04/04/2014

Removing the AA Filter, Marketing Driven or a Great Idea?



Is removing the aliasing filter really such a smart idea?



Full frame image, even in this radically downscaled version you can see alaising on the steel shed, click on the image to zoom in a bit more and its even more obvious. The shot was taken with a Minolta 28-85 Zoom at 28mm on a Sony A900.  The 28mm end is very sharp in the middle of the frame, which can exacerbate alaising and moire'


As both the megapixels counts of cameras and our expectations of acceptable resolution have grown the call has gone out in earnest, "dump the anti aliasing filter".  I have recently come to the opinion that the dumping of the AA filter would be ill founded and there are far better ways to increase the clarity of your images if that's what you need.

For those not versed in the intricacies of digital sensors I will try to illuminate upon the reasons for the filter, its supposed deficits and the widely accepted impact it has on image quality, and for good measure, followed by a little reality and some actual photographic examples.

Virtually all sensors use a matrix of red, green and blue pixels and as the word "matrix" implies they are laid out in a fixed grid pattern.  The structure is completely different to film where the light sensitive elements, (silver halides) are randomly deposited upon the films surface.  This last point is very significant so hold onto it.

The normal method of arranging the pixels is known as a "Bayer pattern pixel matrix" its served us well offering efficiency and relative ease of image data conversion into the final photographic image. Some manufacturers have attempted to use different RGB layouts and/or designs, primarily Foveon (Sigma Corp) and Fuji. These alternative designs have proven successful in use, but regardless remain just minor players in the sensor world as they bring other complexities to the table, like designing raw converters that can actually work with them, much higher unit cost or more demanding in camera processing and thus heavy battery usage.

The Bayer layout has its issues though.  When you shoot subjects that contain repeating patterns the patterns often mis-align with the sensors pixel (sensels) spacings and create moire' patterns.  It's easier to see than explain so have a look at the picture below.  The Bayer layout can also cause closely related issues with colour accuracy, but as pixel counts have increased this later issue has become less of a problem.

Moire' patterns most frequently show up on architectural subjects and cloth fabrics, they are however rare with natural objects unless they have significant repeating patterns, which in practice means mainly macro shots of cacti etc, though I have seen it on blades of grass and reeds in otherwise perfect landscape images.

AA filters (anti aliasing) filters reduce or eliminate the moire'  and aliasing issues by acting as a slight soft focus filter directly in front of the cameras sensor. In essence the filter subtly blurs the resulting image, which effectively means it lowers the local resolution a little bit.  The losses are typically recovered by re-sharpening the image after it has been initially processed.  

Unfortunately the sharpening processes are not a free lunch often creating problems of their own, sometimes referred to as artifacts, especially if the original was a jpeg file. Most typically with heavy sharpening we see halos around details giving images that "digital look " but there are others problems as well.  In some cases the sharpening if overdone can actually obliterate very fine detail and even make image noise appear far worse by aliasing the noise patterns themselves.

AA filters are often held responsible for poor wide angle lens resolution in the image corners.  It's simple really, the AA filter sits in front of the sensor but there is a small air gap between it and the sensor.  Light rays travelling from the rear element of most wide angle lenses to the edges of the sensor do so at a much sharper angle than from a longer focal length lens, where the rear element is further from the sensor.  The light rays on the edges of the frame are more softened by the filter than the ones "on centre" simply because they are passing through a proportionally thicker layer of glass and a larger air gap to reach the actual pixel.  This issue also exacerbates chromatic aberration and is probably why some lenses that exhibited no CA on film cameras can be utterly awful CA monsters on certain digital cameras.

For the above reasons the depth of the air gap and thickness of the filter are quite critical, but are always at best are a compromise because what might be ideal for a tele lens could be complete overkill for a wide angle one.  If the filter was designed to work with one fixed focal length, the results could probably be optimised, for example the filter could be made thinner at the edges or even have a curved profile and equally its strength could be matched to the lenses cross field resolution.   In fact in a fixed lens camera you probably could get rid of the AA filter without dramas because the processing could be perfectly optimised to deal with it, I expect this is exactly how Sony have set the RX1 up.

Alas none of the above is possible with an interchangeable lens camera, a mechanical optimisation for one lens would inevitably make it far worse for many other lenses. and a software optimisation might go somewhere towards solving the problem but only for lenses from the makers own catalogue no doubt.


This is a section from the top image, you can see the moire and alaising  (false detail and colour artefacts) on the corrugated iron in particular,  this is a full frame 24 megapixel sensor with a strong AA filter to start with, imagine how bad it would be without one at all.  The effects also tend to exacerbate chromatic aberration in my expereince.  Click to enlarge for a better look.


In this version the Colour Moire' is fixed, there is still some false detail present but overall the result is good, this was done by using a moire' removal tool in a specialised raw convertor.  But it is not without trade offs, if you look closely you can see the colour is less vibrant and in fact quite a bit of the colour definition has gone missing in action, especially on the railway ballast.


Like many aspects of digital imaging the reality of the AA filters role is not quite as straightforward as is often believed by those on web forums advocating "lets dump the AA filter".

The final clarity of an image is the result of many factors, any of which could have a far more significant effect on the final detail of the image than the AA filter does by itself.  The most obvious issue is lens performance, in fact if you shoot with a soft lens it is highly unlikely you would ever need an AA filter.

But there is more, much more, try this list on for size.

Poor focus will lower resolution enormously and in my experience a very significant proportion of regular images are not critically focused.  The differences between sharp and really sharp focus is incredibly fine.....much finer than most auto focus systems are capable of.

Small scale Camera movement will subtly blur the image reducing potential aliasing and moire'

JPEG compression can seriously mess with resolution post capture,  blurring out fine details that would otherwise show aliasing.

Noise, all noise, degrades potential resolution and thus reduces aliasing and moire'.

All noise reduction methods sap the life out of fine details and mask detail aliasing significantly.

Raw conversion software can have an enormous influence on aliasing,  those that are better at extracting fine detail typically exacerbate alaising

So here is the thing, removing filter will not in all honesty gain you anything unless all of the above have been dealt with and WELL controlled.  

But there is a twist, should you actually take the time to sort all of the above issues then you may well find yourself being very grateful for that AA filter.

You don't believe me?

Allow me to tell you about how I shoot with my current general purpose tool. the Sony Nex 5n.

I use the NEX for the great majority of my non-professional work and occasionally for available light pro stuff, I love it and have shot around 6000 frames in the last 3 months with the micro beast.  The 5n does have an AA filter, but it is known to be fairly weak.

So has the filter limited the resolution I am getting?  Not in the least!  

Have I had issues with aliasing and moire' as a result of the weak filter?  Absolutely!  

I see aliasing often and find myself constantly having to deal with it, removing the filter entirely would just make my editing life miserable.

Here is why.

I shoot a lot of architecture and also subjects that are likely to create aliasing issues, but more importantly, remember that list above,  I have basically removed every single one of those items as a factor in limiting resolution. 

I shoot with sharp legacy lenses at optimal apertures (typically Nikon), the focus is as dialled as I can possibly get it using 14.8x live view, I use a tripod whenever needed, electronic first curtain shutter is used to eliminate vibration.  Going further I don't shoot jpegs, rarely shoot above 400 iso, use raw software that extracts all the detail in the file with all noise reduction turned off and beyond that I often shoot using “sensor balanced capture”.

Ultimately for me the only possible source of resolution loss left is that AA filter and it frankly is not a factor, quite the opposite.  With all the issues dealt with the weakness of the 5n AA filter is a constant problem.  How much of problem?  I would calculate that of the 12000 or so files I have shot on that camera at least 400-500 have shown some aliasing issues unless I compromise something along the work flow to negate it....which usually means I do it in the raw processing.

The inconvenient truth is that in many cases I cannot fully control the aliasing, that dear reader is a universal truth about aliasing, if you have it, getting rid of it can be really really hard, and trust me I know all the methods out there to do so. Basically aliasing has to be dealt with at the sensor, doing anything else is shutting the gate after the horse has bolted.

The camera makers know this and have always known it, hence the fact they virtually all had AA filters till recently.  Frankly I think the only real reason for makers offering cameras without AA filters is marketing, they have worked out the a camera "sans AA filter"offers a pathway to a few extra sales with almost no extra manufacturing costs involved.

The marketing is aided and abetted by the fact that camera reviews pixel peep files and at a 100% pixel peeping view the non AA filtered file may look just a little sharper, and sharper means extra sales!

The Nex 5n is not by any means the only camera I have used or tested that exhibited aliasing when all the detail/resolution masking issues are removed. The examples you see on this page were from a Sony A900 which is full frame 24 megapixels, it has a pretty strong AA filter but still the problems are obvious. So here is my take on the AA filter.

Generally I feel AA filters should be maintained and as much as possible optimised for the camera in question, photographers who want or need more resolved detail should examine and modify their technique or tools before jumping to the conclusion that the AA filter is the cause of their problems or the pathway to optical nirvana. 

Let me me try to make this a crystal as possible, if you for example use the Sony Kit NEX lens on say the NEX 5N at focal lengths between 35-55mm I doubt you could ever see alaising, even without a filter, the lens simply does not resolve well enough, and this is the case with a lot of similar lenses.  Removing the filter will not make a poor or average lens resolve any better.....period!

If your technique is slack and you only shoot jpegs but use really good lenses then perhaps the lack of an AA filter might be a slight bonus as it will remove one of the potential limits.

Should you think, it won't worry me as I only shoot landscapes, well yes perhaps your right, but most folk shoot a wide variety of subjects, and you'd be surprised at how many of them have aliasing inducing patterns present.  I have seen aliasing on corrugated iron, brickwork, mesh, fences, windows, cars, and hundreds of other items, trust me in a resolution optimised system aliasing is nowhere near as rare as most armchair webtoghraphers would have you believe!

As resolution of sensors climb  It is likely the AA filter will not need to be as strong, but I doubt it can be fully dispensed with as higher resolutions will enable larger print sizes aliasing will still eventually show up although it's effect will occur on ever finer repeating patterns than it does now.

I like to think of the AA filter as analogous to the safety features on your car, like airbags, traction control etc.  You don't need them all the time or indeed often, but when you do they will save your bacon.  And it's much much harder to sort out a bingled car than it is to have avoided the accident and injury in the first place.





No comments:

Post a Comment