Anti-aliasing

[TonyW]

I am working through the various technical issues of photography in order to educate myself. One topic that has occupied my mind has been anti-aliasing. I had not heard of it until relatively recently.

As I understand it, anti-aliasing is effected by a filter in front of the sensor in the camera where it distributes some of the light directed towards a pixel to neighbouring pixels, thus tending to reduce the bandwidth of the signal on the sensor. A side effect is some blurring of the image. The Nikon 800E seems to be the only dSLR camera that does not use it.

My question is: Why is this not applied numerically to the raw data either in or out of the camera after the analogue to digital conversion? I would have thought that this would be similarly effective and could be better controlled and switched on and off as needed. Also, the manufacture of the camera would be simpler.

[Mike Buckley]

The Nikon [D]800E seems to be the only dSLR camera that does not use it.

The Nikon D7100 also has no anti-aliasing filter. There may be other cameras as well.

I'll leave the rest of the discussion to others, as I'm not qualified to go any further.

[Dave Humphries]

The Nikon D7100 also has no anti-aliasing filter. There may be other cameras as well.

The new D5300 apparently doesn't have one either, seems Nikon may have decided Tony is right

Cheers,

[Inkanyezi]

The simple answer is that there are advantages of applying anti-aliasing at this level. The necessary blurring would be greater if applied in post production. Pentax has chosen another approach in the latest high-end camera, where the sensor is moved about during exposure in sub-pixel increments, and Sony does the same in its latest FF model with 36 megapixel sensor, where the user may choose to use it or not. A similar method is used by some medium format cameras to another end, increasing the pixel count. Fujifilm has still another approach, with a different pattern, which also gives the advantage of larger dynamic range. So there are different methods. All have drawbacks, and all of them also have advantages in one way or another.

We will probably see more cameras in the near future without the filter as well as using the Pentax method of vibrating the sensor during exposure.

[Manfred M]

So far as I know the Nikon D7100 does not have an low-pass (AA) filter either. The Leica M series and medium format cameras don't use them either. My D800 has a "weak" AA filter and frankly when I compare my images to the ones taken with the D800E, I can't see any difference in the image quality, even when pixel peeping unless the D800E is used in near ideal conditions (pro glass and mounted on a really well anchored tripod).

Sensor pixels are made up repeating patterns, so when certain types of subjects are photographed, moire patterns appear in the image. By applying a light blurr to the light hitting the sensor, this relatively rare effect can be avoided. Unfortunately, it is very difficult to take out after the image has been captured, as there is no way to automatically identify an area the exhibits this problem in a capture. The same thing can be said abot trying to deal with this issue in post. Using an AA filter has been the best way to deal with this issue.

Would the manufacture of the camera be simpler if the AA filter were removed; perhaps, but not by much. Modern sensors detect IR and UV wavelengths, so some form of filter is required, regardless. The AA filter is just part of the assembly that sits infront of the sensor.

[MrB]

The Nikon 800E seems to be the only dSLR camera that does not use it.

For a year now the Pentax K-5 IIs, with no AA filter, has been available. (The K-5 II does have one.)

Philip

[dje]

Hi Tony, I look at it this way.

The array of pixels in a sensor forms a spatial sampling system, with each pixel collecting a sample. The sampling frequency is equal to the number of pixels per mm. Mr Nyquist established that with a sampling system, frequency components in the signal greater than half the sampling frequency will cause aliases or spurious signals. Moire patterns are an example of this. The AA filter restricts the bandwidth of the system to somewhere near the Nyquist frequency (ie half the sampling frequency). This is implemented by blurring over a radius of two or three pixels.

The idea of the AA filter is to prevent the aliases being produced. If you allow them to be produced and try and remove them later in software, you are dealing with a different procedure all together, which by all accounts, is a difficult one to perform.

As the pixel density of sensors continues to increase(higher sampling frequency), this should become less of an issue. The D800 is probably a case in point. As Manfred points out, it only needs a weak AA filter.

Dave

[Glenn NK]

The recently announced Olympus OM-D EM-1 (or is that D-E1 ) has no AA filter. And I think the previous EM-5 has none either.

I've only seen one photo published online that showed the effects of no AA filter (it was taken with a Nikon D800E).

So while it could be a problem (shooting something with a fine repeating detail - cloth for example), for most work it's not an issue at all. Landscape photography seems not to be affected, and this is likely why landscape photographers are so keen on the D800E.

Glenn

[Glenn NK]

Pentax has chosen another approach in the latest high-end camera, where the sensor is moved about during exposure in sub-pixel increments. . . .

We will probably see more cameras in the near future without the filter as well as using the Pentax method of vibrating the sensor during exposure.

Interesting - I did not know that. Thanks for the info.

Glenn

[Colin Southern]

Capture sharpening more or less reverses it's effect, which is needed anway to counter the two other forms of image softening that are part of the process (digitization and demosaicing).

Personally, I'd prefer my camera not to have it, but in reality it doesn't have any real-world affects on an image.

From a manufacturers perspective they need some kind of filter in front of the sensor to protect it and stop it responding to undesirable wavelengths anyway, so they possibly figured it was easier to add it at that stage (since it's essentially countered with correct (and global) capture sharpening) than it was to make people deal with the (more localized) consequences where they occurred.

[Inkanyezi]

In fact, the aliasing occurs anyway, as long as we look at the image on a monitor with a repeating pattern. It is also present when the image is reduced in size, as then it will be presented with fewer pixels than the original. Well anti-aliased parts of the image will then be completely blurred, while those parts where repeating patterns remain will have aliasing artefacts. So in essence, anti-aliasing might have little or no effect on an actual image, when it is captured on a medium with high spatial frequency, but later reduced to a lower spatial frequency that may cause moire in larger repeating patterns.

This can be seen rather clearly in the image below, where the vertical bars in the bridge fence, will appear different, depending on distance. In the first segments, in the upper left corner, they are not aliased to a great degree, so you can count the bars. Looking further out on the bridge, you reach a point where individual bars are more difficult to see, and then a diagonal pattern appears, followed by a few segments where there is a shading with a substantially smaller number of vertical "bars", while farther out, the segments are totally blurred - which was the purpose of anti-aliasing.

The effects are clearly visible when looking at the image in 100%, as when opening it in its own window or tab. You can also see how it changes if you enlarge the image step-wise in your browser with Ctrl+ or Ctrl and moving the mouse wheel. The image has been reduced with the Lanczos3 algorithm, and although the camera has an AA-filter, the aliasing artefacts are also there in the original image, although in different positions than in the reduced image.

So even though there is an AA filter in the camera, the moire might occur when the image is displayed - or printed.

[xpatUSA]

Just for completeness, none of the Sigma digital cameras have 'blur' filters.

Falk Lumo has compared the D800 with the D800e, see here:

http://www.falklumo.com/lumolabs/art...0AAFilter.html

As I understand the Nikon system, it's not just "a filter" - there's beam splitting and all kinds of stuff going on, perhaps a Nikon cognoscentum would enlighten us about that to save us having to post vague guesswork

[edit: Threadstopper Ted strikes again! ]
.