Will anti-aliasing filters doom even forthcoming Canons to early obsolescence?

[JerryBruck]

In recent years we've seen the M-9, offered by a revitalized Leica at about triple the price usual for such a full frame body,  along with the designer casing and restricted functionality evocative of Apple Computer but with more than a twist of retro love -- at last you can buy new triple-priced lenses without such sissy features as speed or auto-focus.  The customers can't love it enough, and fall to almost mystical terms when they try to describe the IQ.   (There have been whispers that Leica's unimagined success at the high end has driven many of Canon's top execs into counseling.)   Pentax entered digital Medium to raves with the 645D; we have gasped at the entrance of Sigma's SD1 with its Foveon X3 (APS-C) at a stonking $7000, body only.   En route is the FujiX-Pro1, a poor person's M-9, in a way, that has the heavies at Luminous-Landscape dancing the frug the twist and the wa-watusi in anticipation.  Then there is the expected Nikon 800E...

What these five bodies have in common is the absence of anti-aliasing filters.  Stripped of this dulling scrim (some say), an APS-C sensor out-resolves the best full frame, and other aspects of lens IQ leap ahead as well.  Moire (they say) is hardly an issue at today's high resolutions anyway and can be eliminated entirely by in-camera digital signal processing (Leica), Fuji's quasi-random 6x6 color filter array or that triple stacking of photo-sites (Sigma).

So tell us, who that knows, just how easy is abandonment of the AA filter, what does Science and Engineering say?  Will it be a new dawn?  What if anything will Canon offer here?

Finally, is this a case -- for now at least -- in which the interests of still and movie shooters clash? (The lower resolution and higher data rates would make moire more of a problem and harder to manage with the latter, I imagine.)  Light: please, heat: not-so-much.  I have a foot in both camps.  Let us love one another.

[pharp]

I've wondered about that myself. www.maxmax.com offers the following service for existing bodies;
HR: High Resolution/Hot Rod.  AA (Blur)/IR Cut Filter (ICF) filter removal and replacement with a new custom ICF.

Has anyone tried this? Big difference?

[sb]

I have to admit, the way they've solved the moire problem on the Pro1 is quite clever (randomization of pixel patterns). Of course, we need adequate sample images to really see if it's working the way they claim, but if it does, I think Canon and other manufacturers could easily come up with their own modifications to Fuji's solution without violating patents etc.

I think this needs to happen. 

[Viggo]

FWIW the Leica S2 is a pretty expensive camera and it suffers very badly from moiré from what I've seen from shots a friend of mine took.

[Policar]

The moire on the Leica is horrible.  I would love that camera but the images look so bad up close.  Fuji's approach sounds novel, but everyting I've seen shot without an AA filter has been frighteningly bad thus far...unless shallow DoF or diffraction kicks in first as is the case with some medium format backs.  Once you're approach diffraction limited on virtually everything losing the AA filter seems to make sense, until then...not so much.  Kodak's CCDs only lack a filter because Kodak and their clients don't have the equipment to make AA filters affordably and reliably in large sizes.

[AprilForever]

Has anyone ever had the Hotrod conversion done?

[thejoyofsobe]

it's the new Fuji array that's got me interested. it's too late for this generation of sensors to adapt what with the 1DX, 5DX and the t4i (likely 7d2) sensors already coming to market and Fuji's sensor not yet in the hands of consumers to see if it's all that it's cracked up to be. unless the patent is airtight or there are problems with Fuji's sensor, i don't see how Canon, Sony and Nikon don't respond in kind in their next iterations.

[Mt Spokane Photography]

Improvements in cameras and sensors is incremental, and nothing is going to suddenly obsolete a camera.  Many older cameras still compare quite favorably with the newer ones.

AA filters are there to reduce Moire.  Its a tradeoff as to hom much AA filter to use, and how much of a high frequency filter to use in the processor.  No one has yet repealed the laws of physics, they just trade off one thing for another.

Presumably, improvements will continue to be made, but todays cameras will still work as well as ever.

[aeturnum]

It's worth pointing out that the Foveon sensors don't have an AA filter because they get all three color values in each pixel and don't have to average from surrounding pixels. They never had an AA filter, as far as I know. The process of getting all three RGB values for each pixel when using a Bayer filter is, as I understand it, what generates the Moiré.

If people knew about a reliable way to eliminate Moiré using software, I am sure it would be implemented at this point. I am interested in seeing how the X-Pro performs, as it's not using a Bayer filter. I'm pretty skeptical of a complete solution to the problem suddenly presenting itself, however, given how long people have been looking for a solution.

[kapanak]

It's worth pointing out that the Foveon sensors don't have an AA filter because they get all three color values in each pixel and don't have to average from surrounding pixels. They never had an AA filter, as far as I know. The process of getting all three RGB values for each pixel when using a Bayer filter is, as I understand it, what generates the Moiré.

Precisely. Which is why a 15mp APS-C Foveon is equivalent to a 45mp medium format sensor. Quite an amazing tech, really. Each wavelength of light penetrates the sensor at different depths, and THAT is how each pixel can see all colours. Really, the only true way to get rid of moire.

[jrista]

Ultimately, moire is a matter of physics when it comes to spatial frequencies. Assuming you have an "ideal" sensor that doesn't exhibit moire, and take a photo of one of those starburst test charts. Moire is still going to be a problem. The computer screen you view the photo on will exhibit moire as you near the frequency limit of the pixels on the screen. Print will exhibit moire as you near the resolution of the print. Moire is a matter of physics...and whenever your spatial frequency approaches the frequency of the viewing medium, it will present. The benefit of Foveon is that it will largely eliminate color moire, however it still can't eliminate luminance moire. It does reduce the need for a low pass filter since you no longer have the funky bayer pixel arrangement to deal with. If recent patent filings by Canon are indicative, we may see a layered sensor from Canon at some point in the future.

The benefit of a low pas filter for bayer type sensors...so long as it is properly designed and tuned for the specific sampling frequency of the sensor...is that it can mitigate recorded moire (sampled frequencies that might produce moire), so that it doesn't compound the problem of moire in other mediums. Sometimes Canon gets a bit overzealous with their low pass filtration (blurring frequencies that can still produce decent contrast and useful detail on the sensor), and they should definitely get better at producing AA filters that are designed for the sensor they are used on, thus producing better results. The problem I see with eliminating the low pass filter is that you then have both moire from a viewing medium...say computer screen, that interferes and compounds with moire hard-coded into the image. I just can't see any reasonable outcome to that...other than lost detail (or worse, useful detail replaced with useless detail) in common viewing mediums.

Lightroom 4 has introduced a moire healing brush, accessible under the custom brush tool. I've poked around with it, and while its nice that you can target moire removal just where it is needed, it doesn't seem to produce a nice or as consistent a result as a proper low pass filter does (one that isn't overzealous.) I'm sure the technology will improve, and for those to whom it matters, it'll probably be the saving grace for filterless cameras. I think it will just be one more thing we have to correct in post, and personally I would like to see low pass filters stick around...however I think they could be improved for sure.

[necator]

Only owining a 5D c and a 1Ds mk II I can only speak for relatively low resolutions, yet moiree is a issue with the relatively weak AA-filter on the 5D c. I regularly photograph fashion, and sometimes a cloth produces moiree ... and it's a beast to get it removed in software.

And I can't imagine how moiree patterns could be identified automatically and automatically corrected.

So, removing the AA filter would result in crispier images overall, but in rare cases it would leave you with a unusable photo. If you never shoot moiree-critical things, no-AA would be the best to you since sliced bread. But if you do, it'll drive you crazy, IMHO.

[JerryBruck]

@necator -- from your experience, and if you have the time, a list of types of "moiree-critical things" would be very interesting.  We hear and see often this effect from repetitive fine patterning on fabric.  What else?

[jrista]

@necator -- from your experience, and if you have the time, a list of types of "moiree-critical things" would be very interesting.  We hear and see often this effect from repetitive fine patterning on fabric.  What else?

Anything with repeated, predictable patterns at fine scale will produce moire. A chain link fence at the right distance will produce moire...a brick wall at the right distance will produce moire. I think it occurs more with fabrics because they are a fine-scale repeating pattern that is frequently photographed as part of fashion photography and the like.

Its ultimately the convergence of the grid pattern of the sensor with fine-scale repeated patterns of the virtual image. The patterns can be a grid like fabric, brick wall, or chain link fence, radial patterns, etc. I don't think there is a discrete list of things that can produce moire...there are quite a few things that could, in the right conditions.

[LetTheRightLensIn]

no