“To Omit an Alias Filter… Is like Building a Sports Car with No Brakes”


Last week, we pointed you to a piece by the New York Times on how Fujifilm is attempting to kill moiré without killing sharpness by designing its sensors in a way that eschews the traditional anti-aliasing filters used in digital cameras. Photographer Martin Doppelbauer disagrees with Fuji’s claims: he has published a piece arguing that, “digital cameras without aliasing filters are cameras with a built-in design flaw“:

To omit an alias filter in front of a digital image sensors is like building a sports car with no brakes. Of course, the car accelerates a little faster due to the lower weight and the cornering ability is also better due to the smaller unsprung weight – but ultimately it lacks an essential functional element.

For analog cameras, an alias filter is not required: ​​Film has no sharply defined limit of resolution. It loses contrast and resolution gradually with increasingly higher frequencies. You could say, the low-pass filter is already incorporated in the film itself.

[…] By omitting the alias filter, the recorded image information […] does not increase! Even though images of cameras without aliasing filters may appear sharper and crisper: Images of cameras with a proper alias filter can easily be re-sharpened to achieve the same visual impression – without side effects.

So according to Doppelbauer, the recent fascination with removing anti-aliasing filters is more based in marketing rather than science.

Alias-filters: Yes or No? [Martin Doppelbauer]

  • Nathan Blaney

    Why is this person’s opinion noteworthy? Is he an expert of some sort? I’ve never heard of him. All I know is my X100 and X-Pro 1 are the best cameras I’ve ever owned.

  • Samcornwell

    The thing with the digital sensor, is that to a point the aesthetic produced is all the same. The major differences are CMOS, CCD and size. Most digital SLR’s today produce exactly the same digital mosaicing, unlike different films producing very different results upon closer examination (See Paul Graham’s 2011 book FILMS).

    The point being that alternative products (omitting the alias filter) which change the structure of the digital image at the very core should be appreciated whether they’re giving or taking away value because they add diversity (to digital photography.)

  • G

    If you’re gonna quote from his reasoning do at least include something that actually explains what he’s on about, instead of just a phoney analogy about sports cars that doesn’t really relate too well with the topic.

  • Tommy Sar

    The thing about using car analogies to explain technical ideas is that they are almost never make sense.

  • AdamV

    Why can’t they just make a camera with a retractable anti-aliasing filter?

  • Thewirehead

    The X100 is an awesome camera with a ton of really quirky things that make me want to throw it sometimes. I’m very torn about how I feel about it. And from what I can find, there is some kind of filter on the X100 sensor, though it sounds like it’s very light and some of the moire is removed by the camera software during processing.

  • Mike Selsky

    Yea the Leica M9 totally blows as well.

  • Matthew Wagg

    Another pixel peeper with no real science to back up his claims.

  • David Liang

    Even if he was correct his analogy about the sports car is not. A camera without an Aliasing filter is like a car that has no speed limiting factor. Break’s have nothing to do with it.

  • Michael Comeau

    I just want to say that all of my cameras have AA filters and that doesn’t mean I’m a bad person.

  • brandon

    i believe it’s spelled “brakes”, and i think his analogy is just fine. a camera without an AA filter still has it’s speed limited. there are only so many pixels, AA filter or not.

  • MarvinB7

    The audacity! How dare everything! I shoot with a 30D, so stop whining about trivia. :D

  • Duke Shin

    I’ll stick with my Konica Pop until all this blows over.

  • Mike Traynor

    I have owned a Nikon D800e for about 6 months now and I have to say the results have been spectacular. I mostly shoot landscapes and portraits, and maybe 1 out of 500 has a bit of aliasing if I look for it. The work involved in fixing those areas is marginal compared to not having to sharpen each image. In most cases I would just not use that image.

  • wickerprints

    He’s not saying those cameras are bad. He’s merely criticizing (and rightly so) the popular misconception that the optical aliasing filter is somehow a detriment to the image. After all, various Leica bodies have also been designed without such a filter, as has the Nikon D800E. Fuji is not being singled out.

    The point he is making is perfectly valid. Moire and aliasing artifacts are the inevitable product of sampling at a fixed spatial frequency. They are impossible to avoid so long as a regular grid of pixels is used to sample the incoming light. It isn’t, as various people might want you to believe, a property of the Bayer color filter array.

    My suggestion is that, rather than questioning his expertise or knowledge of this phenomenon, instead perhaps you should consider his claims as a starting point to investigate the underlying mathematical and engineering principles, so that you can come to your own understanding. I think it’s unfortunate that so many photographers react with hostility when confronted with what they perceive to be negative attacks on their favored gear. Such a response closes one off to the scientific truth.

    And for those who might say, “well, but I’m a photographer, not a scientist–all I care about is taking photos,” then by all means. Take the photos you want to take. Nobody is saying you can’t. But if you have an interest in understanding the workings of the tools you use, and whether a manufacturer is perpetuating misconceptions about their technologies for the purpose of marketing advantage, then yes, you should educate yourself in the technical details. And educating yourself so that you aren’t fooled into believing untrue claims is not mutually exclusive to the enjoyment of taking photos.

  • Tomasz Kulbowski

    Nonsense. I’ve been using x100 for a past year a lot, in various situation, and never noticed any issues with moire. Instead, I find the lack of AA filter very useful and the amount of details is better than my 5D II. What’s the point in worrying about a possible issue which might rarely affect some photos in some very specific circumstances? In some situations my 5d can’t focus properly – does it make it useless?

  • td

    There is no way that a filter does not reduce transmission to the sensor. Anti-aliasing filters consist mostly of lithium niobate layers. If they are removed, transmission is increased. Fuji, Nikon and Pentax cannot be so ignorant.

  • Porter

    Wow. Quite the nerdgasm, going on here. In the overall scheme of things, none of this matters. Is a captivating photograph any less impressive once you know what kind of sensor it was made with? No. It doesn’t matter. Relax. Leave behind the pasty glow of your computers, go outside, and put to use whatever camera you have. Have your mom pack a hot lunch, it’s cold outside.

  • td

    There is no way that a filter does not reduce transmission to the sensor. Anti-aliasing filter in digital cameras mostly consist of lithium niobate layers. If they are removed, transmission is increased. Fuji, Nikon and Pentax cannot be so ignorant.

  • Ivan

    Just to point out that we are looking at the results from two sensors with different resolutions, two different lenses, and different image processing.

    Moire is probably easier to find with X-Trans, but to have a real case here same resolution sensors should be used, same lens at identical aperture, with identically framed test subject, and processed identically. The last requirement is hard to satisfy at this point due to (still) inadequate support for X-Trans raw files, so I would rely on in-camera JPEG and what manufacturers offer to process images from their own cameras and sensors.

    Only then we can discuss the results.

  • jgentsch

    Neither his analogy nor his argument nor his comparison of the two different cameras hold up.

    Yes, Moire is a problem of discrete sampling and has to be corrected for. You can try to do that upfront by using an optical alias filter or after the fact as part of the image processing algorithms. You can also combine both which at least for me and my 5dII is what happens in practice.

    I personally would argue that using an AA-filter you make an upfront assumptions about what parameters (amout of blur) are apropiate for most situations, whereas you could fine-tune the algorithm per scene or even per part of the image. So in the long run as sensors and algorithms improve I believe dropping the AA-filter makes sense.

    Comparing different sensors does not help to support his general theoretical claim either. Even if he had taken the same sensor with and without an AA-filter and compared the results he could only have concluded that the AA-filter solution was or was not more effective than the algorithm used for that particular scene (artficial stress test).

    The analogy doesn’t hold up, because it is not like there is a part missing, the functionality is just provided differently.

  • G

    The X100 actually has an anti-aliasing filter.

  • Chetan Crasta

    A better comparison would be to a car without a rev limiter.

  • Tomasz Kulbowski

    Just checked the specs – you’re right! But still – it’s a very weak filter and you can clearly see it on photos. Many reviews comparing x100 and cameras without AA filer said it was removed completely, that’s what made think so.

  • G

    True, it is weaker than in most cameras.

  • E

    Nikon does this, not for all cameras.. just like Fuji doesn’t do it for all cameras as well.

  • tttulio

    Listen to this guy, if you are a Photographer who enjoys shooting test charts.

  • Cloudmaster1

    well looking at my results from a D800E and a D800 i can say he is spot on.

  • Cloudmaster1

    well i have both the D800E and the D800… there is no visible difference when i do prints on my epson 3800.
    sharpening eliminates the difference to a marginal factor.

  • Cloudmaster1

    wagg… another internet user with a P&S and a opinion.

  • Cloudmaster1

    boy you don´t even know your camera……

  • Laurent Laborde

    No, if you want a less crappy analogy it’s “Manual gear vs automatic” or “brake without or with ABS”. One of them can eventually lead to better performance if you know what you’re doing. This article is just an ugly FUD.

    I want a camera without bayer filter (B&W), without hot mirror (IR/UV), without antialiasing filter (which is just a low pass filter, i don’t ask for filters).

  • wickerprints

    Ignore the analogy, and consider the science for what it is.

  • Laurent Laborde

    sensor may not be the most important, but it doesn’t mean it should be ignored.

  • Jackson Cheese

    Cloudmaster1…another pixel peeper more interested in cameras than photography.

  • Matt

    Maybe he was refering to post impact of the car? :)

  • Matt

    Actually, there is some diversity. Fevron, which I’m trying out now. And Fuji has done several approaches before. They mixed different sized photosites and have played around with the ordering of photosites. Always with good results. I think both techs need supporting. I’m done with buying canon/nikon for a long while, they have enough support for their R&D.

  • Matt

    Sorry, but the Bayer array is a large factor. And, IMO the AA filter is not the best solution to the problem.

  • branden rio

    You mean Foveon?

  • wickerprints

    You have misunderstood my point about the Bayer CFA, so I will elaborate. The Bayer arrangement DOES contribute to more visible moire and aliasing, because the (linear) sampling frequency in the red and blue channels is half that of a sensor with no color filter at all (panchromatic). So you will observe moire and aliasing at lower spatial frequencies in these channels. Some of this is partially offset by the demosaicing interpolation performed in the RAW converter–this is true for any sensor that uses any type of color filter array, but because the Bayer arrangement follows the same lattice structure as the sensor’s pixels, it is prone to color aliasing.

    However, that is NOT the same thing as saying that moire and aliasing are PROPERTIES of the Bayer CFA. Even for a camera without ANY color filter of any type, such as the Leica Monochrom, the simple fact that the pixels are arranged in a square lattice means that the sampling is subject to moire and aliasing artifacts, even if there is no color information present.

    The idea of aliasing in the frequency domain is more intuitively understood if we look at a simple example in the time domain, which is the zoetrope. A zoetrope simulates smooth movement/animation by permitting each image to be seen only at certain intervals. In the time between images, the line of sight is blocked. This is a form of sampling in the time domain–intentional aliasing is introduced by blocking a continuous signal, to fool one’s vision into thinking that the input is coming from a smoothly animated source. The same principle is employed with motion picture film projection. The projector’s light is turned off in between frames, so that you don’t see the movement of the film. If the light were continuously on, you would see a blur. If the frequency of light pulses were half as fast, despite the film moving at the same rate, you would see only half the frames. If the light pulse frequency were twice as fast, you would see a double image, each offset by half the picture height (if the film is moving vertically).

  • Porter

    Perhaps, but according to the back-and-forth going on in the comments there are ways to correct for both in post, or at least ways to get around the issues. So, this all just seems like an argument for argument’s sake. That’s all.

  • wickerprints

    I disagree, because at the optical AA filter prevents spatial frequencies above the pass band from polluting the signal. If the AA filter is not present, then there is no longer any way to tell whether the signal is real or an artifact, or to what extent it is a combination of both. That information is lost.

    The ideal AA filter simply filters out frequencies above the Nyquist limit. You don’t need to do anything else. But as we all know, no cut-off filter is ideal, so you do lose some of the real signal at the highest spatial frequencies. However, this is still far superior to not having the filter there–in fact, it can be argued that this is precisely how film behaves, in terms of its resolving power as a function of spatial frequency. The MTF doesn’t just drop off a cliff–it rolls off gradually.

  • Stewart Doyle

    An analogy is a bit like a sports car with a motorbike welded to the side of it…

  • Tony L.

    Nice link Dominik, thanks for sharing! I agree, I only had 1 moire issue with my Hasselblad (H3D), and that was shooting a model with a jean shirt. Just to add, that the moire issue disappears at 80mp+. The great thing about what fuji is doing is they are producing images just as sharp as my H3D, with a fraction of the cost. I would say that if any photographer has a big issue with moire, they need to go to LF film and not think about it anymore, cause thinking about it takes time away from actually shooting.

  • Juhu

    There should be a sort of ‘continuous’ or analogue-like circular sensor surface from which the image is read or ‘scanned’ with a user-definable resolution.

  • Schlu

    I don’t use a camera to reproduce reality. I use a camera to create art. Not using a AA-filter is of course a valid tool to crate art. The viewers of my pictures love the results.

  • Zos Xavius

    He is using silkypix and capture one. I think the fuji jpeg engine is immune to this sort of thing. Raw processing and x transjust arent there yet……

  • Jonas

    I’ll just say I have yet to see an X series real-world (not lab) photo that was ruined by moirée.