Live Dangerously

Ever wondered why all your digital pictures seem like they're shot through a soft veil, a slight haze? That's not inherent to the Bayer pattern—it's the helpful AA (anti-aliasing) filter that the manufacturers use to prevent moiré. Trouble is, the AA filter also prevents sharpness. For a mere $450, MaxMax.com will hot rod your DSLR for sharpness by removing its AA filter for you. (I suppose they'll let you throw out your own warranty.) In the pictures above, the one on the left is the hot-rod mod, the one on the right is the stock D200. Click in the image area to see it larger. Sure, you'll get moiré occasionally—live with it.

I'm sure someone will quickly correct me if I'm wrong, but I think this is also Leica's trick on the M8—they leave out the AA filter to improve sharpness. (Wonder if a hot-rodded D40x is as sharp as an M8?)

The link leads to the D200 page, but they'll do it on a Canon 5D and other Nikons, too. Turnaround time is one to two days. Live dangerously.

____________________

Mike (Thanks to Oren)

Featured Comment by Carl Weese: "I'd seen links to this before: 'image quality' in this usage means resolution at 850% view. In fact, they want to sacrifice balanced image quality at the altar of resolution.

"Moiré can be a huge problem with all sorts of subjects—backlit leaves come to mind—in landscape work. Restoring resolution lost to the AA filter is what capture sharpening is all about. One can certainly argue with the degree of AA filtration used on various cameras—it's a compromise so no one solution is perfect. Just as you can argue over the amount of in-camera high-iso noise reduction with some implementations of reduction looking worse than the noise itself. But the premise than AA filters are simply a bad idea and you'll get a global improvement in "image quality" simply by throwing away the AA filter, that's just plain silly."

Featured Comment by Andreas Manessinger: "Hmm...I suppose there is more danger to it than one thinks. I use a D200 and for many images, especially when I am satisfied with the colors of JPEG from the camera, I use Capture NX as RAW converter, even when I do all other work in Photoshop. Nikon really understand their camera and the conversion results are top-notch, usually much easier to achieve than in Adobe Camera RAW, but I guess Nikon's algorithms rely on the sensor being unmodified. Actually the same would be true for other software. Just think of DxO and their de-mosaicing and automatic image correction algorithms, and more so the new noise reduction on RAW data that will come in version 5. All that could cease to work when you do that modification. In other words: I would not do it."

Comments

[Chris Crawford]

I have a Kodak 14n and it also was designed with no AA filter. Sharpness is incredible, but Moire is pretty bad in some situations. I use mine for small products and landscapes so i rarely have troubles. Clothing can be a b-tch with it though!

Didn't some of the first digital SLRs, the old Kodak ones made with a film camera body with a big growth on the bottom, have user-removable AA filters so you could take it out and put it in as needed?

[erlik]

Hmmm. I wonder how an E-510/410 would fare compared with a camera that has had its AA filter removed. That is, E-510/410 can have noise reduction turned off completely. The detail is very nice.

For instance, on my http://www.clandestineart.com/photo/410.html, you can see the power and anchoring lines in Big Sky even at 800x600 and even if they were quite distant.

[Richard Tugwell]

Certainly with the M8 i don't feel the same urge to head for the sharpening toolkit that I do with the EOS5D. Do you know if there is any impact on sensor cleaning / dust accretion if you have this mod?

[Dierk Haasis]

Any reason for major surgery with known complications [moiré] when the AA-effect can easily be healed by Unsharp Mask [originally developed in the darkroom exactly to counter such softening effects from lenses] or the various incarnations of smart sharpening by different manufacturers?

Sounds to me a bit like cutting of ones head to get rid of headaches.

[Christer Almqvist]

the URL above does not work becuse the "," is included. try this:

http://www.clandestineart.com/photo/410.html

[phojus]

Flipping heck Mike! You're scaring the hell out of me.

[APScribble]

Ahhhh, that's one reason I love my 2 megapixel Canon D2000, yes, one of the old Kodak models built off of a Canon film body.

Canon D2000: The Awesome Antique Digital Camera

[Louis Doench]

Interesting, perhaps i will have this done with my D1x, who has passed his warranty time anyways, and has had the hotshoe snapped off.

Could this be combined with converting to IR through lifepixel?

[Joe Q]

Sure would be nice to slide an AA filter in and out.

I'll take a little less sharpness over purple people (M8 style) any day. The lack of an AA & IR filter on the M8 was more of a forced compromise than a goal -- they couldn't fit it. For street & people photography, which the M8 is best for, an AA & IR filter is desired. For landscapes, you can do away with these things, but landscapes are not what the M8 is all about.

[Jon Bloom]

Intentionally or otherwise, this example is deceptive. There is much greater depth-of-field in the left image than in the right. (Compare the relative sharpness of the steering-wheel hub in the lower right corner of the two images, then compare the relative sharpness of the key or "VENT BLOWER" label in the two images.)

With a moderate amount of unsharp mask, the image on the right can achieve artifact-free sharpness equivalent, to my eye, to the image on the left -- sans moiré. At least, the in-focus part can be as sharp.

The anti-aliasing filter is there for a reason, you know.

[Alan Rew]

What about the increased hazards when you need to remove dust from the sensor? An AA filter protects the actual sensor, so when you clean the 'sensor' you're actually cleaning the filter. The thought of cleaning the actual sensor scares the hell out of me!

[Mike]

Alan,
As I understand it, the AA filter is bonded to the IR cut filter, and when they replace it, they use a piece of plain optical glass and a new IR cut filter. So you're just cleaning a piece of glass, not the sensor.

Mike J.

[Ctein]

Dear Jon,

First, the function of an article illustration is to illustrate, not to prove. It cannot be deceptive, only the text can be. This is a common mistake readers make-- arguing that an article is wrong because the accompanying illustration doesn't prove the thesis to their satisfaction. Well, that ain't its job.

Second, I can't see any difference that matters in the DoF. Uniformly, at all distances, the unfiltered photo is sharper-- the key, the dash, everything. That will yeild greater DoF if you're holding the blur circle constant; it's not a consequence of it.

(Not that I think removing the filter's smart idea. )

pax / Ctein

[eli]

Oh, it's a physical layer? That sorta sucks. I hope these guys bond a similarly-hard material in its place. Still, I'd like to see cameras do away with this. That's why they put advanced little computers into the cameras. Why do it in hardware when it causes such blurring?

[eli]

Also, I just looked at the raw CR2 files from the 5D they posted on their site. It's not the one AC unit with moire that concerns me, it's the AC unit behind it perspectively:

http://lowmag.net/assets/2007/badAChotrod.jpg

What is that? Is this the real reason behind AA filter?

[Ctein]

Dear Eli,

This is not a problem you can solve in software without extremely high levels of computation that approach human visual feature recognition. Theoretically doable; practically doable when you have a teraflop CPU in your camera. We'll get there, but not tomorrow.

See this column for an explanation of the problems aliasing causes.

http://theonlinephotographer.typepad.com/the_online_photographer/2007/09/sampling-isnt-s.html

How would you "de-alias" my scan illustration, for example, in software? Same problem.

pax / Ctein

[Paul De Zan]

All in all, this strikes me as a strong candidate for Worst Idea Ever.

Think I'll put that $450 in my D3 fund.

[Martin Doonan]

Joe Q wrote
"For street & people photography, which the M8 is best for, an AA & IR filter is desired. For landscapes, you can do away with these things, but landscapes are not what the M8 is all about."

Actually, I've seen the results from an M8 with back lit leaves, dappled light. Results were plain awful. The 5D images they show for example have all sorts of weird colour effects around telegraph poles.
This mod would mess up a whole lot of my landscape work.
Whilst I can correct easily with software, I'll live with the AA filter.

[A. Dias]

The reason for the AA filter is fundamental sampling theory (Nyquist limit). A low-pass filter is needed to band-limit the data to one-half the sensor's sampling frequency. I rather live with that than ugly aliasing artifacts. Artificial accutance at the cost of those artifacts is too high a price to pay. That's what the Leica M8 produces and people love it, but I do not.

[Dierk Haasis]

Dear Eli,
when the information Mike gives us is correct, that they exchange the AA-filter with a piece of optical glass, chances are it is much harder than the original material. AFAIK, the AA/IR-filter is realised through a not so hard synthetic material [vulgo 'plastic'].

[Howard Slavitt]

So answer this: why do the highest end digital backs now commonly available (well, the most expensive anyways) not have an anti-alias filter (i.e., Phase One, Leaf, Imacon-Hasselblad medium format digital backs)? Have they designed their chips somehow differently to take into account the fact that there's no AA filter? Are people who use any of these backs locked into proprietary software -- and, if so, maybe it's being addressed at least partially in software? I'm quite interested in this topic because I've been seriously evaluating whether to "live dangerously." I've been working with and evaluating several Raw files the last few days taken (and generously lended to me) by someone who had his 5D converted by maxmax.com. . . . I do wonder about the comment of whether using such files would throw off, for example, DXO Optics' lens/camera measurements, because I am using DXO and so is the guy who gave me the files and had his 5D converted. . . . My conclusion is that I can't come to a conclusion based on looking at the raw files alone. I'd need to have been there taking the photos, and, much better yet would be to have side-by-side comparisons of with and without an AA filter.

[Antonio Dias]

>So answer this: why do the highest end digital backs now commonly available (well, the most expensive anyways) not have an anti-alias filter (i.e., Phase One, Leaf, Imacon-Hasselblad medium format digital backs)?

With very large sensors, the sensor may resolve more than the optics in front of it. In that case, the optics may very well intrinsically perform the band-limiting required by the sampling process and no additional low-pass filter (AA) is required.

[Paul G.]

What I think would be helpful here, as with most things, is a comment by someone intimately familiar with the physics of sampling (i.e. 'someone who knows what they are talking about', along the lines of the comment by A. Dias). The internet can be a very dangerous place to gather "information" (read: opinion) on which to base your own personal photographic decisions. I would would like to see the results of a super-charged D200 image compared with a stock D200 image with appropriate sharpening and contrast adjustment applied. As RAW images captured with the D200 are meant to be adjusted by the photographer in Photoshop or some other similar program, I think that the comparison would be more prepresentative of a "real-world" situation. Showing unsharpened images is interesting, but ultimately biased and not terribly helpful, although I give them credit for showing examples with some serious moire problems.

As I capture mostly landscape images, this issue is of particular interest to me. I also like taking pictures of my cats looking at birds through the screendoor. I wish I had the funds for a super-charged D200, a stock D200, a IR D70s, and a Nikon FM3a with PanF 50.

Paul G.

[Howard Slavitt]

A. Dias wrote: "With very large sensors, the sensor may resolve more than the optics in front of it. In that case, the optics may very well intrinsically perform the band-limiting required by the sampling process and no additional low-pass filter (AA) is required."

I doubt this explains why medium format backs do not have AA filters. The current 39 megapixel backs have pixels of about the same size as the Canon 1DsIII (about to be released) and the Canon 20D and 30D, all of which have, had, or will have AA filters. I'm pretty sure that until about a year or two ago, the MF backs with the largest pixel counts were 25 megapixels. They had a much larger pixel size than DSLRs at the time, but did not have AA filters. . . .

It's still possible that the current Hasselblad lenses (as an example), because they have greater coverage than 35mm lenses, cannot outresolve the MFDB sensors. But it's highly unlikely. They almost certainly outresolve the 25 megapixel backs, thus creating a high potential for moire; otherwise, what would have been the point of the 39 MP backs? . . . Also, the word on the street is that the current Haselblad lenses resolve as many lpm as state of the art 35mm lenses -- I'm sure they're at least close. . . . No, I highly doubt the answer for why MFDB do not have an AA filter is because the lenses do not ouresolve the backs. . . .

I think the answer is that DSLRs are supposed to be all purpose cameras, and the best all purpose compromise for the vast majority of users is to have an AA filter. It does not require as much post processing, and most of the AA effects can be countered by sharpening. However, for people who are willing to spend the time to individually process each RAW file, and to selectively sharpen, and where necessary, fix chromatic aberration, then going without an AA filter on a Nikon or Canon DSLR may well be the way to go. I think that Mike Johnston captured the decision well in his article title -- are we willing to "live dangerously"?