Absurd DxO D800E results: Are low pass filters a joke?

[Bart_van_der_Wolf]

Why cant they make a screw on filter for when you need the AA? Just like putting a UV filter on the lens.

Hi Arthur,

Because it becomes very difficult to restrict the blur to a very tiny region. When also lenses get involved, the effect will also be very much variable due to the variable lens aberrations. Some claim to have a patent pending product that requires some sort of Raw conversion adjustment. It also seems to be aimed at video resolutions, so there is probably a lot of blurring at the full resolution level. There do seem to be some issues with image quality ...

Cheers,
Bart

[Fine_Art]

Maybe someone can test the black panty hose trick. It's been around forever in film, what is the impact on a AA free digital camera?

[Jack Hogan]

Why cant they make a screw on filter for when you need the AA? Just like putting a UV filter on the lens.

Apparently the Pentax K3 uses its sensor image stabilization system to produce an OLPF effect at will.  Don't know how well it works compared to a fixed AA but it seems like a pretty smart approach.

[Jack Hogan]

 This is an interesting topic. I would like to take this further. In particular, I would like to understand what MTF curves for a real-world AA filter looks like.

In the frequency domain an ideal 4-dot beam splittin' AA looks like a cosine that hits first zero at 1/(4x) cycles per pixel, +/- x being the shift in pixels introduced by it.  In an MTF graph it is an absolute value so it bounces back after the zero.  x tends to be in the 0.35-0.4 pixel range for most current (Nikon) cameras.  Here is a theoretical example on a D610 (dotted blue line, x=0.39), it should look very similar on a D800:



It is hard to isolate because one needs to keep everything the same in the AA and AAless version in order to measure the AA effect accurately.  I think I was able to do it on the D610 and A7 because they appear to have antialiasing action in one direction only, so I was able to use green channel raw data from a single capture in each case:




Things get unreliable after the zero because I am taking the ratio of two small noisy values.  You can read more about where these charts came from here.

Jack

[jduncan]

I saw that, but does it contradict some of the other information around on the resolution differences between the two models ?
Anyway, it's an amazing camera the E..

A newer model with a bit of an update would be welcome here..

I believe that is because lenses. Most lenses will not show.Look at DxO data and you will see a bunch of lenses  (that are good in many ways) rendering
13 or less perceptual megapixels.  So with that glass the difference between a D600 and a D800E will be less than expected, let alone D800 vs D800E.

Resolution is not the all  there is for a lens. Some times resolution is not even the key parameter.  never the less I can't justify a design like the Nikon 55mm f1.4 noc simply because resolution is too bad, but in a D4 it will be all is needed.

I want to see if the D800E can differentiate between  average resolutions of the Sigma 50mm Art an the 55mm otus.
I continue to believe that the 135mm Zeiss is like the best lens there is  (taking into account price), unless you need autofocus.
Zeiss should license Sigma or Tamron auto focus know-how, or use the screw mechanism. Waiting for authorization by Nikon or Canon is non-sense.


Best regards,
J. Duncan

[jduncan]

Now the tests are all shot wide open correct?
I never shoot wide open, if ever above f4 mostly 5.6-11.
Would any diffraction still equalize the bodies at smaller apertures.
Before I bought the d800 everyone said "the difference is so minor" "sharpening will minimize any difference"
These tests seem to indicate a pretty huge gap. What would be a more pragmatic reading?
Would I really be gaining any noticeable difference with the "e"?
Of course I'm happy with my non-e but thinking if I could ramp up sharpness that much....

The article states that  wide open the otus matches or exceeds the  135mm at f2.8  but at  f4.0 the 135mm beets everyone else, at any aperture,  basically. So the tests are done at multiple apertures.
Also when you use the tool for comparing lens you can change the aperture (and the focal length ) and see the results with an specific camera.

The best part of DxO mark is having an standard. I hope that alternatives come along so we can use multiple sources for comparing lenses and sensors.
But having a standard is better than no standard  and marketing scams about looks.

Best regards,
J. Duncan

[BernardLanguillier]

Resolution is not the all  there is for a lens. Some times resolution is not even the key parameter.  never the less I can't justify a design like the Nikon 55mm f1.4 noc simply because resolution is too bad, but in a D4 it will be all is needed.

Interestingly, the Nikon 58mm f1.4 has just been awarded lens of the year in Japan.

Cheers,
Bernard

[Dave Ellis]

Hi Jack

That's very interesting information you've provided. It gives a very good summary of how the different elements of the system contribute to MTF. For me, this is just the sort of information you need to make a proper distinction between lens performance and camera performance.

The AA blur in one direction only is interesting but I'm struggling to understand the rationale behind it !

Dave

In the frequency domain an ideal 4-dot beam splittin' AA looks like a cosine that hits first zero at 1/(4x) cycles per pixel, +/- x being the shift in pixels introduced by it.  In an MTF graph it is an absolute value so it bounces back after the zero.  x tends to be in the 0.35-0.4 pixel range for most current (Nikon) cameras.  Here is a theoretical example on a D610 (dotted blue line, x=0.39), it should look very similar on a D800:



It is hard to isolate because one needs to keep everything the same in the AA and AAless version in order to measure the AA effect accurately.  I think I was able to do it on the D610 and A7 because they appear to have antialiasing action in one direction only, so I was able to use green channel raw data from a single capture in each case:




Things get unreliable after the zero because I am taking the ratio of two small noisy values.  You can read more about where these charts came from here.

Jack

[Bart_van_der_Wolf]

The AA blur in one direction only is interesting but I'm struggling to understand the rationale behind it !

Hi Dave,

Perhaps to reduce Video resolution aliasing, but not lose too much still image resolution?

Cheers,
Bart

[Dave Ellis]

Hi Bart

Yes video shooters seem to be more concerned about AA than still shooters ! I don't know much about that aspect of video but I think that aliasing is probably accentuated with video due to variations in it from frame to frame, which would give a "motion" effect to the aliasing.

With AA blur in the vertical direction only, you'll get a varying amount of alias reduction depending on the direction of the high frequency detail in the image. For high frequencies purely in the vertical direction there will be no alias reduction whereas in the horizontal direction there will be maximum alias reduction. In general, something in between the two extremes will occur.

Dave

Hi Dave,

Perhaps to reduce Video resolution aliasing, but not lose too much still image resolution?

Cheers,
Bart

[fdisilvestro]

THD is a measure of departure from linearity. So far, in this discussion, we've assumed linearity, so I'm not sure what your point is here. Maybe you could expand on that.

In my experience, if you stay away from electron counts near the full well capacity, where there is oftem some compression, today's sensors are remarkably linear.

I'm assuming that your filter comment is unrelated to your THD comment, so I'll assume linearity here. I don't know what you mean by a high-quality signal in this context. Certainly, one of the consequences of steep filter skirts is large phase shifts. That's important in audio. Is that what you mean?

I've wondered about the effect of spatial phase shifts in imaging before, but have not reached any conclusions. They are explicitly ignored in MTF analysis. Is that a bad thing?

Jim

Thanks for your comment. I might have been using the THD term in an imprecise way, since it is usually related to the harmonics created as a result of non-linearity in systems.

Similar to THD, I'm referring to the weight or power of artificially created output signal during sampling or reconstruction, which is not part of the original signal (THD considers harmonic frequencies caused by non-linearties).

In digital imaging we can have aliasing and artifacts caused by the interpolation mehtod.

My point, and I may be wrong, is that as you get closer to Nyquist frequency, the weight of those "artificial components of the signal" or artifacts, starts to weight more and degrade the quality of the signal.

Regarding phase shift, I have no idea if the affect the output.

On another note, related to the non-intuitive reasoning about why you don't need the AA filter if you lenses are not that great, you may think that you need only one low pass filter, not two.

Finally, does anybody knows if the DXO test comparing the D800 and D800E used unsharpened images?

Regards

[Bart_van_der_Wolf]

Finally, does anybody knows if the DXO test comparing the D800 and D800E used unsharpened images?

Hi Francisco,

Sharpening affects the MTF (as does demosaicing) and the visibility of aliasing. Since no sharpening was specified, we can only assume that it was omitted. I'm pretty sure that the folks at DxO would have mentioned it if sharpening had been applied, because the way they could do it would reduce the differences between center and corners.

Cheers,
Bart

[fdisilvestro]

Hi Bart,

Thanks for your comment, It is just that I consider capture sharpening (preferably by deconvolution) as an essential part of the process, especially if there is an AA filter. I understand it is difficult to include it in a comparative test and not cause more controversy.

It would be interesting if DXO applied their own "lens softness" algorithm together with the corresponding camera/lens combinations and post the results for this comparative test. My guess is that the differences will be less dramatic

Regards,

[eronald]

Hi Bart,

 My guess is that the differences will be less dramatic

Regards,

I think this is actually the issue at debate: are the 800 and E similar in real life or not? My reading of the DxO data says "not".

And of course most of us here wonder how much better sharpening/deconvolution/motion cancellation work when you don't throw high frequency information away deliberately.

Edmund

[fdisilvestro]

And of course most of us here wonder how much better sharpening/deconvolution/motion cancellation work when you don't throw high frequency information away deliberately.

Edmund

I think it is a compromise between high frequency information thrown away vs. information "created" or not present in the original signal (aliasing + artifacts). Now, due to the way we perceive images, some people might actually like the effect of those artifacts, giving the idea of increased sharpness.

[Bart_van_der_Wolf]

I think that guess, that the D800 and E are very similar in real life with "normal" lenses and/or mediocre shooting conditions was the general assumption; whether it is true is the issue at the heart of this thread.

Hi Edmund,

Good to have you back from your mishap.

As we can see in other threads here on LuLa and elsewhere, sample density (which is identical between the D800 and D800E) and lens quality are two main deciders for image resolution. All the OLPF does is reduce the modulation of the highest spatial frequencies more than of lower spatial frequencies, but it does not eliminate them all together. The net loss of limiting resolution due to the unmodified OLPF in the D800 was measured as less than 1%, but at the same time the amplitude of aliasing was reduced.

So I am also a bit surprised by the significantly different MTF50 + perception based scores of DxOmark. It must be the perception component (Contrast Sensitivity Function) that weighs in that heavily. Of course, deconvolution sharpening would boost the spatial frequency at which the MTF50 is achieved, and it would boost the modulation at the peak of Contrast sensitivity. It's also easier to deconvolve an OLP filtered image without boosting artifacts than from an unfiltered image.

That would suggest that the Mpix score leaves a lot of real life (which always requires Capture sharpening) quality perception out of the metric, thus reducing its predictive importance. It's just a benchmark, and it leaves a lot of questions unanswered (as most single figure metrics do). In this case it even raises questions, because the lenses in the mix are that good (they do not level the remaining differences).

In particular, it may turn out that deconvolution/sharpening/motion cancellation and lens aberration compensation works a great deal better when information has not been deliberately destroyed

I fully agree. And as other experiments have shown, even a properly behaved destruction (f/32 diffraction with known PSF kernel) can be recovered from, to a large extent. But when e.g. aliasing artifacts get into the mix, then things become very difficult to repair. GIGO still rules.

Of course, aliasing is only an issue in the narrow plane of focus, so not all subject matter is affected in the same way. Stopping down beyond f/5.6 also starts reducing the issues (and differences) on the sensel pitch of the D800/E due to diffraction. Also remember that the DxO mark tests were cherry-picking the best apertures, presumably around f/4, where diffraction does not yet affect the scores as much. Comparing them at e.g. f/8 would have given much closer scores already.

Cheers,
Bart

[eronald]

Hi Edmund,

Good to have you back from your mishap.

As we can see in other threads here on LuLa and elsewhere, sample density (which is identical between the D800 and D800E) and lens quality are two main deciders for image resolution. All the OLPF does is reduce the modulation of the highest spatial frequencies more than of lower spatial frequencies, but it does not eliminate them all together. The net loss of limiting resolution due to the unmodified OLPF in the D800 was measured as less than 1%, but at the same time the amplitude of aliasing was reduced.

So I am also a bit surprised by the significantly different MTF50 + perception based scores of DxOmark. It must be the perception component (Contrast Sensitivity Function) that weighs in that heavily. Of course, deconvolution sharpening would boost the spatial frequency at which the MTF50 is achieved, and it would boost the modulation at the peak of Contrast sensitivity. It's also easier to deconvolve an OLP filtered image without boosting artifacts than from an unfiltered image.

That would suggest that the Mpix score leaves a lot of real life (which always requires Capture sharpening) quality perception out of the metric, thus reducing its predictive importance. It's just a benchmark, and it leaves a lot of questions unanswered (as most single figure metrics do). In this case it even raises questions, because the lenses in the mix are that good (they do not level the remaining differences).

I fully agree. And as other experiments have shown, even a properly behaved destruction (f/32 diffraction with known PSF kernel) can be recovered from, to a large extent. But when e.g. aliasing artifacts get into the mix, then things become very difficult to repair. GIGO still rules.

Of course, aliasing is only an issue in the narrow plane of focus, so not all subject matter is affected in the same way. Stopping down beyond f/5.6 also starts reducing the issues (and differences) on the sensel pitch of the D800/E due to diffraction. Also remember that the DxO mark tests were cherry-picking the best apertures, presumably around f/4, where diffraction does not yet affect the scores as much. Comparing them at e.g. f/8 would have given much closer scores already.

Cheers,
Bart

Bart,

I think it might be worthwhile running some real world tests or spatial domain simulations. The remark about narrow plane of focus may or may not apply to landscape and art repro shooters who are some of the untypical clusters found on this forum.

Also, as Francisco alludes to, the whole issue is clouded by the question of what gives "sparkle" to an image, and how texture is perceived. Head hair, beard stubble, skin, eye detail, feathers etc.

In fact Francisco makes an interesting point: Film grain became an integral and expected part of the "dark" photographic image, and aliasing artefacts may now be culturally expected as an indication of sharpness in digital photos.

I don't feel well served by the standard MTF results, or indeed my own MTF lens tests, in the sense of predicting subjective sharpness, while I do find that DxO's DR and noise figures reflect and predict quite well what I see in the field when using a camera.

I guess if I were more experienced I would feel more comfortable with frequency arguments, but at this point I think spatial simulation might be more illuminating. I wonder whether resampling/filtering/re-enhancing some typical dSLR images down to VGA resolution might not supply the desired enlightenment with low experimental overhead.

Edmund

PS. I do wonder is a decent camera as good as a decent hifi or is it as bad as a transistor radio? Are we really seeing the texture or does an image "only" convey the same amount of information about skin and hair as the Venus de Milo?

[Bart_van_der_Wolf]

Bart,

I think it might be worthwhile running some real world tests or spatial domain simulations. The remark about narrow plane of focus may or may not apply to landscape and art repro shooters who are some of the untypical clusters found on this forum.

Hi Edmund,

I agree, real images are better than charts or single number metrics, but the difficulty with visual comparisons is that they are complex. Many different types of subject matter, and many different post-processing paths, leads to complex comparisons.
 

Also, as Francisco alludes to, the whole issue is clouded by the question of what gives "sparkle" to an image, and how texture is perceived. Head hair, beard stubble, skin, eye detail, feathers etc.

I absolutely agree. As I tried to demonstrate, even the first step in a sharpening workflow, Capture sharpening, can create a level playing field for subsequent post-processing and comparison, but our tools make it hard to achieve it. That's even before Creative sharpening! And even something like Clarity can be implemented in hugely different ways and again change the look of an image immensely. And then there is 'taste', or the lack of it.

In fact Francisco makes an interesting point: Film grain became an integral and expected part of the "dark" photographic image, and aliasing artefacts may now be culturally expected as an indication of sharpness in digital photos.

I agree that expectations have something to do with it, but I've never seen a noisy sky when I look at the real thing, or a stairstepped straight edge with halos. So when I want to capture reality, I'm not going to create an abstraction. When I do want to create an effect/abstraction, anything goes, even creating an image from a photograph (or some over the top HDR tonemapping). However, when artifacts start to distract from conveying the emotion or message, then something needs to be improved, IMHO.

I don't feel well served by the standard MTF results, or indeed my own MTF lens tests, in the sense of predicting subjective sharpness, while I do find that DxO's DR and noise figures reflect and predict quite well what I see in the field when using a camera.

I guess if I were more experienced I would feel more comfortable with frequency arguments, but at this point I think spatial simulation might be more illuminating. I wonder whether resampling/filtering/re-enhancing some typical dSLR images down to VGA resolution might not supply the desired enlightenment with low experimental overhead.

It's tough to devise an objective comparison. Just to illustrate one of the potential variables, look at the two attached examples (first from the "optimal Capture sharpening" thread, second with added Detail / Creative sharpening) and compare. A viewing distance of some 6 feet or 2 metres, at 100% display zoom might give a better sense of detail with our low resolution displays. Also observe how much apparent resolution is gained at closer viewing distances, because I improved spatial frequencies for closer viewing more than those for more distant viewing.

So without a rigorous regime of shooting, processing, viewing distance, and subject standardization, most images can be made to look like another, or something different, very easily.

PS. I do wonder is a decent camera as good as a decent hifi or is it as bad as a transistor radio? Are we really seeing the texture or does an image "only" convey the same amount of information about skin and hair as the Venus de Milo?

LOL, define decent HiFi (and how large a room and its acoustical properties) ...

Cheers,
Bart

[Ray]

We should all know by now that expectations can result in a placebo effect, on all matters, whether it be testing the efficacy of a new drug, or the clarity and 3-dimensionality of a hi-fi or a photographic image experience.

To resolve such matters you need the double blind test. That is, a comparison in which the viewers or listeners have no knowledge of the credentials, brand, or model of the equipment being used to produce the sound or images.

In the world of hi fi, such testing has produced remarkable results. Subtleties of amplifier harmonic distortion often get drowned in the larger deficiencies of loudspeaker performance and room acoustics, often resulting in expensive amplifiers with ultra-low harmonic distortion serving no practical purpose.

Likewise, subtleties of resolution differences in images can get lost depending on print size and viewing distance.

Without a direct comparison, at very large sizes, of D800 and D800E images of the same subject using the same lens and shooting methodology, there can be no meaningful conclusion.
Such comparisons would of course have to include unsharpened images, as well as images sharpened with a variety of different techniques and programs.

[Simon Garrett]

We should all know by now that expectations can result in a placebo effect, on all matters, whether it be testing the efficacy of a new drug, or the clarity and 3-dimensionality of a hi-fi or a photographic image experience.

Reports of miracle improvements from the recent D800/D800E firmware are case in point.

On dpreview people were reporting changes in the new firmware including not just the changes that Nikon described but also faster AF, lower noise and I'm sure someone said it now cures warts.