Looking at the Leica related threads on the Alternative forum I find many people ecstatic about the sharpness of the DMR and even Sigma DSLR shots. The reason for this sharpness is that neither the Leicas or the Sigmas have AA filters. Obviously Canon and Nikon DSLR's could produce results at least as good if they left off the AA filters, since they have cameras with greater resolution and their lenses are comparable to the Leicas and certainly at least as good as the Sigmas.
Canon and Nikon use AA filters presumably to avoid aliasing, AA=anit-aliasing after all. But the Leica and Sigma folks pooh pooh the need for AA saying that aliasing shows up in only a few pictures and is easily dealt with with software.
So would Canon users accept cameras without AA filters if the payoff is sharper looking pictures with only a few shots ruined by aliasing?
Well, the contention was that AA filter would remedy moire effects, which show up from time to time only under certain circumstances. Supposedly, the flip side of the coin is a slight reduction in resolution. To what extent, I am not sure.
The level of AA fiters inside Canon cameras varies. The original 1Ds has a weaker AA filter than that in the newer 1Ds MkII and again, people claim that the older camera is capable to resolve more lines than the newer one. However, the increased pixel count and better performance in high ISO settings make up more than the effect of AA filter between the two, if any.
I am for a weaker AA filter rather than getting rid of AA filter entirely. But who am I ?
My AA filter infested 5D is so sharp I sold off half my lenses as they just didn't cut it anymore. If the images were any dad burn sharper I'd probably cut myself...
The 5D I think has a weaker AA filter too, and its noticably sharper than the 300D I had but its also twice the resolution . I think I'd rather have to sharpen a little more in PP vs having to worry about whether moire is going to ruin a key shot.
The AA filter isn't just for moire and aliasing. It's also because of the staggered R, G, and B pixels on a Bayer sensor (CMOS and CCD), which despite the interpolation algorhithms still produces a ton of color noise. The blur reduces the color noise. The only sensor out there that doesn't face this problem is the Foveon.
To add to what DrPablo said the reason that Sigma does not use one is because they use a Foveon sensor unlike Canon and Nikon. I'm not sure what Leica used in the past but their new M8 uses a CCD apparently without an AA filter opting to get rid of ill effects with software.
BTW if you want to read about a thousand posts on the subject there's a HUGE thread from about a month ago on this subject. https://www.fredmiranda.com/forum/topic2/415203/0
DrPablo wrote:
The AA filter isn't just for moire and aliasing. It's also because of the staggered R, G, and B pixels on a Bayer sensor (CMOS and CCD), which despite the interpolation algorhithms still produces a ton of color noise. The blur reduces the color noise. The only sensor out there that doesn't face this problem is the Foveon.
That's just so wrong dude - sign up for a signals processing class at your local college.
Canon should keep the AA filter.
I'm very happy with it as is. I saw some images from the new Leica M8 that underscore how useful the antialiasing is in some circumstances. With a good lens my images are what I'd consider razor sharp, and with very little moire* to worry about I'd say don't mess with a good thing.
* I actually did start seeing faint moire in certain scenes once I got my 17-40L... damn it's a sharp lens. Fortunately it's reasonably well controlled and manifests as very faint colour abberations rather than more extreme coloured patterns. I have never seen it with the 70-200, probably because offensive patterns like fabrics or whatever would tend to be slightly outside of the zone of focus at telephoto focal lengths...
As long as the color in the image changes slowly in the spatial dimension relative to the filter pattern, color interpolation works well. However, for edges of objects, or fine details, color may be interpolated incorrectly and artifacts can result. For example, a small white dot in a scene might illuminate only a single blue pixel. The white dot might come out blue if it is surrounded by black or some other color, depending on what comes out of the interpolation. This is called aliasing. One way to reduce aliasing is to use a blurring (or “anti-aliasing”) filter, which deliberately discards fine details. Defocussing the camera lens does almost the same thing....Show more →
Some of the challenges that face interpolation algorithms include image artifacts, like moirés and color aliasing (shown as unrelated green, red, and blue pixels or resulting in discoloration). Most cameras fight the aliasing problem by putting an antialiasing (AA) filter in front of the sensor (which actually blurs the image and distributes color information to the neighboring photosites). Of course, blurring and high-quality photography don’t usually go together, and finding the right balance between blurring and aliasing is a true challenge for camera design engineers.
Canon shouldn't lose the AA filter. It's there to reduce the incidence of imaging errors, one of the more visible ones being moire fringing. Without an AA filter in place, there may be other imaging errors present in the image that you don't even know are there. Once a signal is aliased, it contains data that wasn't in the original signal. An AA filter enables a more faithful capture of the scene without digital imaging errors.
The Kodak SLR/c doesn't have an AA filter and look at what happened to that camera. If you really want one, I am sure you can pick one up used very cheap.
When used under the right conditions they take one hell of a picture though!
Thanks Nathan, I'd appreciate that too. If he wants me to be more technical and say 'color moire' instead of 'color noise', then he has every opportunity to make that critical point.
What Canon aught to do is remove the hardware AA filter completely and whip up a software AA filter into the DIGIC IV processor and then throw an on/off switch into custom functions (or even a AA strength slider in the regular menu).
Software filtration unfortunately will never be as good for moire and aliasing. The PS fixes for moire are very very blunt tools, and they probably do more harm than the AA filter. I had an image with a lot of moire from a tight pattern, and I tried Dan Margulis' moire technique. It didn't really work except for dulling a lot of the color info.
Well, Photoshop's built-in noise reduction and sharpeing are no hell either but as time goes on we see new techniques and software developed for these things (Neat Image, the FM Sharpening Plugins, etc.) that do a much better job. I don't think it unreasonable to imagine Canon implementing an extremely effective software AA filter and integrating it in the onboard processor. Or maybe a combination of a much weaker hardware AA filter and a secondary (optional) software filter on the chip.
DrPablo wrote: "The AA filter isn't just for moire and aliasing. It's also because of the staggered R, G, and B pixels on a Bayer sensor (CMOS and CCD), which despite the interpolation algorhithms still produces a ton of color noise. The blur reduces the color noise. The only sensor out there that doesn't face this problem is the Foveon."
Yes, but the Leica DMR and M8 use Bayer sensors and have no AA filter. The pixel spacing of the Leica sensors are not unusually dense but the lenses are supposedly very sharp. In other words the lenses should be providing detail to the sensors that is well above the Nyquist frequency and resulting in a lot of aliasing. Yet the Leica owners think that aliasing is a non-issue.
I've seen only a little aliasing with my 5D but when it did occur it ruined what was otherwise a very nice picture. I took some shots through a window which had a mesh screen over half the window. The mest screen was not resolved by the sensor but in its place was an ugly maze-like pattern with a background of false color. The color could be removed but not the maze-like pattern. Even worse, the rest of the picture looked so good on screen that I didn't even notice the aliasing until I printed a 13X19. I can't think of any way that software could detect and fix such aliasing automatically. Once aliasing artifacts are recorded on the sensor there is no way to fix the situation in software except to alter the image in a destructive way.