone to aliasing/moiré (see accordion) using different AA filters, from no AA filter to 4px gaussian blur, followed by ideal sampling (decimation):
No aliasing in the original file:
(http://www.guillermoluijk.com/misc/aliasing2.jpg)
Rescaling down in Photoshop with different AA filters:
(http://www.guillermoluijk.com/misc/aliasing.jpg)
Aliasing artifacts are clear in all images, being the one with 4px AA filtering the first one acceptable IMO. In the non problematic areas (head and clothes of the musician), the 4px image is less sharp. But does it contain less useful information than the aliased images?.
After sharpening it (centre), I doubt there is really any extra information in the other images, which where still a bit sharper but displayed aliasing artifacts (see accordion and metallic tubes of the chair in the 2px version):
(http://www.guillermoluijk.com/misc/aliasing3.jpg)
Translating this into the D800 vs D800E debate, what is a better choice: more protection against aliasing/moiré at the expense of some (recoverable?) loss of sharpness? or extra sharpness (without additional useful information?) at the expense of more aliasing/moiré artifacts in particular scenes?.
Regards
-
Thanks a lot!
Erik
Thanks in advance. The results will be interesting.
Regards,
Bill
-
Hi,
I played around a bit with sharpening on a screen dump of Michael's image.
The screen dump shows the duplicates:
1) Original image
2) Right hand image sharpened
3) Both images sharpened by the same amount
That was quite extensive sharpening and it of course much affected by JPEG artifacts.
Normally I would use the enclosed settings for low ISO landscape images in Lightroom. May be a bit excessive but I like them.
Best regards
Erik
-
Regarding this, I think it's important to distinguish between capture sharpness (accutance), and real detail (information). Sampling theory says that no detail is lost if an ideal AA filter is used followed by ideal impulse sampling. The differences in real detail (not in capture sharpness) between the D800 and D800E should be explained through the non-ideal AA filter of the D800.
From the simplified frequency-domain theoretical dsp POV, a sin(x)/x function might appear perfect, as it has a perfectly flat, wide passband (true detail), and perfect attenuation of the stop-band (no aliasing). In practice, this could lead to ringing and other artifacts around sharp edges.
The "ideal" linear optical prefilter would probably be close to the kernels preferred for image downsampling. I think that lanczos (windowed sin(x)/x) is close to what most seem to prefer. But an optical aa filter could not have "negative light" contributions, so it might ideally be closer to a gaussian or whatever.
Practice tells us that photographers tend to be less concerned with what is "real" detail, and more concerned with what "gives an impressive popping 3d print". Even if the D800 had the best conceivable OLPF (which it does not) in terms of sampling theory, some users might prefer the aliasing bound to be present in the D800E. I think it is a good thing that Nikon gave users the choice.
-h
-
I have been eagerly awaiting comparisons like these trying to make a decision, after nikon's samples I was surprised to see just how much more detail there with in the E file. What would be interesting is at what apertures of the 85mm this difference remains evident so see how useful it will be in the real world, if it's a wash by f/11 or at f/2.8 the usefulness is limited somewhat.
-
Consider the noise implications of greater sharpening.
Proper sharpening does not have to increase the overall noise to unacceptable levels. Featureless areas can/should be masked, and areas with many details already have a favorable signal to noise ratio.
On top of that, algorithms like an adaptive Richardson-Lucy restoration will already dampen the noise and increase the S/N ratio with each iteration. One can try RawTherapee to see how far one can push the sharpening before noise becomes a problem (which it rarely is with printing).
Cheers,
Bart
-
Proper sharpening does not have to increase the overall noise to unacceptable levels.
Well, I guess that's self-evident, Bart. If the sharpening were to increase noise to unacceptable levels, it couldn't be considered proper. ;D
What occurs to me is, if the D800 shot is always sharpened so that it looks as sharp as the D800E shot, could there be visibly more noise at least somewhere in the full size image, despite one's best efforts.
It may well be the case that at base ISO this is less likely to be a problem. But what about ISO 3200? What about small crops that one wants to print largish?
Featureless areas can/should be masked, and areas with many details already have a favorable signal to noise ratio.
I agree that featureless areas can be masked, but detail in the deep shadows tends to have an unfavourable SNR. When I use Smart Sharpen, I usually fade the amount of sharpening applied to the shadows.
I also wonder what will happen at high ISO when noise may be a problem without any sharpening at all.
On top of that, algorithms like an adaptive Richardson-Lucy restoration will already dampen the noise and increase the S/N ratio with each iteration. One can try RawTherapee to see how far one can push the sharpening before noise becomes a problem (which it rarely is with printing).
Maybe, but I don't have the cameras to try that. I have to make a decision as to which model to order. I'm currently favouring the D800E.
I was very impressed with Focus Magic, but I'm disappointed they are taking so long to develop a version compatible with 64 bit Windows OS.
Cheers!
Ray
-
I find it very surprising that Michael says that he can see the difference between the two Nikon c
ameras in prints. Not long ago he posted a story comparing prints from a Canon G12 and a Hasselblad saying most of his photographer friends could not tell the cameras apart; in prints.
-
I find it very surprising that Michael says that he can see the difference between the two Nikon c
ameras in prints. Not long ago he posted a story comparing prints from a Canon G12 and a Hasselblad saying most of his photographer friends could not tell the cameras apart; in prints.
The situations are quite different. In the case of the G12/P45 comparison, the MF lens was not used at its sharpest aperture whereas the G12 probably was. (From memory, I think it was F3.5 compared with F11).
Secondly, the print size of A3+ would have resulted in most of the resolution benefits of the P45 being thrown away in downsampling, whereas A3+ size is just about right for the G12 without any interpolation of pixels.
The 36.3mp of the D800 is good for a 20"x30" print without interpolation (depending on choice of 240ppi or 300ppi), and that is the size, or larger, one would compare detail and resolution. In practice, people frequently interpolate images to make large prints. I've made 24"x36" prints from my 12mp Canon 5D, which look acceptably sharp from a reasonably close distance, like a viewing distance equal to the diagonal of the print. But I sure wish I'd had a D800E at the time I took those shots.
-
Having examined Michael's second, updated comparison of shots of a $20 bill after both have been sharpened, I'm beginning to think the resolution differences between the D800 and the D800E really are too small to be worth bothering about.
The differences in the unsharpened images are sufficiently great to become a concern, but appropriate sharpening seems to have narrowed the differences to a point where one really wonders under what circumstances, outside of extreme pixel-peeping, such differences would be apparent.
Can I suggest the following test. Michael should make a 24"x36" print from the D800E shot, according to his satisfaction, then ask Jeff Schewe to make another print from the D800 RAW image of the same scene, using the same paper, ink, printer and profile etc, with the request that Jeff try to make the D800 print look as close as possible in quality, sharpness, color and contrast etc to the D800E print which Michael has previously produced.
Both prints should then be circulated to experienced photographers in a double-blind test to see if they can correctly identify which is which.
If most of the viewers are able to see the differences when holding the prints in their hands and viewing them from the close distance one might read a book, the prints should then be placed on a wall, side by side, and viewed from a more sensible distance equal to the diagonal of the print. If the differences can still be correctly identified in that situation, then I think there's no doubt that the D800E is a worthwhile improvement.
However, I'm still uncertain about the noise implications of sharpening under certain circumstances. Is it likely that deep shadows in a D800E shot will be noticeably cleaner as a result of the lower sharpening required?
When the images are shot at a high ISO where noise may be almost unacceptable, will the differences between the two cameras then be more noticeable, the D800E shot being either noticeably sharper or noticeably less noisy?
-
Having examined Michael's second, updated comparison of shots of a $20 bill after both have been sharpened, I'm beginning to think the resolution differences between the D800 and the D800E really are too small to be worth bothering about.
The differences in the unsharpened images are sufficiently great to become a concern, but appropriate sharpening seems to have narrowed the differences to a point where one really wonders under what circumstances, outside of extreme pixel-peeping, such differences would be apparent.
Hi Ray,
Consider this. In digital imaging, broadly (lens specifics aside) speaking, resolution is defined by two things:
- Sampling density (assuming quality optics and good enough technique, i.e. proper use of a tripod, and perfect focus). Sampling the projected image with a finer discrete sampling grid will allow to reconstruct the original analog image more accurately. Since both the D800 and the D800E have an identical sampling density, there is no reason to expect any difference caused by this discriminator.
- Amplitude of the Signal to Noise (S/N) ratio at the limiting resolution. The sole purpose of an Anti-Aliasing or Optical Low-Pass Filter (OLPF) is to reduce the amplitude of the signal at (and thus beyond) the Nyquist frequency. High signal levels above Nyquist will 'fold back' to coarser features with spatial frequencies below Nyquist. Due to the different sampling densities of Green versus Red/Blue, the 'strength' of the OLPF is usually under-designed, i.e. it's too weak to avoid all aliasing, and as a consequence leaves enough S/N to restore, in the focus plane.
That means that at the absolute Nyquist frequency limit (set 'in concrete' by the sampling density), there is usually enough signal left in the focus plane (despite the OLPF) to allow restoration to higher relative levels with clever sharpening. That restoration will not only boost the amplitude of the signal at the Nyquist frequency, but also in lower spatial frequencies, which enhances the so-called 3D look of images (which also explains why it also has an effect in smaller sized prints), because the overall MTF is boosted (also for lower spatial frequencies, coarser detail gains acutance). The lack of an OLPF boosts amplitude of the below-Nyquist spatial frequencies (which is welcome), in addition to the aliasing artifacts (with a high enough amplitude to cause occasional trouble, which is a nuisance).
Therefore it's not a real surprise that proper sharpening pretty well levels the playing field between both versions (with a slight advantage to to 'E' version IF aliasing doesn't spoil the party, which it will when you can least use it, AKA Murphy's law).
Cheers,
Bart
-
- Amplitude of the Signal to Noise (S/N) ratio at the limiting resolution. The sole purpose of an Anti-Aliasing or Optical Low-Pass Filter (OLPF) is to reduce the amplitude of the signal at (and thus beyond) the Nyquist frequency. High signal levels above Nyquist will 'fold back' to coarser features with spatial frequencies below Nyquist. Due to the different sampling densities of Green versus Red/Blue, the 'strength' of the OLPF is usually under-designed, i.e. it's too weak to avoid all aliasing, and as a consequence leaves enough S/N to restore, in the focus plane.
That means that at the absolute Nyquist frequency limit (set 'in concrete' by the sampling density), there is usually enough signal left in the focus plane (despite the OLPF) to allow restoration to higher relative levels with clever sharpening. That restoration will not only boost the amplitude of the signal at the Nyquist frequency, but also in lower spatial frequencies, which enhances the so-called 3D look of images (which also explains why it also has an effect in smaller sized prints), because the overall MTF is boosted (also for lower spatial frequencies, coarser detail gains acutance). The lack of an OLPF boosts amplitude of the below-Nyquist spatial frequencies (which is welcome), in addition to the aliasing artifacts (with a high enough amplitude to cause occasional trouble, which is a nuisance).
Therefore it's not a real surprise that proper sharpening pretty well levels the playing field between both versions (with a slight advantage to to 'E' version IF aliasing doesn't spoil the party, which it will when you can least use it, AKA Murphy's law).
Cheers,
Bart
Hi Bart,
I understand the broad concept here, but there's also another truism; if the detail doesn't exist in the first instance, in the captured data, then no amount of sharpening can create it.
If the fine detail which is close in resolution to the sensor's Nyquist limit, is of high contrast, has a reasonably high MTF, then I can understand that the detail may still be captured despite the OLPF, but with a very faint or reduced contrast. Clever sharpening may well restore such faint detail to the same contrast level as in the D800E capture.
But supposing the contrast of that fine detail in the real scene being photographed, does not have good contrast. It may then be the case that the camera with the OLPF does not capture such detail at all, but the D800E does succeed in capturing the detail, albeit faintly.
Clever sharpening might then restore such faint detail in the D800E image to something which is noticeable in a print.
Cheer! Ray
-
Having examined Michael's second, updated comparison of shots of a $20 bill after both have been sharpened, I'm beginning to think the resolution differences between the D800 and the D800E really are too small to be worth bothering about.
The differences in the unsharpened images are sufficiently great to become a concern, but appropriate sharpening seems to have narrowed the differences to a point where one really wonders under what circumstances, outside of extreme pixel-peeping, such differences would be apparent.
Can I suggest the following test. Michael should make a 24"x36" print from the D800E shot, according to his satisfaction, then ask Jeff Schewe to make another print from the D800 RAW image of the same scene, using the same paper, ink, printer and profile etc, with the request that Jeff try to make the D800 print look as close as possible in quality, sharpness, color and contrast etc to the D800E print which Michael has previously produced.
Both prints should then be circulated to experienced photographers in a double-blind test to see if they can correctly identify which is which.
If most of the viewers are able to see the differences when holding the prints in their hands and viewing them from the close distance one might read a book, the prints should then be placed on a wall, side by side, and viewed from a more sensible distance equal to the diagonal of the print. If the differences can still be correctly identified in that situation, then I think there's no doubt that the D800E is a worthwhile improvement.
However, I'm still uncertain about the noise implications of sharpening under certain circumstances. Is it likely that deep shadows in a D800E shot will be noticeably cleaner as a result of the lower sharpening required?
When the images are shot at a high ISO where noise may be almost unacceptable, will the differences between the two cameras then be more noticeable, the D800E shot being either noticeably sharper or noticeably less noisy?
Thanks Ray for suggesting this test. I was about to suggest the same. Unless you provide such a double blind test no human being can escape from the confirmation bias we all suffer from ;)
-
I am in line for an 800E. I used to be a big photography fan.. In fact my Son worked for a great photographer Herman Leonard for some years. But with the loss of all equipment in Katrina that ferver vanished. But just lately have some Nikon Lens around and decided to pre order an 800E so will be very excited to receive it. Appreciate the thoughts posted. Now if I can just understand much of the digital workflow, this old man can get back to taking some photos.
-
I am in line for an 800E. I used to be a big photography fan.. In fact my Son worked for a great photographer Herman Leonard for some years. But with the loss of all equipment in Katrina that ferver vanished. But just lately have some Nikon Lens around and decided to pre order an 800E so will be very excited to receive it. Appreciate the thoughts posted. Now if I can just understand much of the digital workflow, this old man can get back to taking some photos.
loved Herman Leonard's music photos...and saw a documentary with him going back home, to his studio, after that disaster...really sad stuff...
give yourself some time to get a handle on digital workflow...read a lot about what others do or suggest, then adopt the 'methods' for processing and archiving that make the most sense 'to you' and for your needs...
i've been using a D700 the last three years and it's great...will wait a bit to see more on the D800/e...then maybe upgrade, maybe wait...before I buy I'm hoping to play with one in a store, listen to it...the clack of the shutter/mirror slap on my D700 is very loud...hoping the d800 is much more quiet...one of the few criticisms i have of this lovely nikon body...
-
I am puzzled by what appears to be a rationalization that "a proper sharpening would negate the advantage the 800e has". I am not a scientist, but what I've learned so far about the subject matter is that in post processing you can improve/sharpen lens acutance (i.e., edge contrast) but you can not (much) improve lens resolution (i.e., fine detail). If so, no amount of sharpening would improve (much) the lack of fine detail in 800 (relative to 800e). No?
Unless what you squints are saying is that, given the same lens, the amount of fine detail is the same in both cases, just blurred by AA more in 800 than in 800e, and that, with a proper reverse engineering of that blurring, one can come to the same starting point, i.e., the same amount of detail the lens is otherwise capable of. Is that it?
-
I am puzzled by what appears to be a rationalization that "a proper sharpening would negate the advantage the 800e has". I am not a scientist, but what I've learned so far about the subject matter is that in post processing you can improve/sharpen lens acutance (i.e., edge contrast) but you can not (much) improve lens resolution (i.e., fine detail). If so, no amount of sharpening would improve (much) the lack of fine detail in 800 (relative to 800e). No?
Hi Slobodan,
No, proper sharpening is not the same as enhancing acutance by boosting edge contrast. Proper (deconvolution) sharpening actually restores the resolution of the original signal to what it was before the blur got added. You can see an example of it in this thread (http://www.luminous-landscape.com/forum/index.php?topic=45038.msg378541#msg378541).
So it's more than edge contrast, it's actual resolution that can be restored to a very large extend, only it's minus the aliasing artifacts because those spatial frequencies were attenuated before they were captured by a discrete sampling grid.
If it were only edge contrast, then I'd be as puzzled as you are/were.
Unless what you squints are saying is that, given the same lens, the amount of fine detail is the same in both cases, just blurred by AA more in 800 than in 800e, and that, with a proper reverse engineering of that blurring, one can come to the same starting point, i.e., the same amount of detail the lens is otherwise capable of. Is that it?
You've got it, that's it. Actual restoration instead of simulation.
Cheers,
Bart
-
The simple way to demonstrate if there is any real resolution benefit, is to do an objective comparison of a star target like the free one I proposed (http://www.openphotographyforums.com/forums/showthread.php?t=13217). In its center it will demonstrate if the limiting resolution differs or (probably) not by the diameter of the center blur, and how the sensitivity for aliasing differs at various angles by diverging patterns (in color if the demosaicing fails), all in a single shot. That single shot can be an in-camera JPEG and a Raw file if you want to simultaneously verify the quality of the in-camera JPEGs.
Objectivity will help the readers, more than subjectivity/opinions (which will always be contested, this is the internet...) does.
Cheers,
Bart
Bart,
Agreed. Your sine-radial modified Siemens star chart provides an excellent means of quickly and objectively comparing and evaluating a wide range of sensor/lens characteristics. However, I would also very much like to see the full suite of SFR/MTF test results from Imatest---run by someone who is thoroughly familiar with the proper Imatest methodology to make valid comparisons between the D800 and the D800E. This should include using a lens that substantially out-resolves the 36MP sensor (one of the better Zeiss offerings that performs well at f/4 to f/5.6 should work well here, at least in the center of field). A plain vanilla raw converter like Dave Coffin's DCRaw would provide a pure linear conversion and eliminate possible bias in the raw converter.
Further, the tests could be performed with a combination of no sharpening, optimized ACR/Lightroom sharpening and, as suggested, with a top Deconvolution sharpening algorithm (preferably optimized based on a careful evaluation of the PSF of each lens/sensor system). The charts could be either a well made sine pattern or slant-edge patterns with low to high contrast elements. If the same lens is used on both cameras and since these are essentially identical sensors at heart (except for the AA filters), the results should be quite relevant in isolating the effect of the AA filter difference.
I think seeing the entire SFR/MTF plots for each of these cases, especially the response for both sensors at spatial frequencies near Nyquist (and at medium frequencies where the apparent "3D" effect is seen), would make clear any advantages the D800E might have over the D800. Specifically, if optimized sharpening performed on the image from D800 sensor is able to boost the response curve at and below the Nyquist cutoff enough to substantially match the response curve from the (optimally sharpened) D800E image, then the D800 may be the better choice. It would also be interesting to see the response curve above Nyquist on both sensors to show the likelihood of aliasing, although the radial pattern of your star chart might better illustrate this if the resolution of the test lens is not symmetrical.
I'm sure one of the commercial sites will eventually perform MTF/SFR testing with Imatest or DxO software---but the near total lack of published testing methodology on these sites (raw conversion?, chart contrast?, sharpening?, focus distance?, etc.) has rendered me highly skeptical of their results. :-\
EDIT: The robust Imatest noise module might also objectively answer the difficult question of how much the relative noise levels suffer on the D800, due to more aggressive required sharpening, compared to the D800E.
Kind regards,
Bob
-
Hi Bart,
Restoring an image using deconvolution needs a decent PSF (Point Spread Function), and I'm not sure we have it. The article you refer to was about restoring an image with diffraction, and it seems that using a gaussian (bell curve) as PSF yields decent results, but I don't know what happens with an image formed by an AA-filter.
Diglloyd compared the Leica S2 with a Nikon D800. The Leica had moiré "all over the place" while the Nikon had very little. My guess is that Lloyd Chambers ("Diglloyd") will reshoot the same image with the D800E. Moiré is an interaction between sensel pitch and structure pitch in the subject, so Moiré may or many not arise. Occurrence of moiré pretty much indicates that the imaging system aliases but the absence of moiré doesn't tell anything.
Best regards
Erik
Hi Slobodan,
No, proper sharpening is not the same as enhancing acutance by boosting edge contrast. Proper (deconvolution) sharpening actually restores the resolution of the original signal to what it was before the blur got added. You can see an example of it in this thread (http://www.luminous-landscape.com/forum/index.php?topic=45038.msg378541#msg378541).
So it's more than edge contrast, it's actual resolution that can be restored to a very large extend, only it's minus the aliasing artifacts because those spatial frequencies were attenuated before they were captured by a discrete sampling grid.
If it were only edge contrast, then I'd be as puzzled as you are/were.
You've got it, that's it. Actual restoration instead of simulation.
Cheers,
Bart
-
Hi,
Imaging resource used to publish downloadable raws with patterns usable for SRF/MTF. I'm doing that test once images are available.
Best regards
Erik
Bart,
Agreed. Your sine-radial modified Siemens star chart provides an excellent means of quickly and objectively comparing and evaluating a wide range of sensor/lens characteristics. However, I would also very much like to see the full suite of SRF/MTF test results from Imatest---run by someone who is thoroughly familiar with the proper Imatest methodology to make valid comparisons between the D800 and the D800E. This should include using a lens that substantially out-resolves the 36MP sensor (one of the better Zeiss offerings that performs well at f/4 to f/5.6 should work well here, at least in the center of field). A plain vanilla raw converter like Dave Coffin's DCRaw would provide a pure linear conversion and eliminate possible bias in the raw converter.
Further, the tests could be performed with a combination of no sharpening, optimized ACR/Lightroom sharpening and, as suggested, with a top Deconvolution sharpening algorithm (preferably optimized based on a careful evaluation of the PSF of each lens/sensor system). The charts could be either a well made sine pattern or slant-edge patterns with low to high contrast elements. If the same lens is used on both cameras and since these are essentially identical sensors at heart (except for the AA filters), the results should be quite relevant in isolating the effect of the AA filter difference.
I think seeing the entire SRF/MTF plots for each of these cases, especially the response for both sensors at spatial frequencies near Nyquist (and at medium frequencies where the apparent "3D" effect is seen), would make clear any advantages the D800E might have over the D800. Specifically, if optimized sharpening performed on the image from D800 sensor is able to boost the response curve at and below the Nyquist cutoff enough to substantially match the response curve from the (optimally sharpened) D800E image, then the D800 may be the better choice. It would also be interesting to see the response curve above Nyquist on both sensors to show the likelihood of aliasing, although the radial pattern of your star chart might better illustrate this if the resolution of the test lens is not symmetrical.
I'm sure one of the commercial sites will eventually perform MTF/SRF testing with Imatest or DxO software---but the near total lack of published testing methodology on these sites (raw conversion?, chart contrast?, sharpening?, focus distance?, etc.) has rendered me highly skeptical of their results. :-\
EDIT: The robust Imatest noise module might also objectively answer the difficult question of how much the relative noise levels suffer on the D800, due to more aggressive required sharpening, compared to the D800E.
Kind regards,
Bob
-
Hi Bart,
Restoring an image using deconvolution needs a decent PSF (Point Spread Function), and I'm not sure we have it. The article you refer to was about restoring an image with diffraction, and it seems that using a gaussian (bell curve) as PSF yields decent results, but I don't know what happens with an image formed by an AA-filter.
Hi Erik,
I don't think it's that useful to characterize the PSF of the AA-filter in isolation. We will always have a mix of PSFs in our optical system. The lens with its residual aberrations, (de-)focus, the aperture, IR-filter / OLPF package, microlenses, Bayer CFA, and sensel aperture size and shape, they all add something to the combined PSF and resulting MTF. Fortunately that combined PSF can be approximated quite well with a Gaussian like PSF. Theoretically we might be able to design an even more accurate PSF, but in practice we are confronted with too many variables (such as defocus) to deal with, and the optimal PSF would not be that useful anymore. Only in a laboratory situation would it make sense to devise an optimal PSF for a given lens/camera/subject combination.
As we speak, I'm preparing to launch a free web-based tool that will allow to exploit the slanted edge feature which is present on some test charts, in order to generate a Gaussian PSF kernel. Even though there are not many applications that allow the direct use of a user generated PSF kernel, at least it will allow to calculate the optimal radius setting for capture sharpening of a well focused shot.
One could assemble a personal collection of radius settings for one's own camera/lens combinations at various aperture settings, and use those radii for e.g. Lichtroom or ACR. Maybe the differences between the lenses are small enough to use the same settings for different lenses at various apertures. Alternatively one could, for a specific circumstance, build a specific kernel for a specific situation or shooting scenario.
An example could be a security camera image that needs to be enhanced. One can after the fact record an image of a slanted edge target with that specific camera and use the resulting PSF information to enhance the earlier recordings of an event at the same shooting distance / focus quality. Or one can envision a scenario where e.g. the near end of the DOF zone for a hyperfocal landscape shooter is analyzed, and the radius parameter we find will enhance both the near and the far end of the hyperfocal DOF zone.
Of course we can also use it to determine the actual resolution difference between the D800 and D800E after using the optimal radius settings for capture sharpening in various applications ...
Cheers,
Bart
-
I'm going to get one of these two cameras, but which one is something of a head-scratcher.
After looking at the second set of photos (the Canadian currency), and the exceptional conditions under which the test was shot (no wind, super-rigid set-up, concrete slab, mirror LU, etc.), it seems to me that almost any shots taken under less rigorous conditions would probably not show any difference at all. That is, if you were shooting at the Grand Canyon with a little wind with a lighter tripod & head on a dirt pathway, that you might degrade the quality enough that any detectable differences would depend more on outside factors than on the sensor. And that would certainly be the case with ANY hand-held shots; the mirror vibration alone might be enough to do that.
And if, given even slightly variable conditions, there is no practical difference (that is, you really couldn't predict which one would give you a better shot), then perhaps it would be better to go with the 800...because despite Michael's inability to find moire, it does exist, and the 800 helps lessen that effect, as is demonstrated in some of Nikon's own illustrations.
Maybe the question comes down to this: do near-perfect conditions occur more often than moire?
Goofy as it sounds, I think I'm working my way around to a complaint that is exactly the opposite of the complaints about test conditions that he referred to -- the test conditions are too good for me. I'd actually like to see a test in which Michael shoots, say, six different handheld views, in matching pairs with each of the two cameras, does his best to sharpen them, and then see if Jeff Schewe can pick out the 800e shots.
(I'd also like to see a test in which Michael has four margaritas, jogs down to to the lakefront, and then takes a couiple of handheld shots while being taunted by clowns, so it'd more closely resemble my own photography; but I doubt we'll see that.) ;D
-
Hi,
Imaging resource used to publish downloadable raws with patterns usable for SRF/MTF. I'm doing that test once images are available.
Best regards
Erik
Hi Erik,
Thanks for reminding me to recheck the Imaging Resource site! I downloaded some D800 raw files of general images a few weeks ago, but did not find any raws of the usual test patterns. The raws of resolution charts have since been posted:
http://www.imaging-resource.com/PRODS/nikon-d800/nikon-d800A7.HTM
Nothing yet on the D800E and they have not yet posted the full review of the D800, but Imaging Resource is usually the last to provide their complete review.
I'm still wishing they would update their ancient resolution chart to allow better MTF testing in Imatest. But this should be a good start for comparing the two sensors as long as they are 100% consistant with their methods when testing both the D800 and D800E. The Sigma 70mm macro (at f/5.6) used for the D800 tests looks like a pretty good choice, although I can think of lenses with better center resolution to help reduce the effect of the lens on MTF. When I get a chance, I'll also try running the Imaging Resource slant-edge sample from the D800 through Imatest. I'll be interested to see your results when the D800E images are posted.
Kind regards,
Bob
-
Michael's initial comparison was startling because it showed a greater difference between the two cameras than many would have expected. Not exactly sure what the explanation is there, but the second comparison is more in line with what I'd have expected.
I tend to fall on the side that AA filters are a good thing to have in cameras. I understand the reasoning behind them and I personally find the aliasing that often occurs in AA-free images objectionable. I'm primarily a landscape shooter so moire isn't much of a concern for me, but color aliasing and edge aliasing are. So I'm inclined to go with the D800.
The only thing that gives me some doubt is whether you really need to be shooting at f/4-f/5.6 for aliasing to show up in any quantity on a 36mp 135-format sensor. Even with T/S lenses I'm more likely to be shooting at f/8 than f/4, and f/11 is not unusual for me. So for now I'm waiting for more tests/comparisons to become available before I decide.
-
Hi Bart,
Hi Erik,
As we speak, I'm preparing to launch a free web-based tool that will allow to exploit the slanted edge feature which is present on some test charts, in order to generate a Gaussian PSF kernel. Even though there are not many applications that allow the direct use of a user generated PSF kernel, at least it will allow to calculate the optimal radius setting for capture sharpening of a well focused shot.
That sounds very interesting. I can imagine X-Rite adding a slant-edge pattern to their ColorChecker Passport so one could simultaneously profile a custom PSF along with the custom color profile! ;D
Of course we can also use it to determine the actual resolution difference between the D800 and D800E after using the optimal radius settings for capture sharpening in various applications ...
That's all I'm really expecting/hoping from this exercise.
Kind regards,
Bob
-
I don't think it's that useful to characterize the PSF of the AA-filter in isolation. We will always have a mix of PSFs in our optical system. The lens with its residual aberrations, (de-)focus, the aperture, IR-filter / OLPF package, microlenses, Bayer CFA, and sensel aperture size and shape, they all add something to the combined PSF and resulting MTF. Fortunately that combined PSF can be approximated quite well with a Gaussian like PSF. Theoretically we might be able to design an even more accurate PSF, but in practice we are confronted with too many variables (such as defocus) to deal with, and the optimal PSF would not be that useful anymore. Only in a laboratory situation would it make sense to devise an optimal PSF for a given lens/camera/subject combination.
I see your points.
I still think it would be interesting, perhaps even useful, to estimate the true total PSF for a variety of aperture, distance/focus, etc settings. Perhaps this set of PSFs can be analyzed to extract a constant component (stuff that does not change, such as AA-filter), along with a convolved variable component (function of aperture, focus, distance). Separating a constant function from a variable component in a set of convoluted responses sounds complex, but perhaps the frequency-domain (multiplication) is suitable?
This could (among other things) allow us to predict the system PSF of a Sigma 10-20mm mounted on a Canon 50D, based only on a) measurements of a 10-20mm mounted on a Nikon D7000, and b)measurements of a 17-55mm on a Canon 50D. It might also be easier to predict the performance of a given lense on future, high-resolution sensors before they are available (such as, until recently, the D800).
Do you have example images showing the visual difference between using the true PSF (or an accurate estimate) vs using a quickly estimated gaussian for deconvolution?
-h
-
Of course we can also use it to determine the actual resolution difference between the D800 and D800E after using the optimal radius settings for capture sharpening in various applications ...
I imagine that the "optimal radius setting" will be a function of visible/annoying image noise, meaning that "ISO" setting, scene, print size/viewing distance, curves/levels usage and viewer preferences all affect the optimal radius. I don't think there will be a number saying something ala: "the D800 has 0.8x the resolution of the D800E after optimal sharpening".
If anything, I would guess that in somewhat ideal settings, the D800 resolution would be approx. 1.0x the D800E resolution, both optimally sharpened. But in e.g. low-light conditions, the number would be smaller than 1.0.
-h
-
I still think it would be interesting, perhaps even useful, to estimate the true total PSF for a variety of aperture, distance/focus, etc settings. Perhaps this set of PSFs can be analyzed to extract a constant component (stuff that does not change, such as AA-filter), along with a convolved variable component (function of aperture, focus, distance). Separating a constant function from a variable component in a set of convoluted responses sounds complex, but perhaps the frequency-domain (multiplication) is suitable?
This could (among other things) allow us to predict the system PSF of a Sigma 10-20mm mounted on a Canon 50D, based only on a) measurements of a 10-20mm mounted on a Nikon D7000, and b)measurements of a 17-55mm on a Canon 50D. It might also be easier to predict the performance of a given lense on future, high-resolution sensors before they are available (such as, until recently, the D800).
Hi h,
I agree that there are uses for the characterization of the individual optical components, but I'll move the discussion about such laboratory conditions to another thread in a more appropriate place on LuLa.
Do you have example images showing the visual difference between using the true PSF (or an accurate estimate) vs using a quickly estimated gaussian for deconvolution?
A comparison between a quick-and-dirty and a more fine-tuned kernel will be addressed in another thread where I'll introduce my tool.
Cheers,
Bart
-
I imagine that the "optimal radius setting" will be a function of visible/annoying image noise, meaning that "ISO" setting, scene, print size/viewing distance, curves/levels usage and viewer preferences all affect the optimal radius. I don't think there will be a number saying something ala: "the D800 has 0.8x the resolution of the D800E after optimal sharpening".
Hi h,
It's 'simpler' than that. The deterioration of the perfect (because it exceeds the resolving power of our optics and sensor) slanted edge by the optical system (and Raw converter) can be expressed in an Edge Spread Function (ESF). This will allow to calculate a PSF. Deconvolving the blurred edge with that PSF should reconstruct the original edge (within practical limits). The Gaussian 'radius' or sigma that best describes the 2D PSF is the optimal radius to use in capture sharpening. The only thing I'm not 100% sure about is whether the radius parameter in e.g. Lightroom, is identical to the sigma of a Gaussian. Afterall, a Gaussian doen't have a real radius because it has infinite dimensions.
If anything, I would guess that in somewhat ideal settings, the D800 resolution would be approx. 1.0x the D800E resolution, both optimally sharpened. But in e.g. low-light conditions, the number would be smaller than 1.0.
Indeed, since the sampling density is identical, the Nyquist frequency is also identical. Only the amplitude of the MTF at Nyquist will be different (before sharpening), and therefore micro-contrast at the limiting resolution, and aliasing. Proper sharpening will level the playing field.
Cheers,
Bart
-
Hi h,
It's 'simpler' than that. The deterioration of the perfect (because it exceeds the resolving power of our optics and sensor) slanted edge by the optical system (and Raw converter) can be expressed in an Edge Spread Function (ESF). This will allow to calculate a PSF. Deconvolving the blurred edge with that PSF should reconstruct the original edge (within practical limits). The Gaussian 'radius' or sigma that best describes the 2D PSF is the optimal radius to use in capture sharpening. The only thing I'm not 100% sure about is whether the radius parameter in e.g. Lightroom, is identical to the sigma of a Gaussian. Afterall, a Gaussian doen't have a real radius because it has infinite dimensions.
Is the edge spread function equivalent to a 1-d step function? If so, the relation between a 1-d step-function and a 1-d impulse (and corresponding system responses) is well-known, and probably extends nicely to the 2-d case?
http://en.wikipedia.org/wiki/Step_response#Linear_dynamical_system
If the total system PSF deviates significantly from a gaussian, deconvolving using a gaussian should produce "significant" errors, irrespective of choice of sigma.
-h
-
Thanks Ray for suggesting this test. I was about to suggest the same. Unless you provide such a double blind test no human being can escape from the confirmation bias we all suffer from ;)
Yes, Hans. I think we're both in agreement here. Confirmation bias is a huge problem in science, in general. My suggestion that Michael hand over the RAW D800 file to Jeff Schewe to do his best to match the resolution of Michael's D800 print, is probably flawed.
Jeff is probably also biased in favour of 'no AA filter'. To get him to do his best, Michael might have to offer an inducement along the lines, "If you succeed in restoring the lost detail in this D800 image, Jeff, so that it looks just as sharp and detailed on your print as it does on my print of the D800E image, I shall reward you by presenting you with a free Nikon D800E." (I have to tread carefully here. This is tongue in cheek. ;D )
What might be a better idea, is for Michael to send the RAW D800 file, together with the printer profile that Michael used to produce his own print from the D800E file, to Bernard or Bart.
Bernard would do his best to prove to himself, if no-one else, that he had made the right decision in choosing the D800 over the D800E, and Bart would do his best to demonstrate that his knowledge of Point Spread Function and Deconvolution processes are unparalleled.
However, even if Bart, or Bernard, or both of them, were to succeed in matching the resolution and sharpness of Michael's D800E image, from which he made the D800E print, the matter would still not be settled.
We have to consider the amount of time spent, and the difficulty incurred, in processing that D800 file to match the D800E file. It might be huge, involving arcane processes not readily available to the general public, and involving considerable expertise, whereas Michael's general processing and sharpening of the D800E image may be basic, standard, and easy. That in itself may be considered as an advantage.
-
Ray:
What about using an automated algorithm to match one output to the other, using a limited (sensible) set of parameters? If we are to believe that the only difference between the D800 and the D800E is the presence of an (effective) AA filter, then one would expect to find a relatively simple relationship between their raw files to be reasonably accurate for identical conditions.
Find an estimate to this relationship, and you can debate its form and the remaining residue.
-h
-
Yes, Hans. I think we're both in agreement here. Confirmation bias is a huge problem in science, in general. My suggestion that Michael hand over the RAW D800 file to Jeff Schewe to do his best to match the resolution of Michael's D800 print, is probably flawed.
<snip>
We have to consider the amount of time spent, and the difficulty incurred, in processing that D800 file to match the D800E file. It might be huge, involving arcane processes not readily available to the general public, and involving considerable expertise, whereas Michael's general processing and sharpening of the D800E image may be basic, standard, and easy. That in itself may be considered as an advantage.
What about a single blind test where Michael prepares X number of pictures of different scenes and shoot each with both cameras (identical setup on tripod). Use identical pp except optimized capture sharpening for each shot. Then print them in the largest size possible at 300PPI and mark them with encoded names on the back that only he knows about. Then a second person brings these photos into a gallery and ask a number of people to review the prints and say which is the sharpest looking and which has the best details (maybe even at close inspection and at a defined viewing distance). Then gather the results and do a statistical analysis on the data and determine if there is a statistically significant difference in either case (close inspection and normal (defined) viewing distance.
Well, I guess this is too much effort, but it could be fun to see the results.
-
What about a single blind test where Michael prepares X number of pictures of different scenes and shoot each with both cameras (identical setup on tripod).
...
Sure, if someone tapes up the cameras first so that Michael doesn't know if he's shooting with the D800 or D800E.
-
... We have to consider the amount of time spent, and the difficulty incurred, in processing that D800 file to match the D800E file. It might be huge, involving arcane processes not readily available to the general public, and involving considerable expertise, whereas Michael's general processing and sharpening of the D800E image may be basic, standard, and easy. That in itself may be considered as an advantage.
Very good points, Ray.
Furthermore, given that we are talking about reverse engineering, its algorithm has to be 100% correct in order to reduce the initial D800e advantage to zero. Again, I am not a scientist or software developer, so I do not know how possible it is to achieve 100% reverse engineering (the only bragged-about case I know of is in Iran currently ;))
My further assumptions would be that, for a perfect reverse engineering (i.e., deconvolution sharpening), one needs to know the exact algorithm of AA blurring, and that would be only Nikon engineers, I presume.
Also, even if the algorithm is 100% successful (in restoring lost detail) it is not inconceivable that there could be some "collateral damage" along the way, i.e., unwanted side effects or artifacts.
In other words, unless all the conditions above are met with 100% success, the initial advantage of D800e would remain, however imperceptible it might be, in print or otherwise.
-
To end the speculation, let's look at an actual conversion I made from 2 files, one from the D800, and one from the D800E.
Michael made sure that the test chart was perfectly focused, selecting the best from a focus series for each camera. The same 85mm f/1.4 lens was used at f/4 on both cameras.
For this pair I used Capture One Pro v6.4, set to a linear filmcurve, shifted the exposure up a bit, and I only corrected for Chromatic Aberration (to compare with other converters, more on that later), and applied no sharpening at all. The sharpening was skipped to avoid speculation about the effect of different types of sharpening, and it wouldn't have made any difference to the actual limiting resolution anyway. It also allows to determine the best capture sharpening settings.
Here are the cropped results, 8-b/ch JPEGs at 100% quality converted to sRGB, at 100% zoom without resampling.
Remember, these are without sharpening.
First the D800: (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800_0406_CO64.jpg)
Click here (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800_0406_CO64.png) for a 16-bit/channel PNG version, AdobeRGB as source, Gamma 2.20
Next the D800E: (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800E_0316_CO64.jpg)
Click here (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800E_0316_CO64.png) for a 16-bit/channel PNG version, Adobe RGB as source, Gamma 2.20
I'll elaborate a bit more on the relevant parts in subsequent posts, but I'll already tell that the D800 shows a limiting resolution of 93.9 cycles/mm or 92.0% of Nyquist, and the D800E shows a limiting resolution of 94.9 cycles/mm or 92.9% of Nyquist. In other words, their limiting resolutions are virtually identical which makes sense because the sampling density is also identical. The OLPF in the D800 only reduced the amplitude of the MTF, which helped to reduce aliasing and demosaicing artifacts.
Cheers,
Bart
-
... the D800 shows a limiting resolution of 93.9 cycles/mm or 92.0% of Nyquist, and the D800E shows a limiting resolution of 94.9 cycles/mm or 92.9% of Nyquist. ...
So a 1% difference in resolution: whether that matters probably depends on whether you are part of the "photographic 1%", or the 99%.
-
And here are some details, zoomed to 200%, with a red circle (92 pixel diameter) drawn at the Nyquist frequency, and a green circle at the limiting resolution. Any detail within the red circle is pure aliasing, the sensor cannot resolve detail which is smaller than a single sensel.
The D800 :
(http://bvdwolf.home.xs4all.nl/temp/LuLa/D800_0406_CO64_Nyquist.png) (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800_0406_CO64_100px.png)
And the D800E :
(http://bvdwolf.home.xs4all.nl/temp/LuLa/D800E_0316_CO64_Nyquist.png) (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800E_0316_CO64_99px.png)
The false color artifacting which is common for Bayer CFA sensors, the Nikons are no exception, can be removed by Capture One Pro's Moiré tool, but the luminance aliasing is too complex to solve. Fortunately we're looking at a torture test, so it's not too common to face such challenging subjects. Having said that, it does send a warning to look out for potential trouble.
It looks like the OLPF of the D800 is very well designed, it will reduce most of the common difficulties, false color artifacting, jaggies, and moiré.
Cheers,
Bart
-
Bart,
How do you explain those areas in D800 I pointed to? Some kind of green/orange amebas, blots? Those do not exist in D800e.
-
Bart
I wish you could design a pseudo random pattern and then compare the 2 cameras
Marc
-
How do you explain those areas in D800 I pointed to? Some kind of green/orange amebas, blots? Those do not exist in D800e.
Hi Slobodan,
Those are areas where, probably due to the specific alignment with the pixel grid, the demosaicing failed. They all happen at 45 degrees off normal, which suggests that the angle and specific level of detail triggers that behavior. Due to the target being so ordered and predictable to human vision, anything unpredictable will stand out like a sore thumb. In an ordinary image it would not happen or not get noticed.
To eliminate the possiblity of a mapped out (cluster of) sensels being the cause, one can verify in another Raw converter, which I did. They all make different trade-offs, and goof-ups. Capture One tends to push the hunt for detail a bit further than e.g. Lightroom/ACR, and on rare occasions it fails.
Cheers,
Bart
-
Thank you Bart and Michael for these tests.
From my point of view, they pretty much close the debate. The results you have given us serve to separate out actual resolution from the "impression" of sharpness that can be created by aliasing artifacts. There is apparently virtually no real detail increase as a result of eliminating the aliasing filtration.
Now I'm very interested in your investigations into optimal sharpening routines...
-
I wish you could design a pseudo random pattern and then compare the 2 cameras
Hi Marc,
Nah, random targets are too easy ;)
Just kidding. Random targets are indeed also used for resolution testing (uniform/white noise and Fourier analysis), but they require a lot of number crunching to get some sensible data out of them. An ordered target can already be evaluated with simple means, sometimes even just visual inspection.
Nevertheless, keep in mind that the target tests the extremes, with a good lens, perfectly focused. Every deviation will show. Real life is often more forgiving.
Cheers,
Bart
-
... Those are areas where, probably due to the specific alignment with the pixel grid, the demosaicing failed. They all happen at 45 degrees off normal, which suggests that the angle and specific level of detail triggers that behavior. Due to the target being so ordered and predictable to human vision, anything unpredictable will stand out like a sore thumb. In an ordinary image it would not happen or not get noticed.
To eliminate the possiblity of a mapped out (cluster of) sensels being the cause, one can verify in another Raw converter, which I did. They all make different trade-offs, and goof-ups. Capture One tends to push the hunt for detail a bit further than e.g. Lightroom/ACR, and on rare occasions it fails...
Then again, how come that D800e does not have any?
-
Hi Marc,
Nah, random targets are too easy ;)
Just kidding. Random targets are indeed also used for resolution testing (uniform/white noise and Fourier analysis), but they require a lot of number crunching to get some sensible data out of them. An ordered target can already be evaluated with simple means, sometimes even just visual inspection.
Nevertheless, keep in mind that the target tests the extremes, with a good lens, perfectly focused. Every deviation will show. Real life is often more forgiving.
Bart,
How about the Dead Leaves (http://www.dxo.com/var/dxo/storage/fckeditor/File/embedded/2009_EI_TextureSharpness.pdf) model? Have you investigated it?
Regards,
Bill
-
Then again, how come that D800e does not have any?
Higher contrast at those spatial frequencies?
Cheers,
Bart
-
Bart,
How about the Dead Leaves (http://www.dxo.com/var/dxo/storage/fckeditor/File/embedded/2009_EI_TextureSharpness.pdf) model? Have you investigated it?
Regards,
Bill
Seems like a very good test!
Marc
ps in my audio background if tests don't correlate to what you are hearing with enough diligence you found the flaw in the test and corrected it. My eye says that there is a sharper image without an AA filter and I rarely see false information. Perhaps it is this dead leaf test that the photographic industry is missing?
-
How about the Dead Leaves (http://www.dxo.com/var/dxo/storage/fckeditor/File/embedded/2009_EI_TextureSharpness.pdf) model? Have you investigated it?
Hi Bill,
That's an interesting article which adderesses some of the concerns that I have about the possible benefits of the D800E when considering real-world images, across the entire tonal range, from highlights to deep shadows, where certain parts of the image may contain fine texture at different levels of contrast and different levels of exposure.
I'm reminded of conparisons I made some years ago between telephoto lenses with and without teleconverter. Sometimes, there seemed to be no benefit in using the teleconverter, or hardly any benefit. But sometimes there was a clear benefit. The results seemed to depend on the contrast and general clarity of the subject. With the use of the teleconverter, well-lit, contrasty texture in the subject seemed to be captured with noticeably more detail, due to the greater magnification, whereas faint and low contrast detail was more easily lost.
The analogy I'm drawing here is that it seems possible that certain low-contrast fine detail and texture, in certain parts of an image, at certain exposure levels, may be captured by the D800E, but irretrievably lost when using a D800.
ps. I forgot to add, when photographing a resolution test target containing black and white lines, the telephoto lens with converter attached always provided greater resolution than the lens without converter.
-
Attention - these charts were run with the wrong gamma setting in Imatest
Please see this post for corrections:
http://www.luminous-landscape.com/forum/index.php?topic=65927.140 (http://www.luminous-landscape.com/forum/index.php?topic=65927.140)
To end the speculation, let's look at an actual conversion I made from 2 files, one from the D800, and one from the D800E.
Michael made sure that the test chart was perfectly focused, selecting the best from a focus series for each camera. The same 85mm f/1.4 lens was used at f/4 on both cameras.
Cheers,
Bart
Thanks Bart!
That's about the most useful data I've seen to date on any forum. I've taken the liberty of doing a quick and dirty run of the slant edge patterns on your downloaded png charts through Imatest Studio (version 3.9 Beta). These were completely unmolested (no sharpening etc.). I suspect Erik may be doing the same about now? I didn't have the true chart contrast to enter in the settings or know the quality of the edge transistions on the (printed?) chart (but they looked pretty clean). I did enter gamma at 2.20. There is a screen capture of the Imatest settings page at the end. If anyone has suggested changes to the settings, it's easy enough to renun the charts.
The horizontal and vertical edge profile are nearly identical between the D800 and the D800E. However as expected, the D800 shows lower contrast at higher spatial freqencies near Nyquist. I couldn't run a noise analysis because Imatest complained that the area was too small (this is probably not the best chart for evaluating noise anyway). The chromatic aberration tests are also interesting.
Since I'm not using Imatest for profit, I'll leave the necessary props for Normen Koren and company:
http://www.imatest.com/ (http://www.imatest.com/)
My only remaining concern is how much noise may be apparent in images from the D800 after various sharpening routines are applied to boost the low-contrast, high spatial frequency portion of the MTF plot to match the D800E?
D800 MTF horizontal slant edge
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-and-D800E-Imatest-Results/i-TNNhH58/0/L/D800-MTF-Cy-per-mm-Horiz-Bart-L.jpg)
D800 MTF vertical slant edge
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-and-D800E-Imatest-Results/i-fvs2sw8/0/L/D800-MTF-Cy-per-mm-Vertical-L.jpg)
D800E MTF horizontal
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-and-D800E-Imatest-Results/i-3x9N7mh/0/L/D800E-MTF-Cy-per-mm-Horiz-Bart-L.jpg)
D800E MTF vertical
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-and-D800E-Imatest-Results/i-MHwk2D8/0/L/D800E-MTF-Cy-per-mm-Vertical-L.jpg)
D800 chromatic aberration horizontal
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-and-D800E-Imatest-Results/i-Ss5fsXF/0/L/D800-CA-Horiz-Bart-van-der-L.jpg)
D800 chromatic aberration vertical
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-and-D800E-Imatest-Results/i-3pdVd4S/0/L/D800-CA-Vertical-Bart-van-der-L.jpg)
D800E chromatic aberration horizontal
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-and-D800E-Imatest-Results/i-Qdtx5mx/0/L/D800E-CA-Horiz-Bart-van-der-L.jpg)
D800E chromatic aberration vertical
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-and-D800E-Imatest-Results/i-xZqtr6D/0/L/D800E-CA-Vertical-Bart-van-der-L.jpg)
Imatest settings for all tests
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-and-D800E-Imatest-Results/i-hMQVTjx/0/L/D800-D800E-Imatest-Settings-L.jpg)
Hope others find this useful---or at least entertaining!
Kind regards,
Bob
Edit: Correction---this was done in Imatest version 3.9 Beta, not 3.8 Beta. Imatest have not yet updated the version reported on the charts (at least not in Imatest Studio).
-
To end the speculation, let's look at an actual conversion I made from 2 files, one from the D800, and one from the D800E.
Michael made sure that the test chart was perfectly focused, selecting the best from a focus series for each camera. The same 85mm f/1.4 lens was used at f/4 on both cameras.
Bart,
Very interesting results. The Moire is worse than I would have imagined, but then I had never seen your target without low pass filtering. I eagerly await your additional analysis. Where did you get the images? Did Michael post them publicly?
Regards,
Bill
-
Hi,
A great lot of thanks to Michael and also for your analysis. Are those raw images available for download?
Best regards
Erik
To end the speculation, let's look at an actual conversion I made from 2 files, one from the D800, and one from the D800E.
Michael made sure that the test chart was perfectly focused, selecting the best from a focus series for each camera. The same 85mm f/1.4 lens was used at f/4 on both cameras.
For this pair I used Capture One Pro v6.4, set to a linear filmcurve, shifted the exposure up a bit, and I only corrected for Chromatic Aberration (to compare with other converters, more on that later), and applied no sharpening at all. The sharpening was skipped to avoid speculation about the effect of different types of sharpening, and it wouldn't have made any difference to the actual limiting resolution anyway. It also allows to determine the best capture sharpening settings.
Here are the cropped results, 8-b/ch JPEGs at 100% quality converted to sRGB, at 100% zoom without resampling.
Remember, these are without sharpening.
First the D800: (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800_0406_CO64.jpg)
Click here (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800_0406_CO64.png) for a 16-bit/channel PNG version, AdobeRGB as source, Gamma 2.20
Next the D800E: (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800E_0316_CO64.jpg)
Click here (http://bvdwolf.home.xs4all.nl/temp/LuLa/D800E_0316_CO64.png) for a 16-bit/channel PNG version, Adobe RGB as source, Gamma 2.20
I'll elaborate a bit more on the relevant parts in subsequent posts, but I'll already tell that the D800 shows a limiting resolution of 93.9 cycles/mm or 92.0% of Nyquist, and the D800E shows a limiting resolution of 94.9 cycles/mm or 92.9% of Nyquist. In other words, their limiting resolutions are virtually identical which makes sense because the sampling density is also identical. The OLPF in the D800 only reduced the amplitude of the MTF, which helped to reduce aliasing and demosaicing artifacts.
Cheers,
Bart
-
I'll elaborate a bit more on the relevant parts in subsequent posts, but I'll already tell that the D800 shows a limiting resolution of 93.9 cycles/mm or 92.0% of Nyquist, and the D800E shows a limiting resolution of 94.9 cycles/mm or 92.9% of Nyquist. In other words, their limiting resolutions are virtually identical which makes sense because the sampling density is also identical. The OLPF in the D800 only reduced the amplitude of the MTF, which helped to reduce aliasing and demosaicing artifacts.
Excellent test and conclusions Bart. It seems clear that the D800E doesn't hold more detail, just higher microcontrast that should be recoverable through smart sharpening.
If you were to pick one of the models, which one would you choose with the available information?
Regards
-
Thank you Bart and Michael for these tests.
From my point of view, they pretty much close the debate. The results you have given us serve to separate out actual resolution from the "impression" of sharpness that can be created by aliasing artifacts. There is apparently virtually no real detail increase as a result of eliminating the aliasing filtration.
Now I'm very interested in your investigations into optimal sharpening routines...
You're all welcome, it wouldn't have been possible without Micheal's input. The benefit of a higher modulation near the limiting resolution is in better microdetail with low contrast subject. An example would be woodgrain in wooden objects. However, since there is no free lunch, aliasing can get in the way with sharp edges and repetitive structures. The false color (chroma) aliasing from Bayer CFA demosaicing can be dealt with with a good Raw converter, but it remains a challenge for the Raw converter, and it will struggle with luminosity aliasing.
So the advise remains to choose one's weapons for the task at hand, or go with the safer option if you want to avoid a lot of manual postprocessing.
I'm working on the optimal sharpening tool, but I just found an issue that needs to be resolved first.
Cheers,
Bart
-
How about the Dead Leaves (http://www.dxo.com/var/dxo/storage/fckeditor/File/embedded/2009_EI_TextureSharpness.pdf) model? Have you investigated it?
Hi Bill,
Yes, I've read about it, but have not come to a conclusion for myself yet. One of the issues that photographers face in practice are things like unexpected artifacts which then consume a lot of time (=money) to deal with. What I like is that it gives an impression about lower contrast detail resolution, but I'm not so sure yet that the 'dead leaves' target is e.g. enough of a challenge to push the Raw converter over the edge of its demosaicing capabilities in view of aliasing.
It's interesting to see how different Rawconverters tackle that challenge, and I have some very interesting initial results coming from RawTherapee which also allows to influence the "amaze" algorithm's false color suppression. The more 'amazing' result is that it resolves even closer to the Nyquist frequency ... Luminance aliasing remains a limiting factor though.
Cheers,
Bart
-
...
For this pair I used Capture One Pro v6.4, set to a linear filmcurve, shifted the exposure up a bit, and I only corrected for Chromatic Aberration (to compare with other converters, more on that later), and applied no sharpening at all. The sharpening was skipped to avoid speculation about the effect of different types of sharpening, and it wouldn't have made any difference to the actual limiting resolution anyway. It also allows to determine the best capture sharpening settings.
...
Bart
Hello Bart,
Thank you and Michael for this interesting test.
I ask myself how different the outcome would be with different Raw converters ...? Especially the software from Nikon self : Nikon Capture NX2.32
Also would like to know if their moiré tool adds to the image quality.
-
Thanks Bart!
That's about the most useful data I've seen to date on any forum. I've taken the liberty of doing a quick and dirty run of the slant edge patterns on your downloaded png charts through Imatest Studio (version 3.8 Beta). These were completely unmolested (no sharpening etc.). I suspect Erik may be doing the same about now? I didn't have the true chart contrast to enter in the settings or know the quality of the edge transistions on the (printed?) chart (but they looked pretty clean). I did enter gamma at 2.20. There is a screen capture of the Imatest settings page at the end. If anyone has suggested changes to the settings, it's easy enough to renun the charts.
The horizontal and vertical edge profile are nearly identical between the D800 and the D800E. However as expected, the D800 shows lower contrast at higher spatial freqencies near Nyquist. I couldn't run a noise analysis because Imatest complained that the area was too small (this is probably not the best chart for evaluating noise anyway). The chromatic aberration tests are also interesting.
Hi Bob,
Thanks for doing the Imatest run, I didn't have the time to do it myself yet. The only change to the settings I could suggest is to replace the crop size setting with the full image size of 7416 x 4916 (which is what some converters squeeze out of the Raw) or the formal 7360 x 4912 pixels.
The various Raw converters I've tried sofar do add a reddish tint to the D800E conversions in varying degrees, interpreted by Imatest as CA.
My only remaining concern is how much noise may be apparent in images from the D800 after various sharpening routines are applied to boost the low-contrast, high spatial frequency portion of the MTF plot to match the D800E?
Yes, that is still open for debate, also depending on whether one uses ETTR and low ISO settings. It often helps to do a mild noise reduction before applying significant sharpening, and to mask and blend the sharpening with a bias towards luminosity sharpening. Personally I don't worry about the noise too much, but then I'm a low ISO shooter, and things look better inprint that when pixel peeping on a monitor.
Cheers,
Bart
-
Very interesting results. The Moire is worse than I would have imagined, but then I had never seen your target without low pass filtering. I eagerly await your additional analysis. Where did you get the images? Did Michael post them publicly?
A great lot of thanks to Michael and also for your analysis. Are those raw images available for download?
Bill, Erik,
They're Michael's files, I was fortunate enough to receive a copy. It's up to Michael how he wants to keep control over them.
Cheers,
Bart
-
Excellent test and conclusions Bart. It seems clear that the D800E doesn't hold more detail, just higher microcontrast that should be recoverable through smart sharpening.
If you were to pick one of the models, which one would you choose with the available information?
Hi Guillermo,
Given my personal shooting preferences (including architecture and product photography, but also landscapes and macro), the safest option would be the D800. I'm a bit allergic to aliasing (because it reveals that the fundamental rules of DSP were not taken serious enough), but I've also spent more effort in perfecting my sharpening arsenal of tools than most others have, so restoring sharpness is not that hard. Mind you, even a camera with an AA-filter can fall victim to aliasing, it's usually just a somewhat more benign manifestation, easier to repair.
Cheers,
Bart
-
Hello Bart,
Thank you and Michael for this interesting test.
I ask myself how different the outcome would be with different Raw converters ...? Especially the software from Nikon self : Nikon Capture NX2.32
Also would like to know if their moiré tool adds to the image quality.
Hi kers,
Good questions, and yes there are differences. One of the problems is that they do not only give different results, but also different workflows, and there is more to image quality than resolution versus moiré alone. So it depends a bit on one's goals, ultimate image quality or good enough quality but a more efficient process.
Another thing is that support for the D800/D800E was only recently added to several Raw converters, maybe there is some room for improvement as these solutions mature. I'll try and post some results from different converters if I can find some time, but I don't have Nikon's Capture NX converter.
Cheers,
Bart
-
The D800E together with photoacute (or some other multi-shot super-resolution software) could be interesting for those chasing the highest resolution that $3000 can buy, and unable/unwilling to do stitching:
http://www.photoacute.com/
This is an excerpt from a mail-exchange I had with them:
Dear xxx,
Thank you for your interest.
Removal of AA-filter should drastically improve the gain of our software. But, a special profile will be required for processing images taken with a camera with removed filter.
Does your Canon 7D already have AA-filter removed?
-h
-
If we consider the entire capture path of light and especially all possible aberrations, including atmospheric instabilities and birefringent layers with breaking index etc…, did we ever compute the minimum sensible sensel size that would benefit from additional AA?
It appears that shooting this puppy under any circumstances other than absolute perfect stability, which probably reflects 99.99% of all cases, will completely negate any adverse effects that no AA might have. Then in those 0.01% of the cases most people will hardly ever have problems. Come to think of it: in those 0.01% cases, where one apparently has time to create perfect circumstances, one also can do a double exposure and the blending will be the perfect AA result by sheer imperfections between the 2 shots, but with the additional benefit of extra depth.
Bart, any change that you can make the RAW files available?
-
Bart, any change that you can make the RAW files available?
Hi Oscar,
They're not my files. I'll leave it to Michael to control the distribution of his files as he sees fit.
Cheers,
Bart
-
Bart
I wish you could design a pseudo random pattern and then compare the 2 cameras
Marc
I would agree. Seems to me the test shows moire wiped out any possibility of resolution gain. Maybe such a target can't be made as I can see the logic and use of this target is very cool, but seems it gives the d800 an advantage over the d800e in a circumstance that a shooter of the e will avoid if at all possible. (I have an e on order).
I guess my skepticism is it appears the AA filter was always blamed for loss of quality, suddenly it isn't any more. OF course, the other thing that I"m curious about is the camera doesn't have an "aa" filter per se, the filter they are using isn't exactly the same as having no AA filter. I'm not sure why you bend the light only to "bend it back" ... why not just leave it off? I really can't understand why this concept is better.
I'm not necessarily disagreeing with the test, as I've felt for a long time that well done sharpening seems to compensate pretty well for issues like an AA filter or diffraction. It just seems the target introducing moire biases the results.
-
I guess my skepticism is it appears the AA filter was always blamed for loss of quality, suddenly it isn't any more. OF course, the other thing that I"m curious about is the camera doesn't have an "aa" filter per se, the filter they are using isn't exactly the same as having no AA filter. I'm not sure why you bend the light only to "bend it back" ... why not just leave it off? I really can't understand why this concept is better.
Probably because it keeps the distance between the mount and sensor exactly the same between the D800 and D800e?
Cheers,
Bernard
-
Wayne,
Regarding AA-filter, I don't know. In theory it is needed but lots of very knowledgeable folks shoot MF digital with excellent technique without running into problems. If an P65+ or IQ180 works well than the D800E having much smaller pixels would work equally well.
Regarding the reason that Nikon did not remove the AA-filter but replaced it with a design that effectively is said the anull the effect may be that the AA-filter is an optical element that may affect astigmatism in wide angle lenses. It is reasonable to presume that new Nikon lens designs are taking the AA-filter into account, and removing the OLP filter from the optical path may result in increased astigmatism on certain lenses.
Lloyd Chambers has compared the Leica S2 to the Nikon D800 shooting an architecture subject and the Leica image has Moiré all over the place, but most users of the S2 seem to have little problem with the S2 and Moiré.
Best regards
Erik
I would agree. Seems to me the test shows moire wiped out any possibility of resolution gain. Maybe such a target can't be made as I can see the logic and use of this target is very cool, but seems it gives the d800 an advantage over the d800e in a circumstance that a shooter of the e will avoid if at all possible. (I have an e on order).
I guess my skepticism is it appears the AA filter was always blamed for loss of quality, suddenly it isn't any more. OF course, the other thing that I"m curious about is the camera doesn't have an "aa" filter per se, the filter they are using isn't exactly the same as having no AA filter. I'm not sure why you bend the light only to "bend it back" ... why not just leave it off? I really can't understand why this concept is better.
I'm not necessarily disagreeing with the test, as I've felt for a long time that well done sharpening seems to compensate pretty well for issues like an AA filter or diffraction. It just seems the target introducing moire biases the results.
-
I guess my skepticism is it appears the AA filter was always blamed for loss of quality, suddenly it isn't any more.
Seems to me that at least some people have been arguing for quite some time that well-designed AA filters preserve quality, not lower it.
-
I would agree. Seems to me the test shows moire wiped out any possibility of resolution gain. Maybe such a target can't be made as I can see the logic and use of this target is very cool, but seems it gives the d800 an advantage over the d800e in a circumstance that a shooter of the e will avoid if at all possible. (I have an e on order).
I guess my skepticism is it appears the AA filter was always blamed for loss of quality, suddenly it isn't any more. OF course, the other thing that I"m curious about is the camera doesn't have an "aa" filter per se, the filter they are using isn't exactly the same as having no AA filter. I'm not sure why you bend the light only to "bend it back" ... why not just leave it off? I really can't understand why this concept is better.
I'm not necessarily disagreeing with the test, as I've felt for a long time that well done sharpening seems to compensate pretty well for issues like an AA filter or diffraction. It just seems the target introducing moire biases the results.
Also I think it might polarize the light a bit causing it to hit the micro lenses more perpendicular, just a guess?
Marc
-
One of the reasons I propose a random target is that I have a gut feeling that is contrary to Bart's and Erick's knowledge.
I think that to have these "artifacts" one needs a pattern (Bayer) on top of a pattern (suit or building windows) at near the Nyquist frequency.
I don't think that a Bayer pattern on a random structure (nature) will cause artifacts. The best way to prove a theory is to try and prove yourself wrong.
So I propose taking Bart's chart and removing the lines that are not prime leaving lines 2,3,5,7,11,13 etc and re-shoot the test. I might very well be wrong but I'd like to see it with my own eyes!
Marc
-
I guess my skepticism is it appears the AA filter was always blamed for loss of quality, suddenly it isn't any more.
Hi Wayne, Bernard,
And therein lies the crux. Yes there is some loss from the AA-filter at low contrast. An MTF curve shows that in general contrast will reduce as the detail gets smaller (higher spatial frequencies). When the subject matter is already of low contrast, then the resulting contrast will be lost to the noise floor and quantization accuracy (14-bit/channel in Raw, and 8 or 16-b/ch after conversion).
But with a bit of subject contrast, and a lens with good glare control and good enough MTF helps, there will be enough micro-contrast left to come close to resolving all the way up to the physical limit, the Nyquist frequency. That Nyquist frequency is determined by the sensor's sampling density (the sensel pitch).
Probably because it keeps the distance between the mount and sensor exactly the same between the D800 and D800e?
It's the sensel pitch that makes the difference. When the same optical projection is sampled at the same sampling pitch, there will be no difference in resolution (usually expressed as a spatial frequency in cycles/mm). The only difference will be in the amplitude of the signal at the same spatial frequencies. The AA-filter is designed to reduce the amplitude as the spatial frequencies approach the Nyquist frequency, in order to reduce the signal amplitudes beyond Nyquist. It's those spatial frequencies beyond Nyquist that cause aliasing, so when the amplitudes of those details (that are too fine to resolve reliably anyway) are reduced, then there will be no aliasing.
When the same optical projection is sampled at a lower density, then the resolution will be lower. The only way to compensate for that is to use a longer focal length to magnify the projected image. When we then add more sensels to increase the Field of View again, then there will again be no difference in resolution (because the output magnification can be proportionally reduced). There will be a better signal amplitude (the '3-D' look). And that's exactly the difference with Medium Format cameras and backs. They tend to have a larger sensel pitch, and need a longer focal length to cover the larger image circle with more sensels.
OF course, the other thing that I"m curious about is the camera doesn't have an "aa" filter per se, the filter they are using isn't exactly the same as having no AA filter. I'm not sure why you bend the light only to "bend it back" ... why not just leave it off? I really can't understand why this concept is better.
I agree that bending the light only to "bend it back" doesn't seem like an efficient operation. I'm not so sure that that is what is actually happening, I also don't understand how it could function as descibed in their marketing diagrams. So there is possibly something else going on, but Nikon decided to not change the number of plane parallel surfaces and thicknesses in the lightpath.
That could help to make optical lens designs which incorporate the existence of those surfaces and layers, and deliver a better image quality, they become part of the optical design. BTW, it probably one of the reasons that Leica didn't use AA-filters either because, added to the proximity of the exit pupil to the sensor surface, the lenses are designed to deliver an image to film, without additional optical elements in the lightpath.
I'm not necessarily disagreeing with the test, as I've felt for a long time that well done sharpening seems to compensate pretty well for issues like an AA filter or diffraction. It just seems the target introducing moire biases the results.
Well, that's a matter of opinion. The target doesn't introduce(!) moiré, let's be clear about that. It only allows to make it visible, which opens the opportunity to devise a remedie. I'd rather be aware of a potential issue, than be confronted in the midst of (or even worse, after) an assignment.
Moiré can only manifest itself when 2 sampling grids are out of sync (by rotation or frequency). The sensor array is a given sampling grid, so if we can avoid shooting repetitive structures then we won't have 2 grids that are out of sync. When we reduce the amplitude of one of the grids (the projected image), the moiré will be less visible (e.g. diffraction or de-focus, or an AA-filter). When we use a longer focal length to magnify the subject details the aliasing will be reduced because of the relatively denser sampling of the projected image details (but also our FoV is reduced, so we need to stitch or use a larger sensor to compensate).
An interesting exercise would be for a D800E owner to shoot the target with a good lens at say f/4 or f/5.6 and, without changing focus, at f/16 or f/22, which may be what a landscape shooter would like to do to achieve the DOF required. The higher modulation of the D800E will help, and the aliasing of sharp edged detail is reduced by diffraction, and diffraction can be reasonably well deconvolution sharpened to restore resolution, but without aliasing artifacts, and the low noise images which the camera is capable of will allow more severe deconvolution sharpening.
Cheers,
Bart
-
One of the reasons I propose a random target is that I have a gut feeling that is contrary to Bart's and Erick's knowledge.
I think that to have these "artifacts" one needs a pattern (Bayer) on top of a pattern (suit or building windows) at near the Nyquist frequency.
I don't think that a Bayer pattern on a random structure (nature) will cause artifacts. The best way to prove a theory is to try and prove yourself wrong.
So I propose taking Bart's chart and removing the lines that are not prime leaving lines 2,3,5,7,11,13 etc and re-shoot the test. I might very well be wrong but I'd like to see it with my own eyes!
Marc
I think you're misinterpreting the results. The files simply indicate that both cameras capture the same amount of detail up to the green circle, and that any information within the red circle is no different from random garbage. The difference between the green circle positions is the difference in detail retention, which is negligible. The differences WITHIN the green circle is the kind of garbage that should be selected on personal preference. It is NOT however, an indication of higher resolution.
-
I guess my skepticism is it appears the AA filter was always blamed for loss of quality, suddenly it isn't any more. OF course, the other thing that I"m curious about is the camera doesn't have an "aa" filter per se, the filter they are using isn't exactly the same as having no AA filter. I'm not sure why you bend the light only to "bend it back" ... why not just leave it off? I really can't understand why this concept is better.
I think it may possibly help mitigate glare. The one bouncing off of the sensor surface to the back of the lens.
I can't imagine Nikon engineers take layers into account when designing lenses, because the lenses aren't exclusively used with this camera, nor can they be sure that future sensors require equal types of layers.
-
My speculation on why the Nikon D800E does not simply omit the two (somewhat expensive) layers that make up the OLPF is that
- this would slightly change the optical path and thus where precise focus occurs, but
- this change would not affect the optical path to the PD AF sensors or the OVF,
so there would be a discrepancy, leading to slight focusing errors with both AF and MF. To correct that could require a slight mechanical change to the mirror/focusing/VF assembly, and having two versions of that assembly (one for cameras with anti-aliasing, one for cameras with aliasing) would have cost more.
-
... in those 0.01% cases, where one apparently has time to create perfect circumstances, one also can do a double exposure and the blending will be the perfect AA result by sheer imperfections between the 2 shots, but with the additional benefit of extra depth.
Only if there is not even 0.01% subject movement between the frames! Those sort of cases, allowing combination of multiple frames taken sequentially, can often be handled with stitching, which gives a vastly greater resolution gain than this eye-ball straining search for differences between the D800 and D800E.
-
I think you're misinterpreting the results. The files simply indicate that both cameras capture the same amount of detail up to the green circle, and that any information within the red circle is no different from random garbage. The difference between the green circle positions is the difference in detail retention, which is negligible. The differences WITHIN the green circle is the kind of garbage that should be selected on personal preference. It is NOT however, an indication of higher resolution.
No what I'm thinking is that if the pattern being shot is random (nature landscapes) there would be nothing inside the circle so you might as well have the superior acutance. It is only because of the test pattern and a Bayer array that with the test target the non AA camera has false information inside the circle. To state it differently a non AA filtered camera will create this random "garbage" only with a test target not with a random scene. So why would I sacrifice acutance on 99% of my image for the slight chance of some small fake detail only when the scene has details close to the resolution limit of the system (Nyquist frequency) and a repetative pattern. The test pattern is a great test of the effectiveness of the AA filter to prevent Moire but does not represent what artifacts would be present (if at all) on a random subject or a subject where details are not near the resolution limit of the system. When I was running my loudspeaker company many would argue that certain crossover (filter) designs had better phase characteristics than the 4th order LR filters that I would design. They were correct but only at the crossover frequency how about the other 99% of your music? make 99% worse so that 1% is better, not a good design tradeoff. My gut feeling is that the artifacts are only present with a pattern and only near or below Nyquist frequency at best less than 1% probably .01% of a nature landscape scene. Why make 99.9% of my image worse just to make .1% better? At any rate there is only one way to prove me wrong thats to run the test on a pseudo random target.
Marc
-
No what I'm thinking is that if the pattern being shot is random (nature landscapes) there would be nothing inside the circle so you might as well have the superior acutance. It is only because of the test pattern and a Bayer array that with the test target the non AA camera has false information inside the circle.
No, BOTH cameras produce garbage within the circle. It is just a matter of personal preference and general subject matter whether you like or require blurry garbage, or grainy garbage. Landscape photography may benefit from grainy garbage, but wedding dresses and architecture may benefit from blurry garbage. Especially in case of wedding photography I would suggest to try before you buy to make a decision.
Note also that a well designed AA filter does NOT degrade the other 99% of your image. If it does, it simply is a badly designed filter. The whole purpose of this test is to see if the filter in the 800 is well designed and clearly the test shows that the detail retention is equivalent for both cameras.
This doesn't mean that all of the other effects mentioned in this thread have become irrelevant, especially for one's personal shooting style, subject matter, processing preferences etc…, but I strongly doubt that a random pattern is going to reveal anything new that we don't already know currently.
-
It looks like a very well designed chart/test/approach to me. It shows very clearly that both cameras produce very similar levels of detail in an absolute sense and that the main differences between them are the microcontrast near Nyquist and the handling of fine detail above Nyquist.
The random pattern test MIGHT show the D800e in a slightly better light but it seems to me that test would actually be less revealing of what is going on. One interpretation of why someone might favour the random pattern test is that it might provide a result that is more to comforting i.e. if someone starts off with a belief that omitting the AA filter is bound to produce better detail, they might deep down inside want testing to show that this is the case - even if we have to choose a very special test to make that happen. Not a very scientific approach but very human ;-) Personally, I'd prefer to see the unvarnished truth which why I prefer the test as presented.
No, BOTH cameras produce garbage within the circle. It is just a matter of personal preference and general subject matter whether you like or require blurry garbage, or grainy garbage. Landscape photography may benefit from grainy garbage, but wedding dresses and architecture may benefit from blurry garbage. Especially in case of wedding photography I would suggest to try before you buy to make a decision.
Note also that a well designed AA filter does NOT degrade the other 99% of your image. If it does, it simply is a badly designed filter. The whole purpose of this test is to see if the filter in the 800 is well designed and clearly the test shows that the detail retention is equivalent for both cameras.
This doesn't mean that all of the other effects mentioned in this thread have become irrelevant, especially for one's personal shooting style, subject matter, processing preferences etc…, but I strongly doubt that a random pattern is going to reveal anything new that we don't already know currently.
-
No, BOTH cameras produce garbage within the circle.
Agreed, but you can't tell how much garbage within the circle is caused by moire. Without a target that eliminates the issue of moire I don't know how you can do an apples to apples comparison ...
-
Agreed, but you can't tell how much garbage within the circle is caused by moire. Without a target that eliminates the issue of moire I don't know how you can do an apples to apples comparison ...
Hi Wayne,
Everything within the red (Nyquist frequency) circle is caused by aliasing, because the image detail is too small to resolve reliably. Whether that leads to moiré depends on the structure of the image detail itself, and its orientation. Moiré is a specific manifestation of aliasing caused by small repetitive image detail and a regular sampling grid. So whether it is moiré or other aliasing artifacts, it remains garbage. I also would agree that some garbage can mimic convincing detail, but we have very little control over it.
Cheers,
Bart
-
The random pattern test MIGHT show the D800e in a slightly better light but it seems to me that test would actually be less revealing of what is going on. One interpretation of why someone might favour the random pattern test is that it might provide a result that is more to comforting i.e. if someone starts off with a belief that omitting the AA filter is bound to produce better detail, they might deep down inside want testing to show that this is the case - even if we have to choose a very special test to make that happen. Not a very scientific approach but very human ;-) Personally, I'd prefer to see the unvarnished truth which why I prefer the test as presented.
I think both charts are necessary to get an accurate picture. The first is certainly necessary to properly assess moire but the pseudo random chart is necessary to assess acutance (I think!?) so why argue lets do the test! Bart where can I download a copy of your chart?
Marc
-
I can understand why some folks are reluctant to accept what the test results show. But there is only one logical explanation. The use of a good a.a. filter has virtually no effect on the resolution of actual detail, and helps reduce garbage--false detail and color artifacting as well as moire.
In some situations, a D800e will have a very slight advantage over a D800 in areas of fine low-contrast shadow detail. This advantage is not in resolution, but in micro contrast. It can probably be easily compensated for in most situations by the use of good post-processing routines, including deconvolution sharpening. This compensation may be less effective at higher ISO's because of the complication of more prominent noise.
In all honesty, the files I prepare for printing do have artifacts, viewed at 100%. I find that I have to give extra sharpening because of ink bleeding. So my print files are just on the edge of being "crunchy," with an almost invisible mist of grain from the sharpening. They look better that way in the final print. I use deconvolution sharpening and micro contrast to get them to that point.
On the other hand, aliasing artifacts from the camera can also make a file look sharper. It's not real detail, but it looks good at some sizes and at some magnifications, to many people. Files from cameras without a.a. filters need less sharpening and can handle less sharpening.
There are all kinds of sharpness paradoxes in photography. This is nothing new. Grain can make a file (or a negative) print better sometimes. Synthetic edge effects built into color films can make a photograph look sharper. The psychology of sharpness is fascinating. So in some sense there's no such thing as "pure" sharpness. At least not at our current stage of technology.
What I'm saying, I guess, is pick your poison. Do you want to adopt aliasing artifacts because they give an impression of sharpness? Or would you rather dial up your sharpening process a little? I prefer the second alternative, because I can control it. And sharpening software will keep getting better as time goes on, so my original file can be improved. Plus I hate camera artifacts and shoot a lot in the city. But that's just me.
I'm glad to see some of the mythology of a.a. filtration put to rest. There's been a touch of elitism in talking about a.a.-less sensors; an assumed superior knowledge that doesn't correspond to optical theory or, as we are seeing, real life. I'm happy we have choices, and happy that we're learning more about what the choices actually offer us.
-
I think it was Slobodan who coined the term "Squints" I must have laughed for 10 minutes, a perfect description! Well I guess I am a "LuLu Squint" and proud to be one!...........
After a nice morning walk photographing a beautiful waterfall in Kobe, it dawned on me that we are not talking about resolution we are talking about acutance or the ability of a sensor to define a one pixel line. The test charts clearly showed an increase in resolution of 1% and most dismissed this as negligible. Well acutance is on the pixel level not the global level so lets take a minute to "squint"! so a D800 has 7380 horizontal photo sites, it can resolve 3680 line pairs. and the E model can out resolve the 800 by 1% that is 1% of 3680 lp's. 1% of 3680 is 37 lp's! thats a 3700% difference in acutance!!!! more than an order of magnitude.
Marc
-
I can understand why some folks are reluctant to accept what the test results show. But there is only one logical explanation. The use of a good a.a. filter has virtually no effect on the resolution of actual detail, and helps reduce garbage--false detail and color artifacting as well as moire.
This.
-
Bart where can I download a copy of your chart?
Hi Marc,
There are a few different versions available depending on which printer one uses, and also a version for the real masochists who want to really stress the Bayer CFA on its weakest point (red/blue resolution with approx. equal luminance).
The targets can be downloaded from the first post in this thread (http://www.openphotographyforums.com/forums/showthread.php?t=13217) on another website. I usually do not like to draw people away from a forum, but there are also explanations there as to its use and interpretation, and some examples, which I don't like to copy to multiple websites.
Cheers,
Bart
-
Hi Marc,
There are a few different versions available depending on which printer one uses, and also a version for the real masochists who want to really stress the Bayer CFA on its weakest point (red/blue resolution with approx. equal luminance).
The targets can be downloaded from the first post in this thread (http://www.openphotographyforums.com/forums/showthread.php?t=13217) on another website. I usually do not like to draw people away from a forum, but there are also explanations there as to its use and interpretation, and some examples, which I don't like to copy to multiple websites.
Cheers,
Bart
Thanks Bart
-
The random pattern test MIGHT show the D800e in a slightly better light but it seems to me that test would actually be less revealing of what is going on. One interpretation of why someone might favour the random pattern test is that it might provide a result that is more to comforting i.e. if someone starts off with a belief that omitting the AA filter is bound to produce better detail, they might deep down inside want testing to show that this is the case - even if we have to choose a very special test to make that happen. Not a very scientific approach but very human ;-) Personally, I'd prefer to see the unvarnished truth which why I prefer the test as presented.
Random patterns tend to be the norm. Specialised and artificially constructed test charts are useful to understand what's going on and can provide a degree of certainty.
However, it is quite natural that anyone who is contemplating buying a camera without the usual AA filter, would be more interested in perceived advantages in real-world scenes which are not usually constructed from evenly spaced, black and white lines.
Now it's clear that the removal of the AA filter does put one at risk of occasionally spoiling an image with moire and aliasing artifacts. What I'd like to know is just how effective the Nikon software is at removing such artifacts. Are those, perhaps rare, occasions when moire rears its ugly head, fatal? In removing the moire, does the Nikon software degrade the image so much that the D800 image is clearly superior?
If one is a scientist gathering data which has to be as accurate as possible, the last thing one wants are artificial data created by the measuring equipment.
However, if the purpose of the exercise is to created a pleasing, interesting and/or esthetic image, in the eyes of some, then the presence of artifical data is of little consequence unless it is visually jarring or obtrusive.
To take an extreme example, I have very few strands of hair on my head. If someone were to shoot a portrait of me using a camera without an AA filter, and such camera were to create additional strands of hair due to its capacity to (apparently) resolve beyond the Nyquist limit, would I be angry and furious? ;D
-
To take an extreme example, I have very few strands of hair on my head. If someone were to shoot a portrait of me using a camera without an AA filter, and such camera were to create additional strands of hair due to its capacity to (apparently) resolve beyond the Nyquist limit, would I be angry and furious? ;D
If you hear any further news on such 'positively discriminatory' cameras, would you be kind enough to pass it on?
;-)
Rob C
-
I have a question: The Nyquist rate for the D800 sensor is much higher than has ever been seen before in a FF sensor, as the M9 sensor is 'only' 16MP. Shouldn't the chances of moire be much lesser in the case of the D800E since it resolves a lot more?
-
I have a question: The Nyquist rate for the D800 sensor is much higher than has ever been seen before in a FF sensor, as the M9 sensor is 'only' 16MP. Shouldn't the chances of moire be much lesser in the case of the D800E since it resolves a lot more?
Aliasing happens whenever there is significant signal beyond half the sampling rate. This manifests itself as spurious errors in the digitized signal that can appear at any frequency (e.g. very low frequency). Moire (used in this context) is just the visible periodic errors that appears when there is a periodic pattern in the scene that happens to trigger significant aliasing at the sensor.
If the lense, scene, recording technique etc is scaled accordingly, aliasing should be as likely on a 2MP sensor as a 180MP sensor. Of course, those factors usually does not scale lineary with sensel count, but it seems that many photographers do their best in order to make it so.
In short: perhaps, but not necessarily.
-h
-
I have a question: The Nyquist rate for the D800 sensor is much higher than has ever been seen before in a FF sensor, as the M9 sensor is 'only' 16MP. Shouldn't the chances of moire be much lesser in the case of the D800E since it resolves a lot more?
Hi Sareesh,
If everything else is the same, a higher sampling density will indeed reduce the risk of aliasing artifacts.
Aliasing artifacts can only be avoided by increasing the relative size of the finest image detail to more than twice the sampling pitch, and/or by reducing the contrast of the image detail far enough.
It will probably require a sub-micron sensel pitch to avoid aliasing without additional help.
Cheers,
Bart
-
Attention - these charts were run with the wrong gamma setting in Imatest
Please see this post for corrections:
http://www.luminous-landscape.com/forum/index.php?topic=65927.140 (http://www.luminous-landscape.com/forum/index.php?topic=65927.140)
Hi Bart,
Thanks for doing the Imatest run, I didn't have the time to do it myself yet.
Glad to help.
The only change to the settings I could suggest is to replace the crop size setting with the full image size of 7416 x 4916 (which is what some converters squeeze out of the Raw) or the formal 7360 x 4912 pixels.
Good point. I noticed that after posting. I usually work with the raw files with embedded exif data that automatically fills in most of the image parameters. I've re-run the horizontal profile charts using your 7416 x 4916 image dimensions. Of course, this did not change the key, edge profile results or SFR (MTF) plots from the original run. But the results that depend on picture height (PH) such as LP/PH are now valid. I've also added the Acutance charts since they are another important metric of Subjective Quality Factor (SQF), as was mentioned in another post. As a newbie on this forum, I'll add another very soft request for access to the original raw files, if and when Michael can spare the time and bandwidth? (Maybe there is something in the background of the full frame captures that he doesn't want to distribute on the web---perhaps one of those infamous Secret-Service style orgies rumored to go on in the Lu-La studio after hours?! :))
After a bit of additional research, my feeling on the Dead Leaves pattern method is that the random, variable-contrast edges may indeed be useful in determining how much of the low-contrast, fine detail might be permanently lost below the noise floor on the D800 (especially at higher ISO), relative to the D800E. The Dead Leaves pattern is similar in concept to the Random (Scale-Invariant) patterns discussed by Imatest:
http://www.imatest.com/docs/random/
Imatest master will analyze both patterns. I haven't tried this method, yet.
The various Raw converters I've tried sofar do add a reddish tint to the D800E conversions in varying degrees, interpreted by Imatest as CA.
That's interesting---maybe those converters are not completely “baked” yet? My thought was that Imatest was just detecting the additional color aliasing apparent in the D800E image, at spatial frequencies beyond Nyquist.
Personally I don't worry about the noise too much, but then I'm a low ISO shooter, and things look better inprint that when pixel peeping on a monitor.
That pretty much describes me as well. As a landscape and macro shooter, I shoot at near base ISO perhaps 90% of the time. That said, there are times such as when shooting long-stemmed lupine at twilght out in the ever-windy foothills of the Eastern Sierra Nevada mountains in California, that the option to boost the ISO to enable faster shutter speeds can sure be useful!
_______________________________________________________________________________
Second run with Imatest 3.9 Beta with corrected pixel dimensions.
D800E SFR
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-D800E-Imatest-Tests2/i-gc3vpxN/0/X3/D800E-Bart-van-der-Wolf-D800E-X3.jpg)
D800E SRF LP/PH
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-D800E-Imatest-Tests2/i-X9Bk7xm/0/X3/D800E-Bart-van-der-Wolf-D800E-X3.jpg)
D800E Acutance
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-D800E-Imatest-Tests2/i-59fvcG2/0/X3/D800E-Bart-van-der-Wolf-D800E-X3.jpg)
D800E Chromatic Aberration
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-D800E-Imatest-Tests2/i-bmWbd8M/0/X3/D800E-Bart-van-der-Wolf-D800E-X3.jpg)
D800 SFR
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-D800E-Imatest-Tests2/i-6fdjgh9/0/X3/D800-Bart-van-der-Wolf-D800-X3.jpg)
D800 SRF LP/PH
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-D800E-Imatest-Tests2/i-VNvPKL3/0/X3/D800-Bart-van-der-Wolf-D800-X3.jpg)
D800 Acutance
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-D800E-Imatest-Tests2/i-QWT5wRX/0/X3/D800-Bart-van-der-Wolf-D800-X3.jpg)
D800 Chromatc Aberration
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-D800E-Imatest-Tests2/i-dPR9r87/0/X3/D800-Bart-van-der-Wolf-D800-X3.jpg)
Imatest Settings
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/D800-D800E-Imatest-Tests2/i-v3CKfMG/0/X2/Imatest-Settings-with-Dims-X2.jpg)
Results generated by Imatest Studio 3.9 Beta
http://www.imatest.com/ (http://www.imatest.com/)
Kind regards,
Bob
-
Why is it that the chromatic abberation on the d800e result is so much worse than with the d800? ???
-
Perhaps someone would provide a summary of the acutance difference between the two cameras for those of us not familiar with interpreting imatest charts
Thank
Marc
-
Why is it that the chromatic abberation on the d800e result is so much worse than with the d800? ???
Because it is not CA, but demosaic residue. False edge colors due to the higher acutance between individual sensels which do not have equal colorfilters. It should only happen because these files were created without any color noise reduction.
-
Hi Jan and Oscar,
Why is it that the chromatic abberation on the d800e result is so much worse than with the d800? ???
Because it is not CA, but demosaic residue. False edge colors due to the higher acutance between individual sensels which do not have equal colorfilters. It should only happen because these files were created without any color noise reduction.
Yes, that's my conclusion as well. This is clearly not caused by classic chromatic aberration due to the lens. If you magnify the edges of the images from both the D800 and the D800E (in PS at highest zoom level), you can see a repeating pattern of red fringing in the D800 image, which is not present in the D800 image. Although the Imatest CA module was not specifically designed to analyze such false-color artifacts, I included the CA charts because I believe they may prove useful in testing the effectiveness of various demosaic algorithms. I also agree that, at least for the slant-edge portions of this chart, the false colors could be easily removed by applying color noise reduction (or CA reduction) methods (especially if applied during raw conversion).
However, I am not yet convinced that any of the tools intended to correct such failures of the Bayer CFA demosaic process (including the new moire removal tools in Lightroom, ACR and Capture NX2) will completely eliminate this issue in all cases.
As a side note, it's often been stated that nature and landscape photographers do not need to worry about artifacts caused by the absence of an effective OLP or AA filter on the sensor. But there are a class of subjects that I have not seen mentioned yet, where it can definitely be an issue. One of my photographic interests is astrophotography, both star trails and guided (on an equatorial mount) fixed images of star fields. Such images are best captured at relatively wide apertures using sharp lenses, under the best possible atmospheric “seeing conditions”. Due to the inherently small PSF (or Line Spread Function (LSF) in the case of star trails), the star images (Airy-disks) projected on the sensor may not cover enough elements in the Bayer array to allow proper color rendering, resulting in stars or trails with randomly-distributed false colors.
(Hey, why are all the stars in this image red, green or blue, with twice as many green stars as red or blue?! :o ---yes, I know that most of the stars won't be rendered pure R, G or B.)
So why not just apply color correction in pp? Part of the appeal of celestial images are the subtle natural colors of the stars and planets. The false-color correction routines I've seen to date also damage these natural color tones This is more than conjecture; I've seen it in my own images and a number of examples online. My current D700 is fairly immune, due to the relatively strong AA filter. But my old D70 (weaker AA filter) did have this issue. One common way to avoid this is to slightly defocus the star images. But this causes loss of the dimmer stars.
Those who like to shoot night scenes involving distant city lights (or any small specular highlight), may also experience this problem.
The high sampling density of the D800/D800E sensor should reduce the occurrence of the above issues. But it is one more thing to be aware of when choosing correct camera.
Kind Regards,
Bob
-
Here is a crude example of what I had in mind for a random target
not sure how to construct it from scratch so it is neat but you can get the idea
I just left the lines that are prime, and the density of the pseudo random target should probably be 2X or 3X
Marc
-
Perhaps someone would provide a summary of the acutance difference between the two cameras for those of us not familiar with interpreting imatest charts
Hi Marc,
I'll take a shot at it. I don't want to drive this thread too far OT. But here's my understanding of the acutance results, along with some links for further reading.
For the sake of completeness, acutance is related to Subjective Quality Factor (SQF) and is a measure of perceived sharpness or edge sharpness of an image that includes factors of resolution (MTF) combined with standardized assumptions about human vision and viewing distance. Resolution is not affected by sharpening---acutance is affected by sharpening.
http://en.wikipedia.org/wiki/Acutance (http://en.wikipedia.org/wiki/Acutance)
First, be aware that Imatest automatically rescaled the vertical (y-axis) of the charts to match the data from each camera. You need to look at the values next to the left axis (labeled Acutance %) to properly interpret the results. Here you'll see that the acutance plot of D800E is somewhat higher than that of the D800, running from about 80% for a 2.5 cm (1 inch) high print to 47% for a 40 cm (16 inch) high print. The D800 plot runs from about 78% to 39% acutance over the same range of print sizes. So there is a 2% difference in acutance for a 2.5 cm high print and an 8% difference for a 40 cm high print, both in favor of the D800E. The horizontal “Picture Height” (print or display height) chart scale can be extended to include much higher print sizes (not shown), where the improvement in acutance of the D800E over the D800 increases slowly to about 9% difference for a 200 cm (80 inch) high print.
My personal opinion is that these results do not represent a huge difference in acutance between the two cameras. It confirms the previous statements that most of the difference in perceived sharpness (at least for higher contrast edges) between output from the D800 and D800E (in print and display) can be mitigated by proper sharpening. My feeling is that the jury is still out regarding detail that is simultaneously both of high spatial frequency and low contrast, especially as ISO is increased.
Links:
Reading the Imatest SQF and Acutance charts:
http://www.imatest.com/docs/sqf/ (http://www.imatest.com/docs/sqf/)
A good description of SQF by Bob Atkins:
http://www.bobatkins.com/photography/technical/mtf/mtf4.html (http://www.bobatkins.com/photography/technical/mtf/mtf4.html)
My favorite paper on the value and limitations of MTF testing by Hubert Nasse at Zeiss (2 parts):
http://www.zeiss.de/C12567A8003B8B6F/EmbedTitelIntern/CLN_30_MTF_en/$File/CLN_MTF_Kurven_EN.pdf (http://www.zeiss.de/C12567A8003B8B6F/EmbedTitelIntern/CLN_30_MTF_en/$File/CLN_MTF_Kurven_EN.pdf)
http://www.zeiss.de/C12567A8003B8B6F/EmbedTitelIntern/CLN_31_MTF_en/$File/CLN_MTF_Kurven_2_en.pdf (http://www.zeiss.de/C12567A8003B8B6F/EmbedTitelIntern/CLN_31_MTF_en/$File/CLN_MTF_Kurven_2_en.pdf)
And of course Michael's tutorial from a few years ago:
http://www.luminous-landscape.com/tutorials/understanding-series/understanding-mtf.shtml (http://www.luminous-landscape.com/tutorials/understanding-series/understanding-mtf.shtml)
The last four links are blessedly free of large complex equations that, upon exposure, have been known to cause some folks to begin sweating, then rapidly degrade into convulsions, followed by invagination of the temples and gray matter dribbling from the ears.
Kind regards,
Bob
-
Why is it that the chromatic abberation on the d800e result is so much worse than with the d800? ???
If I understand the replies correctly, this ugly artifact is
- a manifestation of aliasing
- worse without an OLP filter than with
- quite likely to occur in natural scenes when there is a sharp change of luminosity along a line, such as at the edge of a shadow.
Maybe we need to move beyond the ideas that
a) the only aliasing problem is moiré,
b) moiré can only occur with certain made made subjects, and
c) us nature oriented photographers never photograph these sort of man-made subjects, so
d) we have nothing to worry about from a camera with no OLP filter.
It seems to me that all three of (a), (b) and (c) are at least over-simplifications, if not outright wrong, so I am unconvinced of (d). Of course, those who can afford several bodies for use in different situations might find a no-OLP camera to he useful as one part of the kit.
-
As a user of Foveon based cameras and mirrorless cameras with no or weak OLP filtration, it definitely isn't moire that bothers me in landscapes. It's stair stepped jaggies that destroy the integrity of the edges of twigs and branches and blades of grass and make thin lines look like they are constructed from lego bricks or laminated from long planks that appear to peeling apart.
If I understand the replies correctly, this ugly artifact is
- a manifestation of aliasing
- worse without an OLP filter than with
- quite likely to occur in natural scenes when there is a sharp change of luminosity along a line, such as at the edge of a shadow.
Maybe we need to move beyond the ideas that
a) the only aliasing problem is moiré,
b) moiré can only occur with certain made made subjects, and
c) us nature oriented photographers never photograph these sort of man-made subjects, so
d) we have nothing to worry about from a camera with no OLP filter.
It seems to me that all three of (a), (b) and (c) are at least over-simplifications, if not outright wrong, so I am unconvinced of (d). Of course, those who can afford several bodies for use in different situations might find a no-OLP camera to he useful as one part of the kit.
-
Hi Marc,
I usually look at the MTF curves and the MTF at Nyquist. I got the impression that MTF should not exceed around 20% at Nyquist if aliasing is to be avoided.
According to Norman Koren, the author of Imatest, the resolution at which the system transfers 50% MTF is a good measure of the visual acuity of the system. Sorry for the phrasing. I may also check LW/PH at 18% MTF. To be able to restore detail by sharpening, there needs to be some contrast. There are some different criteria, the Raleigh criteria is the most commonly used, but it is intended to separate to images of nearby stars. I sort of decided to use 18% as criteria. My use of the 18% criteria is most intended for study of diffraction, essentially saying that detail can be restored by deconvolution if MTF is higher than 18%. my choice of 18% is quite arbitrary.
Best regards
Erik
Perhaps someone would provide a summary of the acutance difference between the two cameras for those of us not familiar with interpreting imatest charts
Thank
Marc
-
Oops...hold the presses! Ah shucks too late.
I got a PM from Bart this morning that there is an error in my Imatest analysis. He reminding me that the dng files provided for download were converted into AdobeRGB color space with gamma 2.2 already applied. Therefore, the data must be re-linearized by applying the reciprocal gamma value 1 / 2.2 or 0.4545 in Imatest to correct the data and plots. When I screw up, it has always been my policy to offer an immediate retraction and re-publish the corrected results, which are shown below. I've also edited the previous posts to point to this one. Thanks Bart, I should have caught that one! Sorry for the confusion.
Note that this has made a significant difference in the previously posted results. In absolute terms, the response from both cameras have improved considerably, but the relative differences, while still significant, were not effected as much. One exception is the CA chart (possible demosaic failure), where the deviation is still present for the D800E, but is much improved. I'm sure there will be additional changes in the results when the original files are processed with other raw converters and when capture sharpening is applied.
Hopefully these are now all correct. Note that I've expanded the Picture Height scale on the Acutance chart to include print heights up to 100 cm (40 inches).
D800E SRF cycles/mm
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/3rd-LuLa-Post-Corrected-Gamma/i-Nq4TNNb/0/X3/D800E-Bart-van-der-Wolf-D800E-X3.jpg)
D800E SRF LP/PH
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/3rd-LuLa-Post-Corrected-Gamma/i-FMkPQsm/0/X3/D800E-Bart-van-der-Wolf-D800E-X3.jpg)
D800E Chromatic Aberration
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/3rd-LuLa-Post-Corrected-Gamma/i-6bPnRBJ/0/X3/D800E-Bart-van-der-Wolf-D800E-X3.jpg)
D800E Acutance
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/3rd-LuLa-Post-Corrected-Gamma/i-jcHNFMm/0/X3/D800E-Bart-van-der-Wolf-D800E-X3.jpg)
___________________________________________________________________________________________
D800 SRF cycles/mm
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/3rd-LuLa-Post-Corrected-Gamma/i-jZS4FLB/0/X3/D800-Bart-van-der-Wolf-D800-X3.jpg)
D800 SRF LP/PH
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/3rd-LuLa-Post-Corrected-Gamma/i-6mJw5kH/0/X3/D800-Bart-van-der-Wolf-D800-X3.jpg)
D800 Chromatic Aberration
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/3rd-LuLa-Post-Corrected-Gamma/i-sMc6R4D/0/X3/D800-Bart-van-der-Wolf-D800-X3.jpg)
D800 Acutance
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/3rd-LuLa-Post-Corrected-Gamma/i-rGdb9TV/0/X3/D800-Bart-van-der-Wolf-D800-X3.jpg)
Imatest Settings
(http://bobtothe3rd.smugmug.com/D800-D800E-Tests/3rd-LuLa-Post-Corrected-Gamma/i-6Gt4sQZ/0/X3/Imatest-Settings-Corrected-X3.jpg)
I offer my favorite circular-logic quote from an unknown author, “I thought I made a mistake once, but I was wrong!” :-[
Kind regards,
Bob
-
I'm trying to wrap my brain around this but don't have any backround with Imatest. IRT acutance if at 50cm picture height the E is 56% and the 800 is 45% is 11% a significant difference? as a comparison what would the acutance difference be between a high end lens and a kit lens? much more or less?
IRT edge profile is an infinite slope perfect? so the closer to 0 the better? again just to get a gut feeling what is the difference one would expect between a high end lens and a kit lens?
Thanks
Marc
-
I'm trying to wrap my brain around this but don't have any backround with Imatest. IRT acutance if at 50cm picture height the E is 56% and the 800 is 45% is 11% a significant difference? as a comparison what would the acutance difference be between a high end lens and a kit lens? much more or less?
Acutance or SQF is expressed on a quality scale.
(http://www.imatest.com/images/sqf_bror_quailty.gif)
A blurry picture viewed from a large distance looks sharp and scores high (dotted line in the acutance plot). That same blurry picture shown at a close distance but also at a small size also scores high (solid line in the acutance plot). So to reach a better than 'very good' score of 80, the D800 image should be viewed from 71 cm distance and further and the D800E can be viewed from 66 cm distance and further for the same subjective quality. The same quality can be achieved by printing a 2.5 cm (approx. 1 inch) tall image (landscape orientation) or less and viewing it from some 13-15 cm distance (half of the normal reading distance).
IRT edge profile is an infinite slope perfect? so the closer to 0 the better? again just to get a gut feeling what is the difference one would expect between a high end lens and a kit lens?
Yes, steeper is better, but steeper than 1 pixel is not really possible. An edge profile (edge spread function, or ESF) shows how the abrupt transition from the dark to the light side of a sharp edge is captured. A perfect edge could be represented by going from dark to light in one pixel to the next if, and only if, the edge is positioned exactly between the sensels. Of course, if the edge falls exactly halfway a row or column of pixels, the sensel response would be 50% of the luminance difference, but Imatest measures at a sub-pixel accuracy.
The D800 takes 2.3 pixels to transition the edge, and the D800E takes 1.91 pixels to transition the edge, both unsharpened and with Capture One v6.4 as Raw converter. That means that the D800 needs to be sharpened with a sharpening radius of 0.9, and the D800E requires a sharpening radius of 0.75 to achieve the same sharpening effect in that Raw converter.
Cheers,
Bart
-
I'm trying to wrap my brain around this but don't have any backround with Imatest. IRT acutance if at 50cm picture height the E is 56% and the 800 is 45% is 11% a significant difference? as a comparison what would the acutance difference be between a high end lens and a kit lens? much more or less?
IRT edge profile is an infinite slope perfect? so the closer to 0 the better? again just to get a gut feeling what is the difference one would expect between a high end lens and a kit lens.
Hi Marc,
Yes, 11% acutance increase is significant but probably only marginally perceptible in a 50cm print. Keep in mind that these charts were performed on the dng files provided by Bart and they have not been sharpened. My goal was to get a feeling for the base difference in MTF and acutance on the images as output from the cameras. Acutance values are highly affected by sharpening---in fact, sharpening is required on both D800 and D800E images to get optimum acutance in the final output (especially true with the D800 images due to the AA filter). Proper capture sharpening requires access to the raw files so I will refer to a newer thread started by Erik Kaffehr here:
http://www.luminous-landscape.com/forum/index.php?topic=66477.0
He apparently now has access to Michael’s raw files. So his Imatest results are more realistic than mine for comparing the acutance of the D800 and D800E images after sharpening is applied.
Yes, in general the steeper the edge profile or Edge Spread Function (ESF), the higher the MTF---but it can never go smaller than 1.0 pixels. Sharpening can increase the steepness of the ESF but will not increase the resolution.
Kit lens vs high-end lens is a hard one to answer since kit lenses vary so widely in image quality. Many consider the $100 Nikon 50mm f/1.8 a “kit” lens. But used at f/4 to f/5.6, I would not expect a huge difference in MTF compared to Michael’s 85mm (f/1.4G?) used for the tests in this thread (at least not at CoF). The corners and edges would probably be a different story. For whatever lenses you have in mind, a good place to look strictly for comparing SQF/Acutance data at various print sizes for specific lenses is Popular Photography (but I find Pop Photo lens testing a bit light on other important details).
Looks like Bart beat me to it. But at least we both answered different parts of the questions.
I look forward to Bart and Erik's ongoing sharpening and (hopefully) noise tests.
King regards,
Bob
-
I guess I was trying to gage whether the difference between the acutance/SQF between the 800 and the E is in the same ball park as the difference between a high end lens and a average kit lens or is it an order of magnitude less or more?
Also what jumps out is that for a given print sharpness (personal taste) the E needs less sharpening, in my past experiance I always preferred the look of a less sharpened AA/less print to a more sharpened AA print. It will be interesting to compare real world prints of the 800 sharpened at .9 and the E sharpened at .75.
Marc
-
Well, I guess that's self-evident, Bart. If the sharpening were to increase noise to unacceptable levels, it couldn't be considered proper. ;D
What occurs to me is, if the D800 shot is always sharpened so that it looks as sharp as the D800E shot, could there be visibly more noise at least somewhere in the full size image, despite one's best efforts.
It may well be the case that at base ISO this is less likely to be a problem. But what about ISO 3200? What about small crops that one wants to print largish?
I agree that featureless areas can be masked, but detail in the deep shadows tends to have an unfavourable SNR. When I use Smart Sharpen, I usually fade the amount of sharpening applied to the shadows.
I also wonder what will happen at high ISO when noise may be a problem without any sharpening at all.
Maybe, but I don't have the cameras to try that. I have to make a decision as to which model to order. I'm currently favouring the D800E.
I was very impressed with Focus Magic, but I'm disappointed they are taking so long to develop a version compatible with 64 bit Windows OS.
Cheers!
Ray
Ray,
Consider Images Plus. It is a 64 bit multi-threaded program written by a math prof. for astronomy shots. I have been using it for years for regular photography. It has a very nice adaptive RL that limits noise buildup.
-
Probably because it keeps the distance between the mount and sensor exactly the same between the D800 and D800e?
Cheers,
Bernard
The distance from my eyes to my screen wont change if i take off my glasses. What will change is the angle of light entering my eye. In the same way the AA filter may have an impact that is a part of the design of the microlenses in front of the sensor. Without it the light might miss the microlenses.
-
The distance from my eyes to my screen wont change if i take off my glasses. What will change is the angle of light entering my eye. In the same way the AA filter may have an impact that is a part of the design of the microlenses in front of the sensor. Without it the light might miss the microlenses.
I guess that it depends on the set of phenomena you are trying to deal with.
I would think that complex reflexion can occur between the sensor and the rear elements of lenses and that complex coatings are applied to sensors to reduce these. I wouldn't be surprised if the distance were to be accurately preserved. Just guessing though.
Cheers,
Bernard
-
The distance from my eyes to my screen wont change if i take off my glasses. What will change is the angle of light entering my eye. In the same way the AA filter may have an impact that is a part of the design of the microlenses in front of the sensor. Without it the light might miss the microlenses.
1. If your head is pressed up against the display using fixed-force screws, then adding any material in-between (such as glasses) would change the physical distance.
Not that I know how sensors are aligned to the camera frame/lense.
2. If a (flat, nonconcave/convex) object is inserted whose refractive index is very different from air, could it not alter the "optical distance"?
-h
-
"One thing to bear in mind is that the whole camera is only as fast as the fastest card installed. "
i wonder whether you possibly meant "as fast as the slowest card installed"...
cheers,
pedro
-
Oops...hold the presses! Ah shucks too late.
I got a PM from Bart this morning that there is an error in my Imatest analysis. He reminding me that the dng files provided for download were converted into AdobeRGB color space with gamma 2.2 already applied. Therefore, the data must be re-linearized by applying the reciprocal gamma value 1 / 2.2 or 0.4545 in Imatest to correct the data and plots. When I screw up, it has always been my policy to offer an immediate retraction and re-publish the corrected results, which are shown below. I've also edited the previous posts to point to this one. Thanks Bart, I should have caught that one! Sorry for the confusion.
Bob,
That was an easy error to make, since Imatest uses a gamma of 1/2.0 or 0.5 as a default for SFR as shown in Norman's doc page (http://www.imatest.com/docs/sfr_instructions2/#gamma) and many users (myself included) do not change the default. Did you use 2.2 for your original tests? On re-reading the docs, I see that Norman suggests including a Q-14 chart along with the slanted edge target so one can determine the actual gamma in use. With ACR using PV2010 one can use a linear tone curve (sliders on main tab set to zero and point curve to linear) and obtain reasonable results. However, with PV2012 it is not so easy to obtain a linear tone curve and the default is far from linear.
Shown below are my Imatest results for gamma using a Stouffer wedge and ACR 7 with a linear tone curve in PV1010 and the default tone curve in PV2012. The wedge was exposed with step 1 just short of clipping, but PV2012 uses a bright tone curve and the highlights appear washed out but are not actually blown. Bart's target does include 20 density steps ranging from 255 down to zero. On his OpenPhoto forum post (http://www.imatest.com/docs/sfr_instructions2/#gamma), he states that the image has no attached color space, but on downloading it today, I see that it is in ProPhotoRGB and the steps have even pixel spacing. If one knew more about the encoding of the image, one could determine the gamma of an image of the target. Perhaps Bart will provide some additional information.
Regards,
Bill
-
The distance from my eyes to my screen wont change if i take off my glasses. What will change is the angle of light entering my eye. In the same way the AA filter may have an impact that is a part of the design of the microlenses in front of the sensor. Without it the light might miss the microlenses.
The "optical" distance changes, meaning the distance and time of travel for the light, because of the different refractive index (lower speed of light) in the birefringent (Lithium Niobate) slabs compared to air, and the slightly longer path taken: light is turned a bit to the left or right by the first filter element, and then turned back by the second. This will slightly move the focus position, and the focus mechanisms (auto or manual) would need to be slightly adjusted for that. But I agree that the angle of incidence on subsequent optical elements could also be an issue.
Anyway, I am not sure why we are agonizing over this: rather clearly there is a reason, else Nikon would have had an obvious way to reduce the manufacturing cost by simply omitting those two not inexpensive birefringent slabs.
-
Bob,
That was an easy error to make, since Imatest uses a gamma of 1/2.0 or 0.5 as a default for SFR as shown in Norman's doc page (http://www.imatest.com/docs/sfr_instructions2/#gamma) and many users (myself included) do not change the default. Did you use 2.2 for your original tests? On re-reading the docs, I see that Norman suggests including a Q-14 chart along with the slanted edge target so one can determine the actual gamma in use. With ACR using PV2010 one can use a linear tone curve (sliders on main tab set to zero and point curve to linear) and obtain reasonable results. However, with PV2012 it is not so easy to obtain a linear tone curve and the default is far from linear.
Shown below are my Imatest results for gamma using a Stouffer wedge and ACR 7 with a linear tone curve in PV1010 and the default tone curve in PV2012. The wedge was exposed with step 1 just short of clipping, but PV2012 uses a bright tone curve and the highlights appear washed out but are not actually blown. Bart's target does include 20 density steps ranging from 255 down to zero. On his OpenPhoto forum post (http://www.imatest.com/docs/sfr_instructions2/#gamma), he states that the image has no attached color space, but on downloading it today, I see that it is in ProPhotoRGB and the steps have even pixel spacing. If one knew more about the encoding of the image, one could determine the gamma of an image of the target. Perhaps Bart will provide some additional information.
Thanks, Bill. Yes, unfortunately I did use gamma 2.2 for the first tests, so the results where way off mark, a simple brain fart. I thought of using the density steps in Bart’s chart to determine the gamma but was also unsure how to interpret them. I do have an old Kodak Q-13 (equivalent to the Q-14) that I use for my own testing.
Thanks for pointing out the differences in PV2012 vs PV2010; I was not aware of this issue. It will avoid some head scratching when I start working with raw files from the D800/E.
This may be straying a bit too far off topic into technical complexity for this thread, so I’ll try to start a new one this week to further explore Imatest methods and results. This is one of the few forums that provide such in depth discussions, with folks who have a deep understanding of camera/lens testing and software.
Kind regards,
Bob
-
Anyway, I am not sure why we are agonizing over this: rather clearly there is a reason, else Nikon would have had an obvious way to reduce the manufacturing cost by simply omitting those two not inexpensive birefringent slabs.
That's sort of been my feeling about it and why I'm waiting for the e. I would love to see some official word from Nikon (other than the market speak) as to why they opted for this instead of just omitting it.
-
A question for squints: in another thread, about sharpening and desaturation (http://www.luminous-landscape.com/forum/index.php?topic=66270.msg525623#msg525623), I posted a little demonstration how aggressive sharpening creates desaturation. So, the question is: if D800 requires more aggressive sharpening, wouldn't the collateral damage be desaturation?
-
Bart's target does include 20 density steps ranging from 255 down to zero. On his OpenPhoto forum post (http://www.imatest.com/docs/sfr_instructions2/#gamma), he states that the image has no attached color space, but on downloading it today, I see that it is in ProPhotoRGB and the steps have even pixel spacing. If one knew more about the encoding of the image, one could determine the gamma of an image of the target. Perhaps Bart will provide some additional information.
Hi Bill,
The target was created without colorspace, and saved without colorspace/profile as a PNG (which IIRC doesn't have a colorspace but does have a 'gamma-chunk'). I have no control over what Photoshop writes in the PNG metadata, but I suppose the Gamma chunk is set, I don't know to which value. Where it gets the ProPhotoRGB colorspace from, I have no idea (the file was saved without profile).
In my instructions for use I suggest to assign the printer profile to the target before printing, which should by-pass gamma adjustments and such since we're just sending R=G=B values to the printer. When the target is photographed the data is also recorded as is, and upon Raw conversion there will be a gamma precompensation so we should see and measure the same (more or less) linear step spacing as in the original. That means that theoretically the image has a gamma close to 2.2 . A plot of the luminosities of the PNG version of the D800 shot I uploaded does indeed show an almost (but not perfectly) straight tone stepping (see attached ImageJ profile plot). Therefore the correct Imatest gamma setting to linearize the data would be approx. 0.4545 .
Of course the easiest way to deal with these things is to use Imatest on linear gamma Raw data, but as it is it is also usable. We are not operating in a laboratory here, but the tone scale patches can be used to calibrate the image before measuring (that's why the steps are included in my target). The gray background of the target can also be used to compensate for uneven lighting of the target (and the target was indeed not uniformly lit).
Cheers,
Bart
P.S. I used ImageJ's trend curve plotting, and it decided that the gamma variate curve with the c parameter representing the gamma component of the profile plot data is gamma 2.28352 (see attachment) . So besides some other non-linearities, the best gamma setting for linearization would be 0.438 .
-
A question for squints: in another thread, about sharpening and desaturation (http://www.luminous-landscape.com/forum/index.php?topic=66270.msg525623#msg525623), I posted a little demonstration how aggressive sharpening creates desaturation. So, the question is: if D800 requires more aggressive sharpening, wouldn't the collateral damage be desaturation?
Hi Slobodan,
That nice demonstration also involved using a very (too) large sharpening radius. When the correct radius is used, then the risk you mentioned is not an issue.
That is also part of what I am driving at, using the correct sharpening settings (and they differ between the 2 Nikons). I'm close to publishing a tool with which users can determine their personal equipment's best sharpening settings. It does require some more squinting though ...
Cheers,
Bart
-
My speculation on why the Nikon D800E does not simply omit the two (somewhat expensive) layers that make up the OLPF is that
- this would slightly change the optical path and thus where precise focus occurs, but
- this change would not affect the optical path to the PD AF sensors or the OVF,
so there would be a discrepancy, leading to slight focusing errors with both AF and MF. To correct that could require a slight mechanical change to the mirror/focusing/VF assembly, and having two versions of that assembly (one for cameras with anti-aliasing, one for cameras with aliasing) would have cost more.
Hi BJL,
I believe this is precisely the reason Nikon took this path with the OLP filter implementation (pun intended post writing). Thirty years in high-tech manufacturing has taught me that it’s all about production costs and (hopefully) continuous improvement, providing a better product with each generation, while lowering production costs and price to the end user.
I agree that the effective light path must be made effectively same for both cameras to maintain consistency and commonality in the production line and avoid mechanical differences or adjustments between the two cameras. However, this might also be accomplished by using appropriate thicknesses of less expensive optical glass plates in the D800E to replace the crossed birefringent elements in the D800 (similar to what companies like LifePixel do when they remove the OLPF).
But there is one possibility I haven’t seen in this discussion that may explain Nikon’s choice to re-use the D800 birefringent filter(s) on the D800E. A while back I read a technical white paper on trends in manufacturing of digital camera sensors. It wasn’t specific to Nikon, but did mention Sony. With the goal of simplifying the manufacture of both sensors and cameras, the optical components in the filter stack on the sensors are being reduced. One of those components is the optical glass cover normally installed on the sensor package during sensor fab. (presumably done by Sony for the D800/E cameras).
The clear sensor filter is being replaced with one of the two birefringent filters, making it integral with the CMOS sensor package (and not economically feasible to remove). If this is the case with the sensor Sony delivers to Nikon for the D800/E (and I think this highly likely), it would explain Nikon’s approach of simply rotating that second birefringent filter (along with the necessity of replacing the integral wave plate/IR filter set between the two birefringent filters to eliminate the un-need wave plate). So I think Nikon’s published diagram of the D800/E filter stack is essentially correct:
http://www.nikonusa.com/en_US/IMG/Images/Learn-And-Explore/2012/Camera-Technology/D-SLR-Series/Moire-D800-D800E/Media/OLPF_schematic.pdf (http://www.nikonusa.com/en_US/IMG/Images/Learn-And-Explore/2012/Camera-Technology/D-SLR-Series/Moire-D800-D800E/Media/OLPF_schematic.pdf)
Except that it doesn’t show if the one of the birefringent plates is permanently installed on the sensor.
The cost savings here are obvious---Sony supply one identical part to Nikon for both cameras simplifying their sensor fab. line. And Nikon only has to change their line toward the end when the filter stack is installed to determine if the camera will be a D800 or D800E. Also, the reduction in components and air-glass interfaces should result in both lower production costs and higher final image quality. The savings should more than offset the possible extra cost of the birefringent filters in the D800E. A company like LifePixel could confirm if this is indeed the case for the D800 and other newer cameras. It will make a bit more difficult for them when converting such a camera to remove the OLPF. Their only choice when converting a D800 to eliminate the OLPF effect may be to do what Nikon does.
For future possibilities, there’s been a lot of progress over the past few years on the development of electrically-tunable birefringent filters. They are already being successfully used in high-end space and military imaging applications. The current costs to implement these in a high-volume consumer cameras may still be prohibitive. But I wouldn’t be surprised to see these in future digital cameras. Imagine having a dial or menu item where you can select a variable range of OLP filter effect from say 0 to 10, that you can set according to the subject!
EDIT:
And taking this concept further, since most tunable birefringent filters are based on LCD technology, with clear thin-film electrodes applied to the active LCD plates, there is no reason that the conductive layer couldn’t be patterned into a grid that would match the sensel grid (or the Bayer pattern). If placed directly over the sensel grid, the OLPF effect could be controlled for each individual sensel (or RGB group in the Bayer filter). If combined with an in-camera image processor smart enough to detect aliasing and moiré in the scene, it could dynamically, locally resolve any aliasing issues during capture for the affected areas of the image, while leaving the unaffected parts of the scene alone.
And I’ll offer one more trip into current science fiction land, based on the same concepts. It is also possible to construct electrically-tunable neutral density filters using LCD technology. Place such an ND filter grid over the sensor and it could be possible to individually control the sensitivity of each sensel. If combined with something like Tony Kuyper’s PS luminosity masks (but implemented in the camera’s image processor), the sensor response might be locally tuned to dramatically increase dynamic range during capture---operating as an adaptable, super-GND filter automatically reducing exposure in highlight areas.
Kind regards,
Bob
-
P.S. I used ImageJ's trend curve plotting, and it decided that the gamma variate curve with the c parameter representing the gamma component of the profile plot data is gamma 2.28352 (see attachment) . So besides some other non-linearities, the best gamma setting for linearization would be 0.438 .
Bart,
Please correct me if I am wrong, by my interpretation of the value to enter into Imatest is the actual gamma of the image taken of the target for testing, not the gamma of the file used to print the image for testing. That is why Norman suggests including a Q14 target along with the slanted edge target. In that way, the actual gamma can be computed with the Stepchart module.
Regards,
Bill
-
Please correct me if I am wrong, by my interpretation of the value to enter into Imatest is the actual gamma of the image taken of the target for testing, not the gamma of the file used to print the image for testing. That is why Norman suggests including a Q14 target along with the slanted edge target. In that way, the actual gamma can be computed with the Stepchart module.
That's correct, one uses the gamma of the image taken of the target (e.g. Adobe RGB = gamma 1/2.2, Prophoto RGB = gamma 1/1.8). To achieve a linear gamma for those images, Imatest requires those gammas as input. Norman Koren suggests to use 0.5 when we don't have a better calibration (such as from a stepchart).
The confusion comes from the sloppy use of the gamma term (I'm also guilty of that sloppy use, but try to use the correct values to linearize). The images are brighter than their linear gamma version, so it's the reciprocal value that was applied to counter-act (pre-compensate for) the display gamma and adjust for human vision.
Cheers,
Bart
-
1. If your head is pressed up against the display using fixed-force screws, then adding any material in-between (such as glasses) would change the physical distance.
Not that I know how sensors are aligned to the camera frame/lense.
2. If a (flat, nonconcave/convex) object is inserted whose refractive index is very different from air, could it not alter the "optical distance"?
-h
Most cameras seem to have the chip mounted from behind. You can look at the chip when the lens is removed.
If bending the light = optical distance, then we are all saying the same thing.
-
So to summarize, your theory is the AA filter is an integral part of the sensor when manufacturing, so the only option was to apply a reverse version of the filter to neutralize it's effect, since it's not removable.
I guess this makes the most sense of any theory I've heard so far ... I think all the "optical" path issues could have been resolved with options other than OLP filters.
Hi BJL,
I believe this is precisely the reason Nikon took this path with the OLP filter implementation (pun intended post writing). Thirty years in high-tech manufacturing has taught me that it’s all about production costs and (hopefully) continuous improvement, providing a better product with each generation, while lowering production costs and price to the end user.
I agree that the effective light path must be made effectively same for both cameras to maintain consistency and commonality in the production line and avoid mechanical differences or adjustments between the two cameras. However, this might also be accomplished by using appropriate thicknesses of less expensive optical glass plates in the D800E to replace the crossed birefringent elements in the D800 (similar to what companies like LifePixel do when they remove the OLPF).
But there is one possibility I haven’t seen in this discussion that may explain Nikon’s choice to re-use the D800 birefringent filter(s) on the D800E. A while back I read a technical white paper on trends in manufacturing of digital camera sensors. It wasn’t specific to Nikon, but did mention Sony. With the goal of simplifying the manufacture of both sensors and cameras, the optical components in the filter stack on the sensors are being reduced. One of those components is the optical glass cover normally installed on the sensor package during sensor fab. (presumably done by Sony for the D800/E cameras).
The clear sensor filter is being replaced with one of the two birefringent filters, making it integral with the CMOS sensor package (and not economically feasible to remove). If this is the case with the sensor Sony delivers to Nikon for the D800/E (and I think this highly likely), it would explain Nikon’s approach of simply rotating that second birefringent filter (along with the necessity of replacing the integral wave plate/IR filter set between the two birefringent filters to eliminate the un-need wave plate). So I think Nikon’s published diagram of the D800/E filter stack is essentially correct:
http://www.nikonusa.com/en_US/IMG/Images/Learn-And-Explore/2012/Camera-Technology/D-SLR-Series/Moire-D800-D800E/Media/OLPF_schematic.pdf (http://www.nikonusa.com/en_US/IMG/Images/Learn-And-Explore/2012/Camera-Technology/D-SLR-Series/Moire-D800-D800E/Media/OLPF_schematic.pdf)
Except that it doesn’t show if the one of the birefringent plates is permanently installed on the sensor.
The cost savings here are obvious---Sony supply one identical part to Nikon for both cameras simplifying their sensor fab. line. And Nikon only has to change their line toward the end when the filter stack is installed to determine if the camera will be a D800 or D800E. Also, the reduction in components and air-glass interfaces should result in both lower production costs and higher final image quality. The savings should more than offset the possible extra cost of the birefringent filters in the D800E. A company like LifePixel could confirm if this is indeed the case for the D800 and other newer cameras. It will make a bit more difficult for them when converting such a camera to remove the OLPF. Their only choice when converting a D800 to eliminate the OLPF effect may be to do what Nikon does.
For future possibilities, there’s been a lot of progress over the past few years on the development of electrically-tunable birefringent filters. They are already being successfully used in high-end space and military imaging applications. The current costs to implement these in a high-volume consumer cameras may still be prohibitive. But I wouldn’t be surprised to see these in future digital cameras. Imagine having a dial or menu item where you can select a variable range of OLP filter effect from say 0 to 10, that you can set according to the subject!
EDIT:
And taking this concept further, since most tunable birefringent filters are based on LCD technology, with clear thin-film electrodes applied to the active LCD plates, there is no reason that the conductive layer couldn’t be patterned into a grid that would match the sensel grid (or the Bayer pattern). If placed directly over the sensel grid, the OLPF effect could be controlled for each individual sensel (or RGB group in the Bayer filter). If combined with an in-camera image processor smart enough to detect aliasing and moiré in the scene, it could dynamically, locally resolve any aliasing issues during capture for the affected areas of the image, while leaving the unaffected parts of the scene alone.
And I’ll offer one more trip into current science fiction land, based on the same concepts. It is also possible to construct electrically-tunable neutral density filters using LCD technology. Place such an ND filter grid over the sensor and it could be possible to individually control the sensitivity of each sensel. If combined with something like Tony Kuyper’s PS luminosity masks (but implemented in the camera’s image processor), the sensor response might be locally tuned to dramatically increase dynamic range during capture---operating as an adaptable, super-GND filter automatically reducing exposure in highlight areas.
Kind regards,
Bob
-
So to summarize, your theory is the AA filter is an integral part of the sensor when manufacturing, so the only option was to apply a reverse version of the filter to neutralize it's effect, since it's not removable.
I guess this makes the most sense of any theory I've heard so far ... I think all the "optical" path issues could have been resolved with options other than OLP filters.
Hi Wayne,
Yes, that’s the theory. There is a lot of internet “noise” on this subject at the moment. But I did find a recent thread on photonet where Joseph Wisniewski is offering essentially the same explanation for Nikon’s approach. He seems quite sure this is how the D800E filter set is configured, but I have not yet seen firm confirmation of this by Nikon, Sony or a third party organization that has actually disassembled the D800E, like LifePixel or MaxMax.
I think perhaps Nikon is not likely to confirm this for the same reason they are reluctant to openly acknowledge that the base D800/E sensor chip may be manufactured by a competitor (Sony).
Kind regards,
Bob
-
... But I did find a recent thread on photonet where Joseph Wisniewski is offering essentially the same explanation for Nikon’s approach. He seems quite sure this is how the D800E filter set is configured ...
OK, that's fairly persuasive. Partly because I tend to trust Joe W. on topics like this: he works with sensors and such professionally, so has surely seen Sony sensors in the raw, if not this particular one. Though it just passes the buck to why Sony (or whoever fabs the sensor) cannot simply replace the Lithium Niobate slabs by cheaper optical flats. Too small a production run to be worth recomputing the slightly modified optical path cause by changing to glass?
-
OK, that's fairly persuasive. Partly because I tend to trust Joe W. on topics like this: he works with sensors and such professionally, so has surely seen Sony sensors in the raw, if not this particular one. Toughbit just passes rhe buck to why Sony (or whoever fabs the sensor) cannot dimply replace the Lithium Niobate slabs by cheaper optical flats. Too small a production run to be worth recomputing the slightly modified optical path cause by changing to glass?
Yes, but I have learned to remain skeptical until I see conclusive evidence. That’s why I called it highly-likely (but still unconfirmed). And after all, even Joe can be wrong, occasionally. If you push him on the issue, he might even admit it! ;)
Kind regards,
Bob
-
Rob Galbraith's samples (http://www.robgalbraith.com/bins/content_page.asp?cid=7-11676-12555):
Nikon D800
(http://www.robgalbraith.com/rollovers/d800_d800e/downtown_metal_d800.jpg)
Nikon D800E
(http://www.robgalbraith.com/rollovers/d800_d800e/downtown_metal_d800e.jpg)
-
Rob Galbraith's samples (http://www.robgalbraith.com/bins/content_page.asp?cid=7-11676-12555):
Nikon D800
(http://www.robgalbraith.com/rollovers/d800_d800e/downtown_metal_d800.jpg)
Nikon D800E
(http://www.robgalbraith.com/rollovers/d800_d800e/downtown_metal_d800e.jpg)
Hey! Guillermo! Someone should fix that awful barrel distortion. ;D
-
These Rob Galbraith samples of false color fringing on fine details (not just "moire", but all kinds of bogus color information caused by the bayer filter, which the anti-aliasing filter does a lot to correct) is why I am quite mystified by all the folks who prefer cameras without AA filters.
Then again, people would rather spend 2-3 minutes fiddling with the right focus plane tilt on a T/S lens for a single 36MP image than spend the same 2-3 minutes taking a couple dozen frames that will become a 120MP focus stacked and stitched image with enormously better IQ.
The high-end aspirational photo equipment marketers have trained us well! :-)
-
These Rob Galbraith samples of false color fringing on fine details (not just "moire", but all kinds of bogus color information caused by the bayer filter, which the anti-aliasing filter does a lot to correct) is why I am quite mystified by all the folks who prefer cameras without AA filters.
The answer is, selection bias. It happens all the time. We want to demonstrate that a particular thing is either undesirable, or desirable, as the case may be, then we can always do so by selecting the appropriate data that demonstrates the point we are trying to get across, whilst ignoring the contrary data.
This problem is not just confined to assessments of photographic equipment. It's a world-wide human problem.
-
Kind of a 800/800E and LR4 question. I Went to the Gailbraith site and downloaded the 800 and 800E nefs. And opened in LR4 both with auto camera adjts on and off and chromatic aberation on and off. My questions/observations. On the left side of intersection image the engine of the vertical motorcycle- on the 800E there is "stuff" not sure what to call it but it is not there or hardly there on the 800 image. And going into LR4 with the moire adjustment brush option set to plus 100 it is gone in one swipe (so far so good) on the right hand side of the Image looking at the double arrow sign and concentrating on the white areas. Is this Moire/beat pattern/atrifacting
again not sure what to call it. And the moire Adjustmnet brush in LR4 at +100 has no visible effect on either the 800 or 800E (the pattern is slightly less on the 800 image to start) And one more place just looking this time above the double arrow on the right a group of smaller signs again with lots of pattern etc. what ever it is and it is just slightly less on the 800 version. I am not posting pics at this time (have not done it and don't want to mess it up, My first post here in quit a while) but if anyone else downloads the nefs are they seeing the same things and if someone knows what to call this stuff. I am waiting for mine from B+H and have been going back and forth on which one to get. Currently have 800E ordered but these are the first raws I have seen.
-
Hello to all you lucky D800E owners. What a great sensor you now own :)
http://www.imaging-resource.com did some really great test shoots in a studio and I noticed interesting cross over points that may interest you and others debating which of the two 800's to purchase.
At ISO 400 the sharpness of the D800E is equal to the D800 at ISO 50
At ISO 400 moiré on the D800E is greatly reduced but still there if you look. The D800 has moiré too but the AA filter cleans most up but loosing some detail (D800 at ISO 50)
At ISO 800 moiré on the D800E is now starting to approach the level of the D800 yet sharpness is very similar. That increase in ISO does remove a tiny tiny bit a sharpness but noise levels are similar (D800 at ISO 50)
At ISO 1600 moiré on the D800E is just a hair way from that of the D800 at ISO 50
At ISO 3200 moiré on the D800E is now almost the same (D800 at ISO 50)
At ISO 6400 moiré on the D800E is virtually the same. Noise is now more obvious (D800 at ISO 50)
Incidentally the infamous Red Swatch that pretty much catches out every sensor made except Fovean does not affect many Nikon's and even at ISO 6400 this D800E is simply amazing. Still pulling detail from the swatch where other sensors were giving up at ISO 200 !
Like many other sensors the D800 seems to blow highlights and detail in white cloth. The D800E does not and just keeps showing detail deeper into higher ISO's
In the same still life test, the Larger Beer bottle green label is your moiré detector and I have also matched with other sources in the same images.
Now, how the shutter and aperture were changed in the tests I do not know.
Ultimately it will come down to real life tests to see if a D800E user can play a quick get out of moiré trouble trick by simply shooting at ISO 400 for as good as it could get on a D800 pics or say ISO 800 for a wee bit more added safety and hardly any loss of sharpness.
It is obvious the D800 can get into moiré trouble but the AA seems to hide it fairly well.
The ultimate moiré remover is your brain! The D800E images look so good moiré is not that offensive...
-
Agreed, but both the AA filter and the Bayer filter cause blur, so some sharpening will be beneficial on both cases (I guess preaching to the choir :)). One approach could be to show some examples both without capture sharpening and with the (subjective) capture sharpening that is found optimal in each case. Then the reader can judge the cases. Yet another approach like on dpreview (that I like) is that it is possible to double check by the reader by downloading the RAW files for individual check in the personal favorite RAW processor. When that has been said, I think people read the articles here on LuLa as personal judgement of equipment in use in contrast to the camera reviews on the major review sites where personal opinions by the reviewer takes a back seat. I think the personal opinions are both respected and valued when they are stated as such and not as absolutes. No matter how it is done there will be long threads here and there about how wrong it was done ;)
Another issue raised is the false detail from a sensor without an AA filter. It seems to me that one approach to really investigate this rather than debate the theoretical aspects of it, would be to use a much higher resolution image to check to which degree there is false detail. This could be done by zooming in on high frequency detail by going much closer to the subject and stitch the pictures to a (very) high resolution picture where the false detail will be revealed in comparison. There may be other ways to do it. I know it would be quite work intensive to do this, but it may put an end to some of the discussion and make some conclusions about this issue. And see how the AA filter equipped D800 will fare in this comparison as well.
I think this false detail is often what some see as more detail with AA-less cameras. As you said, not only does removing the AA filter cause color moire, but false detail occurs, which some apparently seem to like.
This picture posted by Joakim on the miranda forums shows a good example of this. Notice the window frames in the M9 picture.
As Joakim says, "I feel deeply sorry for the poor unfortunate souls that live in the building to the left, the contractor obviously used scrap material for the entire project. None of the window-sills or frames have the same measurements / thicknesses. Doesn't bode well for the internal structures..." :)
(http://i1101.photobucket.com/albums/g429/theSuedeII/M9C3.jpg)
-
As Joakim says, "I feel deeply sorry for the poor unfortunate souls that live in the building to the left, the contractor obviously used scrap material for the entire project. None of the window-sills or frames have the same measurements / thicknesses. Doesn't bode well for the internal structures..." :)
Note the ladder (?) up on the roof to the right
-
One thought on the reversed AA filter is each layer has some Infra red and UV filtering properties
perhaps Nikon wanted to keep those properties constant between cameras and finding optical glass the same thickness and with the same filtering properties didn't make sense just stacking a reversed AA filter got them what they wanted without having to inventory extra parts, clever.
Marc
-
I haven't seen any review so far mention if D800 has silent first curtain (electronic shutter) for shooting in live view. It's on of the advantages that the Canon 5D mkII has over e.g. my 1Ds mkIII which does not have this feature. Even MLU is not as good as electronic shutter in live view since absolutely no mechanical movement results in starting the exposure.
-
I haven't seen any review so far mention if D800 has silent first curtain (electronic shutter) for shooting in live view. It's on of the advantages that the Canon 5D mkII has over e.g. my 1Ds mkIII which does not have this feature. Even MLU is not as good as electronic shutter in live view since absolutely no mechanical movement results in starting the exposure.
a
As far as I know you can use Liveview in the d800 in a way only the shutter opens and closes- so it is not completely silent.
For Nikon it is huge progress for it used to move the mirror as well. ( now they have an extra motor to deal with the mirror - to let it stay open it does not need power anymore)
-
a
As far as I know you can use Liveview in the d800 in a way only the shutter opens and closes- so it is not completely silent.
For Nikon it is huge progress for it use to move the mirror as well. ( now they have an extra motor to deal with the mirror - to let it stay open it does not need power anymore)
On my 1Ds mkIII when in live view and take a picture in live view, there is a devastating vibration from the shutter with a fairly long lens and in the critical range of 1 second and 1/30s. This creates really visible blur. What happens is that the the shutter closes and then opens again to start the exposure. The mirror does not move, of course. But the sheer inertia of the shutter is enough to cause blurred images. So if this is how the D800 does it, then my advice is not to shoot in live view. What I do with the 1Ds mkIII is to use live view for critical focusing which includes checking DOF and then exit live view and use MLU for taking the picture(s). For the critical exposure ranges, I will use my cable release and open the mirror by one click and then wait 3-4 seconds and then take the picture. If I bracket I will step through the bracketing sequence using the cable release and click the mirror up and wait for each exposure. This works fine, but I would rather have live view as an option to take the pictures.
-
On my 1Ds mkIII when in live view and take a picture in live view, there is a devastating vibration from the shutter with a fairly long lens and in the critical range of 1 second and 1/30s. This creates really visible blur. What happens is that the the shutter closes and then opens again to start the exposure. The mirror does not move, of course. But the sheer inertia of the shutter is enough to cause blurred images. So if this is how the D800 does it, then my advice is not to shoot in live view. What I do with the 1Ds mkIII is to use live view for critical focusing which includes checking DOF and then exit live view and use MLU for taking the picture(s). For the critical exposure ranges, I will use my cable release and open the mirror by one click and then wait 3-4 seconds and then take the picture. If I bracket I will step through the bracketing sequence using the cable release and click the mirror up and wait for each exposure. This works fine, but I would rather have live view as an option to take the pictures.
This is another reason the Sony Alpha SLT version with the 36MP chip will have some really nice advantages for practical photo-taking, particularly around stability and sharp images:
* built in stabilization with all lenses
* electronic first curtain shutter
* no mirror slap ever (no need for fiddly MLU type functions)
And also
* ease of nailing exposure via preshot live histogram and/or WYSIWYGish EVF display.
-
This is another reason the Sony Alpha SLT version with the 36MP chip will have some really nice advantages for practical photo-taking, particularly around stability and sharp images:
* built in stabilization with all lenses
* electronic first curtain shutter
* no mirror slap ever (no need for fiddly MLU type functions)
And also
* ease of nailing exposure via preshot live histogram and/or WYSIWYGish EVF display.
Well, it was not to start a discussion on different cameras, but simply a question about the D800.
The Canon 5D mkII and mkIII have the ability to shoot without any physical movement, so it's not needed to go to mirrorless to get that. The Canon also has live histogram and the SLT is a crop sensor and 24MP.
-
These Rob Galbraith samples of false color fringing on fine details (not just "moire", but all kinds of bogus color information caused by the bayer filter, which the anti-aliasing filter does a lot to correct) is why I am quite mystified by all the folks who prefer cameras without AA filters.)
I can solve this mystery: because I spend very little time photographing apartment balconies, and prefer slightly greater post-production effort to clean-up any false colourto gain the benefit of an image with slightly greater resolution and accutance,visible to my eye (and to my surprise) even on these web-pegs.
The odds of any of the artefacts uniquely appearing on the 800E impacting a finished print are also so close to zero as to make it a non-issue for me. Most photographers will do vastly greater visible damage to their prints through bad sharpening than a few colour artefacts could wreak.
- N.
-
Well, it was not to start a discussion on different cameras, but simply a question about the D800.
The Canon 5D mkII and mkIII have the ability to shoot without any physical movement, so it's not needed to go to mirrorless to get that. The Canon also has live histogram and the SLT is a crop sensor and 24MP.
I was referring to the upcoming FF 36MP Sony, not the current APS 24MP Sony SLT models.
The Nikon does not offer Electronic First Curtain Shutter as far as I can determine from googling. That is unfortunate, as it makes a big difference with long lenses.
Regarding the APS Alpha 65 and 77 -- both offer IQ comparable to the 5D3 at low ISOs with a bit more DR for the 77 than the 5D3 (and no shadow banding problems).
SLT definitely offers stability advantages for anyone who ever shoots handheld. I am definitely looking forward to playing with the new Sony Alpha.
-
I was referring to the upcoming FF 36MP Sony, not the current APS 24MP Sony SLT models.
The Nikon does not offer Electronic First Curtain Shutter as far as I can determine from googling. That is unfortunate, as it makes a big difference with long lenses.
Regarding the APS Alpha 65 and 77 -- both offer IQ comparable to the 5D3 at low ISOs with a bit more DR for the 77 than the 5D3 (and no shadow banding problems).
SLT definitely offers stability advantages for anyone who ever shoots handheld. I am definitely looking forward to playing with the new Sony Alpha.
I haven't seen an announcement from Sony about a full frame (36MP) SLT model or even a successor the DSLR models.
I don't think the APS-C models offer similar IQ (well anything can be compared ;)) as the full frame 5D mkIII. I haven't so far seen an APS-C sensor providing the same IQ as a full frame sensor with similar amounts of pixels. 2,25 to 2,6 times more are of glass does help the system MTF one should think.
I do agree that a mirrorless system will have advantages for shooting hand held where the mirror slap causes vibrations.
-
The Canon 5D mkII and mkIII have the ability to shoot without any physical movement ...
How do they do that? No mirror movement maybe, but I think there is still shutter movement, because AFAIK, none of the DSLR sensors so far has a global electronic shutter for stills. So I would think it is a matter of _less_ physical movement, in the same way as with a rangefinder camera.
-
Hi,
I have both the Alpha 77 SLT and the Alpha 900, the SLT is APS-C and the Alpha 900 is FF. Both are 24 MP. My feeling so far is that the SLT is a pretty fair match for the Alpha 900, but much more demanding on lenses. I have not done any "scientific testing", though.
The SLT has some advantage in better DoF at equal aperture and live view AF is a real benefit. Right now I use the SLT for tele work, for normal work, either does fine. The Alpha 900 has better viewfinder but lacks live view. The Alpha has a virtual horizon I miss on the Alpha 900.
As is now, I would not buy an Alpha 99 with 24 MP, as it would offer little benefit over the combo I have. An Alpha 99 with 36 MP would be much more interesting. I would also say that user interface is important for me. So would the Alpha 99 arrive with 24MP and a really excellent UI, I may buy it anyway.
Best regards
Erik
-
How do they do that? No mirror movement maybe, but I think there is still shutter movement, because AFAIK, none of the DSLR sensors so far has a global electronic shutter for stills. So I would think it is a matter of _less_ physical movement, in the same way as with a rangefinder camera.
Yes, the 5D mkII, 5D mkIII and 7D (and some other Canon cameras) has electronic first curtain in live view, so no shutter movement as I said. They do an electronic "reset" of the sensor to start the exposure.
-
Yes, the 5D mkII, 5D mkIII and 7D (and some other Canon cameras) has electronic first curtain in live view, so no shutter movement as I said. They do an electronic "reset" of the sensor to start the exposure.
Thanks; so no that I bothered to read up on it, this reduces the mechanical movement to that of the second curtain (the one that ends exposure at each part of the sensor) with the start of the exposure done electronically by resetting the photosites in an incremental sweep across the sensor, rather than by incrementally starting exposure with the mechanical motion of the first curtain.
-
Thanks; so no that I bothered to read up on it, this reduces the mechanical movement to that of the second curtain (the one that ends exposure at each part of the sensor) with the start of the exposure done electronically by resetting the photosites in an incremental sweep across the sensor, rather than by incrementally starting exposure with the mechanical motion of the first curtain.
Yes, the exposure is ended by the mechanical shutter.
It would be interesting if any of the D800 reviewers and owners would do a simple test to check if it use the mechanical shutter to start the exposure.
The test is very simple. Use low ISO in a situation where the exposure time would be long, say 10 seconds, so it is easy to hear if there is a mechanical movement at all to start the exposure.
-
Yes, the exposure is ended by the mechanical shutter.
It would be interesting if any of the D800 reviewers and owners would do a simple test to check if it use the mechanical shutter to start the exposure.
The test is very simple. Use low ISO in a situation where the exposure time would be long, say 10 seconds, so it is easy to hear if there is a mechanical movement at all to start the exposure.
Hans, yes, it does use the mechanical shutter to begin an exposure in LV (tested by setting MUP, LV and triggering the shutter).
-
Hans, yes, it does use the mechanical shutter to begin an exposure in LV (tested by setting MUP, LV and triggering the shutter).
Thanks and that is disappointing. I wonder why this is, since the benefit for shooting on slow shutter speeds on a tripod is so obvious a help for a camera with a high resolution as the D800.
-
Thanks and that is disappointing. I wonder why this is, since the benefit for shooting on slow shutter speeds on a tripod is so obvious a help for a camera with a high resolution as the D800.
Hi Hans,
Probably because the sensor benefits from being reset in total darkness, with the shutter curtains closed. Besides, although I know how critical camera shake is with long FL lenses (and with > 1:1 macro!), the fact that the camera still shakes shows that the support is not adequate enough, unfortunately.
Cheers,
Bart
-
Besides, although I know how critical camera shake is with long FL lenses (and with > 1:1 macro!), the fact that the camera still shakes shows that the support is not adequate enough, unfortunately.
Cheers,
Bart
So how heavy/rigid an object (stand) do you have to fix the camera to in order to make the acceleration of mechanical parts not move/rotate the camera significantly compared to one sensel (or an equivalent fraction of a degree for some focal length)?
-h
-
So how heavy/rigid an object (stand) do you have to fix the camera to in order to make the acceleration of mechanical parts not move/rotate the camera significantly compared to one sensel (or an equivalent fraction of a degree for some focal length)?
Hi h,
That depends on how the lens camera is mounted. When the support is positioned closely beneath the center of gravity it can be lighter, but if not only inertia plays a role but vibration (e.g. wind or trafic) as well, then carbon fiber or wood have better damping characteristics.
There is a reason why stuff like this (http://reallyrightstuff.com/ProductDesc.aspx?code=CB-YS-QR-Pkg&type=3&eq=CB-YS-QR-Pkg-001&desc=Lens-Support-Pkg%2c-dual-Quick-Release&key=ait) is sold. For really critical situations 2 tripods can be used, one for the camera body and one for the lens.
The need for such drastic measures can be determined by shooting the stationary laser dot projected by a separated laser level, or pen. Or one can tie it to the lens and visually inspect how much the dot dances when the tripod or the camera is touched/tapped.
Cheers,
Bart
-
A couple of test shots from my balcony, D800E, tripod, hanging bag, liveview and cable release. Capture 1 highlight slider 99% focus fixer .7 f4.5
first @50mm f4.5 Leica 28-90 R
second @270mm f4.5 Leica 270 4.0 R
third and fourth 100% crops @ 270mm f4.5
Wow!
Marc
The posterizing is a product of downloading to the LL server the jpegs looked clean on my screen!