D800E Moire question

[BJL]

Actually I used to focus on my 39MP back for the moire.

Aha: moiré focusing is the new grain focusing (for dark-room printing old-timers).

[JohnHeerema]

While a poor lens design, or diffraction, or defocus, all help to reduce the amplitude of the higher spatial frequencies, they are much less efficient than an AA-filter. An AA-filtered image can be largely restored to the original detail level by deconvolution sharpening, but without many of the artifacts. An image taken without an AA-filter will only suffer more from sharpening.

Having just ordered a Nikon D800E, I've also been interested in the question of whether to get the version with anti-alias filtering or the version without.

Signal processing theory tells us that since the spatial size of a chroma sensel is double the size of a luminance sensel in each axis, colour aliasing will occur for frequencies corresponding to range between the luminance sensel pitch and the chroma sensel pitch.

Bart van der Wolf has summarized it very nicely, earlier in this thread:

One of the design trade-offs is that the Green and the Red/Blue filtered sensels have a different Nyquist frequency for the chroma part of their information. The Luminance information has a uniform Nyquist frequency because all sensels contribute some luminance information. It's the difference between luminance and (Green vs Red/Blue) chroma aliasing that can cause false color artifacts. AA-filters reduce that risk. They also reduce stairstepping (AKA jaggies) on high contrast edges and lines. It is not only repetitive structures near the limiting resolution that cause aliasing artifacts, but they stand out more because they disrupt a predictable pattern, something human vision is sensitive to. 

For me, the question is the degree to which a raw converter can identify and remove colour aliasing artifacts such as moire. Deconvolution kernels implicitly assume that there is spatially uniform aliasing, which we know is not the case.

While aliasing occurs for all frequencies within the vulnerable range, casual observation tells us that these frequencies are not uniformly present in the subjects that we photograph. My personal belief is that what is needed is a context-aware analysis of the information from the Bayer array, that seeks to identify the lower-frequency manifestations of colour aliasing, and either removes them, or restores them to their correct frequency range.

I have a more than casual personal interest in this kind of signal analysis, as my PhD research is in this area - time-frequency analysis of sparsely sampled, rapidly changing signals. My gut feeling is that it needs to be done in the raw converter, but that significantly more can be done than we are seeing right now.

Lightroom takes the approach of having the photographer identify areas of interest, which I think is a good first step towards building an arsenal of tools that can deal with colour aliasing after the fact.

[madmanchan]

Moire can show in any image regardless of the res (for higher-res sensors like D800/D800E, just step farther back ... eventually your scene details will be near pixel-sized and then they will alias).  If it does show up, simply downsizing the image is often not going to help ... the artifacts may still be clearly visible on web-sized output, even if starting with a huge original.  That's because the nature of moire and (more generally) aliasing means that high-frequency details that can't be accurately recorded will "masquerade" as bogus low frequencies ... like big oscillating color bands which are visible even in small-sized output.

[JeffKohn]

Jeff, whats the other thing?

Edge aliasing and color aliasing. It's the color aliasing that I find particularly objectionable ("color sparklies" in fine details or along edges). Some people just don't seem bothered by them, to me they make an image look over-sharpened and digital.

[benmar]

I have a lot of experience with moire, since I shoot a lot of fabrics (tethered shooting). Used to have more trouble when I shot medium format backs- no AA filter, but now much less shooting 5Dmk2. Sharply defined moire is often very hard to get rid of with software (C1 moire tool, and the old PhaseOne De-Moirize Photoshop plugin which only worked through CS2) since the color part of the moire gets eliminated, leaving a luminosity pattern moire in its place. Stopping way down, merging different captures with different focus have worked for me. I'll be interested to see how well the Lightroom 4 brush works.

[jonathanlung]

Not sure what you are getting at here, unless you are nitpicking on terms.

Sorry; I wasn't trying to nitpick. I'm trying to map your experiences onto what I might expect if I spring for a D800E and I thought that the resolution of the sensor was what mattered (when attempting to capture the same image in different formats). Thus, I wasn't sure whether your experiences with your 39MP back were more reflective of what could be expected with a D800E than your experiences with the 60MP.

Edit: I'm still not sure if I'm any closer to my answer; I know that pixel pitch determines moiré, but to get the same FOV and DOF and resolution, doesn't the pixel pitch scale proportionally with the format and, hence, two sensors with the same number of MPs should be equally susceptible to moiré? If I'm being dense — or simply not making any sense, I blame having caught an early flight this morning and still being awake.

[BernardLanguillier]

Edit: I'm still not sure if I'm any closer to my answer; I know that pixel pitch determines moiré, but to get the same FOV and DOF and resolution, doesn't the pixel pitch scale proportionally with the format and, hence, two sensors with the same number of MPs should be equally susceptible to moiré? If I'm being dense — or simply not making any sense, I blame having caught an early flight this morning and still being awake.

You are correct, resolution is the only thing that matters when comparing the chance of having moire of 2 camera system with sensors of different size (assuming everything else is equal).

Now, the moire will result of an similarity of frequency btwn the details in the image and the pixel pitch. It makes it rather easy to compute the kind of authorized framing for a given subject depending on the size of the repeated patterns (typically in clothes).

Conversely it also makes it easy to identify a worst case scenario that should generate moire so as to understand to what extend a given camera make is likely to exhibit moire.

Besides, in camera processing if the raw data will most probably play a major role also.

Cheers,
Bernard

[ErikKaffehr]

Hi,

Moiré arises when the lens has sufficient fine detail contrast at the pixel level. If the lens cannot transfer enough contrast between two pixels moiré will not arise.

Stopping down reduces contrast at the pixel level because of diffraction and can therefore reduce or eliminate moiré.

So, if you have say 5 micron pixels f/8 would be good enough to eliminate moiré, but at 9 micron pixels f/16 would be needed.

Best regards
Erik

Edit: I'm still not sure if I'm any closer to my answer; I know that pixel pitch determines moiré, but to get the same FOV and DOF and resolution, doesn't the pixel pitch scale proportionally with the format and, hence, two sensors with the same number of MPs should be equally susceptible to moiré? If I'm being dense — or simply not making any sense, I blame having caught an early flight this morning and still being awake.

[Bart_van_der_Wolf]

Edit: I'm still not sure if I'm any closer to my answer; I know that pixel pitch determines moiré, but to get the same FOV and DOF and resolution, doesn't the pixel pitch scale proportionally with the format and, hence, two sensors with the same number of MPs should be equally susceptible to moiré?

Hi Jonathan,

Aliasing, of which moiré is a manifestation, occurs when trying to record detail that is smaller than the Nyquist frequency of a regular discrete sampling system. This means that, for a camera with a 4.88 micron sensel pitch, all detail smaller than 102.5 cycles/mm will produce aliasing.

A lens projects an image of the outside world with a reduced magnification, which is a function of focal length and shooting distance. Therefore one can calculate for which scenario there will be trouble in paradise. Mind you, it's not only repetitive patterns that aliase when they are small enough, also sharp edges and lines and small spots will aliase, not as moiré patterns but rather as stairstepping artifacts (AKA jaggies) and blocking artifacts (e.g. a star in the sky becomes a square).

Paradoxically, aliased images are poor candidates for upsampling, because the artifacts become easier to see. One can mitigate that a bit by preblurring the image.

Using diffraction caused by narrower apertures is not a very efficient anti-aliasing method, but it does help a bit. Defocus is more effective for suppressing high frequency detail. An AA-filter helps to reduce False color artifacts and a lot of the above issues, but they are not strong enough to totally avoid them.

Cheers,
Bart

[Dustbak]

Sorry; I wasn't trying to nitpick. I'm trying to map your experiences onto what I might expect if I spring for a D800E and I thought that the resolution of the sensor was what mattered (when attempting to capture the same image in different formats). Thus, I wasn't sure whether your experiences with your 39MP back were more reflective of what could be expected with a D800E than your experiences with the 60MP.

Edit: I'm still not sure if I'm any closer to my answer; I know that pixel pitch determines moiré, but to get the same FOV and DOF and resolution, doesn't the pixel pitch scale proportionally with the format and, hence, two sensors with the same number of MPs should be equally susceptible to moiré? If I'm being dense — or simply not making any sense, I blame having caught an early flight this morning and still being awake.

I think the D800e will be more like my experience with my 60MP back and probably even more like the tests I did with the 80MP back. Considering the pixel pitch of the D800 is around that of these 2 backs. I expect it to be a minor concern which is also why I put my money where my mouth is If I have one I can tell whether I was right....

Defocussing for moire is probably the lamest thing I have heard, unfortunately too many people give that advice. Why for heavens sake get rid of the AA filter to get sharper images and start making out of focus images. In most cases just a small part of the image is affected so why blurring the whole thing?? Narrower apertures do help and are  IMO better than a blurry image. What is even better is to use moire reduction in software and get efficient in getting rid of it in post processing....

[BJL]

Jonathan,
This is yet another of these aperture ratio vs aperture size issues in format comparisons, with the closest equivalence occurring at equal aperture size, so wi aperture ratio scaling with format size and focal length.

I think you are right that (assuming lenses good enough to produce aliasing and a sensor with no AA filter), when you choose aperture to get the same DOF in the different formats, you will be equally prone to aliasing effects like moirė when the pixel count is equal. Because in that case, you have equal aperture diameter and so equal angular spread by diffraction pairing with equal angular resolution by the sensor. Or on other words the aperture ratio, diffraction spot size (Airy disk) and pixel size have all been scaled in proportion to format size and focal length.

On the other hand, if you compare at equal aperture ratio, diffraction spot size is equal and aliasing will be similar at equal pixel size. That would put the D800E on par with a roughly 90MP back in full 645 format, so close to the IQ180 experience.

[Bart_van_der_Wolf]

I think the D800e will be more like my experience with my 60MP back and probably even more like the tests I did with the 80MP back.

Hi,

I hope you realise that for the same FOV, the image details on the D800E sensor will be significantly smaller, hence more prone to inducing aliasing. The denser sampling of the D800E only compensates for some of that.

Defocussing for moire is probably the lamest thing I have heard, unfortunately too many people give that advice.

Two observations.
First, DOF is not infinite, and thus there will be a lot of defocus on either side of the focal plane. For example, using a 50mm lens at f/4, and focusing at 3 metres (approx. 10 feet), will give a circle of confusion with a maximum diameter of one sensel for subject matter from 2.932 m to 3.071 m, or 138 millimetres (5.4 inches). Everything in front or behind will hit more than one sensel and gradually starts reducing aliasing. Stopping down to f/5.6 will create a diffraction blur pattern for green light that's more than 50% larger than the sensel is, and thus starts reducing aliasing and per pixel resolution. So DOF defocus and diffraction will help to reduce aliasing, but not as efficiently as an AA-filter.
Second, a modest amount of defocus can be very effectively restored by deconvolution sharpening, although the in-focus areas must be masked to avoid oversharpening.

So, it's not like people have to completely defocus their image to reduce aliasing, a lot of the image, except for the focal plane, is automatically defocused whether you want it or not.

Cheers,
Bart

[BJL]

I hope you realise that for the same FOV, the image details on the D800E sensor will be significantly smaller, hence more prone to inducing aliasing. The denser sampling of the D800E only compensates for some of that.

It compensates for all of that if the pixel size is reduced in the same proportion as the image size, which is to say, equal pixel counts. So exisitng 36-40MP options with no low pass filters could be a good guide.

[Bart_van_der_Wolf]

It compensates for all of that if the pixel size is reduced in the same proportion as the image size, which is to say, equal pixel counts. So exisitng 36-40MP options with no low pass filters could be a good guide.

That's correct.

IQ180, CCD height is 40.4mm sensel pitch is 5.2 micron.
D800E, CMOS height is 24mm, sensel pitch 4.88 micron.
Thus, 40.4/24 = 1.68x, and 5.2/4.88 = 1.066x, therefore D800E image detail is reduced much more than the sensel pitch.

IQ160, CCD height is 40.4mm, sensel pitch is 6 micron.
D800E, CMOS height is 24mm, sensel pitch 4.88 micron.
Thus, 40.4/24 = 1.68x, and 6/4.88 = 1.23x, and D800E image detail is reduced more than the sensel pitch.

IQ140, CCD height is 32.9mm, sensel pitch is 6 micron.
D800E, CMOS height is 24mm, sensel pitch 4.88 micron.
Thus, 32.9/24 = 1.37x, and 6/4.88 = 1.23x, smaller difference but D800E image detail is still reduced more than the sensel pitch.

As I said, smaller detail will potentially produce more aliasing, but (DOF) defocus and diffraction may mitigate that a bit. We'll have to wait and see how that plays together with the lens MTF.

Cheers,
Bart

[BJL]

IQ140, CCD height is 32.9mm, sensel pitch is 6 micron.
D800E, CMOS height is 24mm, sensel pitch 4.88 micron.
Thus, 32.9/24 = 1.37x, and 6/4.88 = 1.23x, smaller difference but D800E image detail is still reduced more than the sensel pitch.

Or when ones uses the full width of the sensor, for print shapes 3:2 or wider, the ratio of sensor widths is about 43.9/36 = 1.22, so almost perfectly matching the sensel pitch scaling.

A simpler way of seeing this is with pixel counts: 7320x5484 for the IQ140, 7360x4912 for the D800, and that almost equal count in the long direction tells the same story as the scaling factors being about the same 1.22.


So for prediction on the D800E, check out the moiré on the IQ140 or Aptus-II 8 or Pentax 645D or Leica S2?

[BernardLanguillier]

So for prediction on the D800E, check out the moiré on the IQ140 or Aptus-II 8 or Pentax 645D or Leica S2?

Yes, as mentioned before pixel count is the only thing to care about.

Cheers,
Bernard

[Guillermo Luijk]

Aliasing, of which moiré is a manifestation, occurs when trying to record detail that is smaller than the Nyquist frequency of a regular discrete sampling system. This means that, for a camera with a 4.88 micron sensel pitch, all detail smaller than 102.5 cycles/mm will produce aliasing.

I'd rather point that in a digital camera 'Aliasing, of which moiré is a manifestation, occurs when trying to record detail that is around the Nyquist frequency'.

Aliasing occurs for detail smaller than the Nyquist frequency in an ideal delta sampling system (i.e. a system that takes ideal samples with 0 [spatial] duration). But a digital sensor is far from performing an ideal delta sampling. It performs instead an averaging sampling over a certain period (spatial period, each photosite collects the photons falling into a non-zero area).

So when detail is coarser than the Nyquist frequency, the sensor captures it fine. And when detail gets finer than the Nyquist frequency, the sensor itself performs a low pass filtering that blurs the detail preventing aliasing and moire. It's only when the spatial frequencies are close to Nyquist that problems arise.

I have many times thought about this: a digital sensor maximizing the effective photon collecting surface will not only be more efficient in terms of photon collection (i.e. noise), but also more robust against aliasing and moire. On the other side a digital sensor with more unused gaps between the sensels, will be not only less noise efficient, but also sharper and more prone to aliasing and moire.

Regards

[jonathanlung]

Thanks, all! It sounds like I might take a pass on the 'E' version.