Anti-Aliasing Filter Primer
Last updated February 12, 2012
With the recent announcement of the new Nikon D800 and its sibling the D800E - there's been a lot of renewed discussion about the anti-aliasing filter. It's present in the former and "cancelled out" in the latter (effectively lacking one). Of course, the Leica M8 lacked it entirely since its release in 2006 as with the M9 that followed in 2009. In this article, we'll take a look at the anti-aliasing filter and how the lack of one impacts digital M shooters.
When designing the digital M cameras, Leica made two very important decisions regarding the sensor in an effort to maximize the sharpness of the image. The first and perhaps most controversial was the decision to minimize the ultraviolet/infrared (or UV/IR) filter to its absolute minimum. In the M8, this had the unfortunate side effect, and gave rise to the "magenta issue." You can read more about that in our "UV/IR Filters" article. The second was to borrow a page from more pro-level, medium format digital backs and eliminate the anti-aliasing filter.
Technically called an optical lowpass filter, but often called a blur or AA filter - we'll just call it the latter for short, or anti-aliasing filter. Its purpose is to slightly blur the details (or optical points) in an image as it is recorded by your camera's sensor to eliminate an undesirable image aberration or artifact called moiré which we'll explain below. The blur is tiny - on a sub-pixel level. Each optical point is doubled in the horizontal and vertical axis - effectively blurring small details. Of course the pixels aren't fuzzy - they are what they are, a single point of color. But how they blend with neighboring pixels in color and brightness to represent detail is. The easiest way to explain the effect is by looking at two images and pointing out the differences.
Below we're going to compare two cameras that are roughly similar in specification in order to compare the effect of the anti-aliasing filter. It's not meant as a scientific test, but reasonable steps were taken to ensure a fair comparison. The cameras were tripod-mounted at a distance of 1m from the test target. The lenses used are both considered "professional" level, or the best each manufacturer has to offer - and were shot at f/5.6. Here is a comparison of the bodies, which represent the remaining variables:
|Camera||Crop Factor||Image Dim.||Resolution||Native ISO|
|Canon 1D Mark IIn||1.3x||3510x2345||8.2MP||100|
The differences account for a resolution that is 2,222,964 pixels (~2MP) apart and thus the slightly different scale of the two images. We didn't want to resize one or the other, or "split the difference" to avoid changing the images as little as possible from their "native" state. Thus, no image resizing was done and they are 100% crops of the middle. Once the crops were made, an "auto levels" was performed and the images saved. No sharpening whatsoever. It's best to click on either of the two images below for a pop-up window which will show each crop at full size (800x600) and you can quickly switch between them using the left or right arrow keys for comparison.
First the image from the Leica M8, shot at ISO 160 with the Leica Summilux-M 50mm f/1.4 ASPH:
And now the image from the Canon 1D Mark IIn, shot at ISO 100 with the Canon EF 50mm f/1.2L:
The difference is slight, but it's there. Forget the scale, white balance and contrast for a moment and focus instead on the definition of the black against white outlines. You'll notice the Canon image is softer than the Leica image. Granted, the performance of the two lenses is a bit of a factor and why we shot at f/5.6 - to minimize any focus shift or inaccuracy and to get both lenses into their optimum zone for center sharpness.
So what else can account for the difference in sharpness? The anti-aliasing filter. While the slight blurring of the black and white edges is one indication, look at the moiré (colored pattern) that is produced in the high frequency (fine detail) areas of the Leica image. This essentially means that the lens is providing more detail than the sensor can record. Between the two, the difference in sharpness becomes more obvious. Here's a blow-up of the area we're talking about:
Like we said, this wasn't meant to be a strict scientific test, but it's close enough to make the point. You can certainly sharpen either image, especially the one from the Canon to increase perceived sharpness. Often called "USM" for the "UnSharp Mask" or "Smart Sharpen" as it relates to Photoshop, or just "sharpening" in general. They work by increasing contrast between neighboring pixels on high-contrast borders. However, detail is detail and the more you have to begin with, the more (and more accurate) you end up with.
The downside to this increased sharpness and what the anti-aliasing filter is meant to combat in the first place is the phenomenon we mentioned above; moiré. Wikipedia defines moiré as "when two grids are overlaid at an angle, or when they have slightly different mesh sizes." This slight shift is what causes the color variations in the way light is passed through the bayer filter and onto the sensor. It occurs more readily with higher resolution sensors, finer details and sharper lenses. Things like finely textured fabric, bricks in buildings, or even fences. Often it can be eliminated by a simple rotation of the camera or change in distance (or focus). The problem is, there's no easy way to see this unless viewing the LCD preview at its highest magnification - and even then it's difficult if not impossible to accurately gauge. Thankfully, it can be cleaned up either partially or completely when post-processing. One popular program that handles it particularly well is Capture One Pro (by Phase One).
Using the definition of moiré from Wikipedia then, how does it apply in the photos above? Think of one "grid" being the bayer filter and pixels in your sensor and the second "grid" being the fine lines on the test chart. If you were to rotate the camera slightly in either direction, the moiré pattern would change, possibly disappearing for the most part at just the right alignment. In any event, in real world shooting this is next to impossible to do and while moiré might disappear in one pattern, the world is full of them at all sorts of different angles. You might make it disappear in one pattern only to have it show up in another elsewhere.
How often then, is moiré a problem in day-to-day shooting? It really comes down to your subject matter mostly, how much resolution you have and how sharp your lenses are. When they combine in just the right way - you end up with moiré. But unless you shoot fine fabrics and moderately distant fine details more often than not, it's not generally a real problem and relatively rare. A fair trade-off, then - for sharper images the other 99% of the time.
Don't take this comparison as any sort of disparagement of anti-aliasing filters, or Canon and other manufacturers that use them (as most do). They serve a valid purpose for the "typical" digital camera user. It usually only concerns those of us that look at such details, figuratively and literally - when you're looking to get the very highest resolution from a particular format. Currently, most small format cameras (APS-C, 4/3 and m4/3, etc.) all have one as do most 35mm cameras (with some notable exceptions) while most medium format digital backs don't have one. Another way to look at it is most consumer cameras have the filter while more pro (or now prosumer, slowly) level cameras lean towards not having it.
The Leica digital M cameras having never had an anti-aliasing filter from the M8 through the M9-P, allow the sharpest possible image capture with the given hardware (sensor, bayer filter, etc.). Combined with some of the best optics available from Leica or other manufacturers such as Zeiss - can result in amazingly sharp and detailed images if your technique is up to snuff.