September 30, 2014, 04:45:00 PM

Author Topic: Will Canon ditch the AA Filter?  (Read 20590 times)

CarlTN

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 2227
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #30 on: January 10, 2014, 09:00:18 PM »
It seems to me that if one wants more resolution, one should design and build, or buy a camera with enough pixels on the sensor.  That way, you don't have to worry about whether it has an AA filter (or a strong AA filter).  You can get a camera that has an AA filter, yet also has "enough" pixels.

20 MP should be enough for most situations.  If it isn't, stripping off, or weakening the AA filter, is the wrong way to go about trying to extract more resolution.  Unless of course the sensor doesn't need an AA filter, because it's not a bayer mosaic color array in the first place...Not sure how the Fuji array compares, but it might have less inherent moire. 

In any case, the desire to get extra resolution, should really only be an issue for photogs who print somewhat large, or who crop moderately or heavily into an image.  In both cases, skill, lenses, and technique play as much a role as anything hardware related.  Post process editing skill, also plays a big role.  You can even correct moire in Lightroom, it works pretty well.  You can also add it, which can enhance resolution if combined in the right way with the other sliders...and get a slight increase in resolution in parts of the image.  Or at least I have.  This has worked interestingly, with some of my images shot with sensors that weren't a bayer array...

canon rumors FORUM

Re: Will Canon ditch the AA Filter?
« Reply #30 on: January 10, 2014, 09:00:18 PM »

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4723
  • POTATO
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #31 on: January 11, 2014, 02:23:58 AM »
I don't buy that. Many things can be corrected in post, and moire isn't one of them. pcpro.co.uk aren't renowned for their photographic expertise. In this case I get the impression they just read a couple of marketing gimmicks and wrote it as fact.

So you would be saying that Adobe's moire correction brush cannot work, right?

Moire correction in post currently only takes care of color moire. In that, it only eliminates the color, but not really the moire. There are some moire patterns that Adobe's moire brush can deal with better than others, and if all you have is very light moire (like the kind you might bet with a just barely too weak AA filter), you might be able to clean it up entirely...albeit with consequences, you will lose detail in one way or another.

However, if you completely remove the AA filter from the sensor, there is absolutely no way that Adobe's moire brush is going to clean it up. It will remove color from it, but the actual moire pattern is baked into the image. Moire is a complex convolution, and without knowing the parameters of the interference that created it, you can't do anything to reverse it once it is baked in.

That's WHY we have optical low pass filters...they are really the only way to prevent aliasing and moire.
My Photography
Current Gear: Canon 5D III | Canon 7D | Canon EF 600mm f/4 L IS II | EF 100-400mm f/4.5-5.6 L IS | EF 16-35mm f/2.8 L | EF 100mm f/2.8 Macro | 50mm f/1.4
New Gear List: SBIG STT-8300M | Canon EF 300mm f/2.8 L II

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4723
  • POTATO
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #32 on: January 11, 2014, 03:49:09 AM »
Yes, you can get slightly more detail without an AA filter, but for many forms of photography the minor advantage is nothing compared to the major disadvantage. Also, some here have argued that if the number of MP are increased to the point where the sensor out resolves most lenses, you can get away without it. My question is, in that case, why do you need no AA filter? If the output is already optically blurred at a pixel level by the lens, no lack of AA filter will sharpen it back up. And then what happens if you buy a new yet-to-be-released super sharp lens in the future and it does out resolve the sensor at certain apertures? Moire, and you being forced to defocus some shots. So why not just keep the AA filter?

If we had 32MP APS-C sensors (83MP FF), I would just be jumping for joy if a lens came out that out-resolved it. Really, I doubt any lens will ever come out that significantly out-resolves current lenses at their best (usually macro lenses at f5.6). If we pick something good enough for those it shouldn't be a problem.

Resolving power increases with aperture. At f/5.6, MTF 50% (standard measure for photography), with a diffraction limited lens you have 123lp/mm of resolving power. We already have sensors that resolve that much, and it is certainly no stretch to say that many lenses are diffraction limited at f/5.6. At f/4, diffraction limited MTF50 resolving power increases to 173lp/mm. At f/2.8, diffraction limited MTF50 resolving power increases to 247lp/mm. Now we are really pushing resolution, however making a truly diffraction limited f/2.8 lens is a much more difficult ordeal than making a diffraction limited f/4 or f/5.6 lens. Zeiss once had a lens they had designed explicitly to test high resolution films, which was capable of resolving about 400lp/mm (which would have been diffraction limited around f/1.7-1.8)...however I'd be doubtful if many major brand name lenses, including any from Ziess these days, was actually diffraction limited at apertures above f/4. If we assume future lenses get better, then the only way they could resolve more would be to increase resolving power at apertures wider than f/4...maybe up to 200lp/mm at f/2.8 (not diffraction limited, but still better than anything we have today.)

Now, don't forget that the final resolving power of an entire camera system is a convolution of its component parts. We can't really know the exact PSF of any lens or camera sensor (some manufacturers probably do, but they don't actually publish the information), but we can use a simple formula to approximate: SQRT( lensBlur^2 + sensorBlur^2) allows us to determine total system blur, and from there we can extend the formula to tell us a whole lot of things. Because of the nature of system resolving power as demonstrated by this formula, you can never actually reach the maximum resolving power of your least powerful component...you can only approach it. That means, no matter how high your sensor resolution is, you could never actually "out"-resolve a diffraction limited f/4 lens at f/4...at best, you would be able to resolve 172.99999 with the whole system. The notion of either a sensor or a lens "outresolving" the other is a bit of a misnomer.

To demonstrate how ironic and even a little ridiculous this little relationship is...you would need an APS-C (1.6x) sensor with 1115000x745000 pixels (22.3mmx14.9mm) to render 172lp/mm spatial resolution in an output image with a diffraction limited f/4 lens...that is an 830.7 gigapixel sensor! :P Obviously that's impossible...the pixels would be 10 nanometers in size, and pixels that small would be too small to allow light through, so the sensor simply wouldn't function (not, at least, with visible light...it might function with gamma rays. :P)

The only other way to increase total system resolution is to use a lens with higher resolving power, which can only be achieved at wider apertures. If we assume we have our 200lp/mm f/2.8 lens (which gives us plenty of headroom to work with), to resolve 173lp/mm we would need a 158mp APS-C sensor (15380x10277 pixels @ 22.3mmx14.9mm). That comes out to 411mp FF (24830x16553 pixels @ 36mmx24mm). At any aperture below f/2.8, your total system resolving power would again become diffraction limited, and would be less than 173lp/mm...yet still higher than if you used a sensor with fewer megapixels.

Sorry if that comes off as too complicated. However it is the only way to be clear about "resolution". Sensors do not outresolve lenses. Lenses do not outresolve sensors. The two work together to create a final outcome, and that final outcome will continue benefit by increasing the resolution of either lens or sensor for a long time to come. At the moment, the best APS-C sensors are resolving around 124lp/mm. The best FF sensors resolve far less, and further still the actual output resolution of our actual photographs is even less (and if we apply noise reduction, EVEN LESS!). We have a LOT of headroom before we run into the limitations offered by current f/4 and faster lenses (many of which, while not diffraction limited, still offer resolving power that can be well above 124lp/mm) and experience diminishing returns. An 32mp APS-C/83mp FF sensor doesn't even scratch the surface of how far we could take sensor resolution before we started experiencing diminishing returns the the degree where there was no point in investing in shrinking pixels any further.

I'd actually bet that we could take pixels right down to the wavelength size of light (the point at which pixels become too small to allow visible wavelengths of light through), and still have a useful gain. As a matter of fact, with the new generation of small form factor CMOS image sensors (CIS) that will become mainstream through 2014-2015...the tiny sensors used in smartphones and the like...reaching 0.95µm, or 950nm, we are already well into the realm of wavelength-size or smaller for about half the range of infrared light. Within the next generation or two of smartphone CIS designs, pixels will be as small as they possibly can be...about 750nm, and we won't be able to shrink them any further without filtering out red light!!

This convolution of lens and sensor resolution should also make it clear that there won't necessarily be any benefit from removing AA filters for a very long time to come as well. Everything I've discussed here is at MTF50, or a 50% contrast level. As pixel size shrinks, sensors will be able to resolve detail at lower and lower levels of contrast as well, so the spatial resolutions scale up (more lp/mm) as pixel size drops. I wouldn't go so far as to say a sensor could produce any real meaningful result at MTF10 (10% contrast), but it is very likely that small pixels will be able to resolve detail at 20%, 15% contrast...and lens resolving power at those levels is considerably higher than it is at 50% contrast (i.e. at f/4 MTF15, resolution is closer to 350lp/mm!) We will always need AA filters...there really isn't any good reason to get rid of them.

The only use case where not having an AA filter might result in sharper detail would be if you only ever photograph scenes without ANY repeating patterns of any kind whatsoever. Landscape photography is the most likely scenario where you would encounter only purely random detail at nyquist, but even in the case of landscapes...personally, I don't like them being too sharp. I find that the best landscape photography tends to have that certain "softish" quality to them, clean edges, rather than finger-dicing sharp edges, kind of like "bloom" in modern games makes everything "soft", but in general just much better.
« Last Edit: January 11, 2014, 03:52:59 AM by jrista »
My Photography
Current Gear: Canon 5D III | Canon 7D | Canon EF 600mm f/4 L IS II | EF 100-400mm f/4.5-5.6 L IS | EF 16-35mm f/2.8 L | EF 100mm f/2.8 Macro | 50mm f/1.4
New Gear List: SBIG STT-8300M | Canon EF 300mm f/2.8 L II

chromophore

  • Power Shot G16
  • **
  • Posts: 25
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #33 on: January 11, 2014, 11:11:53 AM »
Anyone who thinks AA filters are intrinsically bad should be required to take a graduate level course in digital signal processing.  Too many people talk about this stuff as if they actually understand the relevant engineering principles, when in fact they're just repeating marketing gimmicks.

ajfotofilmagem

  • 1D Mark IV
  • ******
  • Posts: 944
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #34 on: January 11, 2014, 11:34:35 AM »
Anyone who thinks AA filters are intrinsically bad should be required to take a graduate level course in digital signal processing.  Too many people talk about this stuff as if they actually understand the relevant engineering principles, when in fact they're just repeating marketing gimmicks.
Yeah, those people obsessed by the highest possible sharpness forget that very sharp images seem artificial to the human eye. Ok, we see with the eye-brain together, and exceedingly sharp images bother our vision. That's why all movies cartoon latest 3D, simulate "imperfections" of the lens, lighting, texture objects, etc.. If objects appear too perfect, is not convincing human vision, and become blatantly artificial.
« Last Edit: January 11, 2014, 11:36:35 AM by ajfotofilmagem »

Don Haines

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3239
  • Posting cat pictures on the internet since 1986
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #35 on: January 11, 2014, 12:00:13 PM »
Yes, you can get slightly more detail without an AA filter, but for many forms of photography the minor advantage is nothing compared to the major disadvantage. Also, some here have argued that if the number of MP are increased to the point where the sensor out resolves most lenses, you can get away without it. My question is, in that case, why do you need no AA filter? If the output is already optically blurred at a pixel level by the lens, no lack of AA filter will sharpen it back up. And then what happens if you buy a new yet-to-be-released super sharp lens in the future and it does out resolve the sensor at certain apertures? Moire, and you being forced to defocus some shots. So why not just keep the AA filter?

If we had 32MP APS-C sensors (83MP FF), I would just be jumping for joy if a lens came out that out-resolved it. Really, I doubt any lens will ever come out that significantly out-resolves current lenses at their best (usually macro lenses at f5.6). If we pick something good enough for those it shouldn't be a problem.

Resolving power increases with aperture. At f/5.6, MTF 50% (standard measure for photography), with a diffraction limited lens you have 123lp/mm of resolving power. We already have sensors that resolve that much, and it is certainly no stretch to say that many lenses are diffraction limited at f/5.6. At f/4, diffraction limited MTF50 resolving power increases to 173lp/mm. At f/2.8, diffraction limited MTF50 resolving power increases to 247lp/mm. Now we are really pushing resolution, however making a truly diffraction limited f/2.8 lens is a much more difficult ordeal than making a diffraction limited f/4 or f/5.6 lens. Zeiss once had a lens they had designed explicitly to test high resolution films, which was capable of resolving about 400lp/mm (which would have been diffraction limited around f/1.7-1.8)...however I'd be doubtful if many major brand name lenses, including any from Ziess these days, was actually diffraction limited at apertures above f/4. If we assume future lenses get better, then the only way they could resolve more would be to increase resolving power at apertures wider than f/4...maybe up to 200lp/mm at f/2.8 (not diffraction limited, but still better than anything we have today.)

Now, don't forget that the final resolving power of an entire camera system is a convolution of its component parts. We can't really know the exact PSF of any lens or camera sensor (some manufacturers probably do, but they don't actually publish the information), but we can use a simple formula to approximate: SQRT( lensBlur^2 + sensorBlur^2) allows us to determine total system blur, and from there we can extend the formula to tell us a whole lot of things. Because of the nature of system resolving power as demonstrated by this formula, you can never actually reach the maximum resolving power of your least powerful component...you can only approach it. That means, no matter how high your sensor resolution is, you could never actually "out"-resolve a diffraction limited f/4 lens at f/4...at best, you would be able to resolve 172.99999 with the whole system. The notion of either a sensor or a lens "outresolving" the other is a bit of a misnomer.

To demonstrate how ironic and even a little ridiculous this little relationship is...you would need an APS-C (1.6x) sensor with 1115000x745000 pixels (22.3mmx14.9mm) to render 172lp/mm spatial resolution in an output image with a diffraction limited f/4 lens...that is an 830.7 gigapixel sensor! :P Obviously that's impossible...the pixels would be 10 nanometers in size, and pixels that small would be too small to allow light through, so the sensor simply wouldn't function (not, at least, with visible light...it might function with gamma rays. :P)

The only other way to increase total system resolution is to use a lens with higher resolving power, which can only be achieved at wider apertures. If we assume we have our 200lp/mm f/2.8 lens (which gives us plenty of headroom to work with), to resolve 173lp/mm we would need a 158mp APS-C sensor (15380x10277 pixels @ 22.3mmx14.9mm). That comes out to 411mp FF (24830x16553 pixels @ 36mmx24mm). At any aperture below f/2.8, your total system resolving power would again become diffraction limited, and would be less than 173lp/mm...yet still higher than if you used a sensor with fewer megapixels.

Sorry if that comes off as too complicated. However it is the only way to be clear about "resolution". Sensors do not outresolve lenses. Lenses do not outresolve sensors. The two work together to create a final outcome, and that final outcome will continue benefit by increasing the resolution of either lens or sensor for a long time to come. At the moment, the best APS-C sensors are resolving around 124lp/mm. The best FF sensors resolve far less, and further still the actual output resolution of our actual photographs is even less (and if we apply noise reduction, EVEN LESS!). We have a LOT of headroom before we run into the limitations offered by current f/4 and faster lenses (many of which, while not diffraction limited, still offer resolving power that can be well above 124lp/mm) and experience diminishing returns. An 32mp APS-C/83mp FF sensor doesn't even scratch the surface of how far we could take sensor resolution before we started experiencing diminishing returns the the degree where there was no point in investing in shrinking pixels any further.

I'd actually bet that we could take pixels right down to the wavelength size of light (the point at which pixels become too small to allow visible wavelengths of light through), and still have a useful gain. As a matter of fact, with the new generation of small form factor CMOS image sensors (CIS) that will become mainstream through 2014-2015...the tiny sensors used in smartphones and the like...reaching 0.95µm, or 950nm, we are already well into the realm of wavelength-size or smaller for about half the range of infrared light. Within the next generation or two of smartphone CIS designs, pixels will be as small as they possibly can be...about 750nm, and we won't be able to shrink them any further without filtering out red light!!

This convolution of lens and sensor resolution should also make it clear that there won't necessarily be any benefit from removing AA filters for a very long time to come as well. Everything I've discussed here is at MTF50, or a 50% contrast level. As pixel size shrinks, sensors will be able to resolve detail at lower and lower levels of contrast as well, so the spatial resolutions scale up (more lp/mm) as pixel size drops. I wouldn't go so far as to say a sensor could produce any real meaningful result at MTF10 (10% contrast), but it is very likely that small pixels will be able to resolve detail at 20%, 15% contrast...and lens resolving power at those levels is considerably higher than it is at 50% contrast (i.e. at f/4 MTF15, resolution is closer to 350lp/mm!) We will always need AA filters...there really isn't any good reason to get rid of them.

The only use case where not having an AA filter might result in sharper detail would be if you only ever photograph scenes without ANY repeating patterns of any kind whatsoever. Landscape photography is the most likely scenario where you would encounter only purely random detail at nyquist, but even in the case of landscapes...personally, I don't like them being too sharp. I find that the best landscape photography tends to have that certain "softish" quality to them, clean edges, rather than finger-dicing sharp edges, kind of like "bloom" in modern games makes everything "soft", but in general just much better.
Wow! That certainly makes for a thought-provoking read....

On a completely different way to look at AA filters, Does anyone really think that Canon (and other companies) would go through the trouble and expense of designing and installing AA filters if they were not needed?

Perhaps it can be done through software.... I don't know. What I do know is that there is no such thing as a free lunch. How good is software de-aliasing? What does it do to the picture? What are the effects on IQ? Does it kill the burst rate? Can you turn it on/off in camera or adjust sensitivity? What about "false positives" and misses? What effect does it have on editing the picture later?.

Keep in mind that a LOT!!! of people shot in Jpeg.... RAW shooters are the minority. To do the anti-aliasing in software means real-time in the camera and that's a huge gob of computing power... and that has the potential to cripple your frame rate and suck the life out of your battery.
The best camera is the one in your hands

chromophore

  • Power Shot G16
  • **
  • Posts: 25
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #36 on: January 11, 2014, 01:28:41 PM »
It's kind of funny how someone can, in the same breath, vilify optical AA filters for killing sharpness yet extoll the virtues of anti-moire image processing algorithms, and expect to be taken seriously.

I should also mention that just because something is "sharp" doesn't mean it's a component of a real signal.  That's the whole reason why moire artifacts are observable in discretized data--it is sharp, but it is also spurious.

canon rumors FORUM

Re: Will Canon ditch the AA Filter?
« Reply #36 on: January 11, 2014, 01:28:41 PM »

seekthedragon

  • Power Shot G16
  • **
  • Posts: 20
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #37 on: January 11, 2014, 01:42:05 PM »


I don't buy that. Many things can be corrected in post, and moire isn't one of them. pcpro.co.uk aren't renowned for their photographic expertise. In this case I get the impression they just read a couple of marketing gimmicks and wrote it as fact.

So you would be saying that Adobe's moire correction brush cannot work, right?

Right. It is like P mode: it does this and that, but only takes the intended shot by chance, since the camera has no idea about the current scenario.

If you would like to remove moire, you need to know exactly, what should be on the picture. You can't use the picture for that, as it already contains aliasing. But if you saw the scene, you may know what is wrong, and what is right. If you saw millions of scenes, you may put this knowledge in a computer, and correct these moire errors, if they appear. It may happen, that you make mistakes, as you weren't present at each of the actual shots, and you have only a guess based on your experience. But more or less, it may be useful.

However, there is no way to correct moire using only the picture and having no information about what should be there. It is not technology, not even physics, it is pure mathematics. So it won't change even in another universe.

dilbert

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 2975
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #38 on: January 11, 2014, 05:49:13 PM »
So for all those that believe the AA filter is necessary, how often have your photographs been saved by it?

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4723
  • POTATO
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #39 on: January 11, 2014, 05:54:21 PM »
So for all those that believe the AA filter is necessary, how often have your photographs been saved by it?

Oh, I'd say at least 60,000 times. ;) I photograph birds. Bird feathers are massively prone to creating moire patterns. I also have another 2000+ landscape photos, where my AA filter avoided jagged-edged tree & bush branch, grass blade, and mountain peak edges. I could probably scrounge up more cases. Lot of architectural stuff around here would definitely create moire patterns. The majority of my family photographs would have exhibited moire in clothing. So, yeah, at LEAST 60,000 times, probably closer to 75,000 times.

If Canon was going to offer one major improvement in their next sensors, and it was a choice between DR and no AA filter, I'd pick DR no question. If the choice was between fewer megapixels and better high ISO and no AA filter, I'd pick the fewer megapixels. If the choice was between pretty much anything and no AA filter, I'd pick the other thing. There is no other way to eliminate moire and other forms of aliasing...it has to be done before the aliasing is baked into the image.
« Last Edit: January 11, 2014, 05:56:27 PM by jrista »
My Photography
Current Gear: Canon 5D III | Canon 7D | Canon EF 600mm f/4 L IS II | EF 100-400mm f/4.5-5.6 L IS | EF 16-35mm f/2.8 L | EF 100mm f/2.8 Macro | 50mm f/1.4
New Gear List: SBIG STT-8300M | Canon EF 300mm f/2.8 L II

tcmatthews

  • Rebel T5i
  • ****
  • Posts: 143
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #40 on: January 11, 2014, 07:50:55 PM »
It seems as if the usual suspects are taking sides again.  The biggest thing to remember about engineering anything is that there are always trade-offs.  An AA filter is not magic.  It has a cost in image quality.  And the higher the megapixels the less benefit in having one.  No implementation of software can be a magic solve all problems bullet. 

First all an AA filter or an optical low pass filter does is blur the image at a sub-pixel level.  Possibly up to a 2-3 pixel level in the worst case of a strong filter.   Then the sensor basically down samples the resulting image.   You have no control over the level of the blur because it was carefully designed and implemented by the camera manufacture in hardware.  After the senor reading sharpening is applied to the image to lessen the effects of the AA filter normally in post-processing. 

The important thing to remember is you can remove moire in software.  I did it all the time when using my scanner.  But it is not free it also has drawbacks  mainly the need to down-sample and then sharpen the image.  But it also has the benefit of being controllable.  If you had true raw sensor readings for video you could use this technique to remove moire in video selectively.  (Do not know if the software exists but this is done all the time in video games and CGI.  So it should not be hard to create.  Given video card hardware is designed to do this sort of work is should be fast.)   

Personally on DSLR I prefer the AA filter to be weak.  To keep the images as sharp as possible.  I do not mind a little moire or selectively applying Gaussian blur to remove it.  Another option would be to produce to version one with and another without the AA filter.  On a high megapixel I could careless even if they removed it completely.  That said I think some on this forum are allergic to even the slightest moire. 

For a high megapixel mirrorless such as the Sony Ar7 it makes more sense to remove the AA filter.  This was likely done not because it improved sharpness but because the thickness of the filter caused more problems with color shift on wide angle lenses than the AA filter solved with moire. 

Will Canon ditch the AA filter well that will depend on the engineering trade-offs.  So only Canon knows.  And on this forum we will complain regardless. 
6D, 60D, Rebel XS, EF-S 15-85,  EF 70-200f4L, EF 100f2.8L,  Tamron 70-300VC,  Tamaron 28-75f2.8,   EF 40f2.8,  EF 50f1.8, EF 85f1.8, EF 24mmf2.8IS  EOS M 22f2, 18-55, EOS M-EF adapter, Nex6,  SEL50f18,  SEL1018,  DN 30f2.8, adpaters and Fd and M42 lens.

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4723
  • POTATO
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #41 on: January 11, 2014, 09:49:42 PM »
It seems as if the usual suspects are taking sides again.  The biggest thing to remember about engineering anything is that there are always trade-offs.  An AA filter is not magic.  It has a cost in image quality.  And the higher the megapixels the less benefit in having one.  No implementation of software can be a magic solve all problems bullet. 

First all an AA filter or an optical low pass filter does is blur the image at a sub-pixel level.  Possibly up to a 2-3 pixel level in the worst case of a strong filter.   Then the sensor basically down samples the resulting image.   You have no control over the level of the blur because it was carefully designed and implemented by the camera manufacture in hardware.  After the senor reading sharpening is applied to the image to lessen the effects of the AA filter normally in post-processing. 

The important thing to remember is you can remove moire in software.  I did it all the time when using my scanner.  But it is not free it also has drawbacks  mainly the need to down-sample and then sharpen the image.  But it also has the benefit of being controllable.  If you had true raw sensor readings for video you could use this technique to remove moire in video selectively.  (Do not know if the software exists but this is done all the time in video games and CGI.  So it should not be hard to create.  Given video card hardware is designed to do this sort of work is should be fast.)   

Personally on DSLR I prefer the AA filter to be weak.  To keep the images as sharp as possible.  I do not mind a little moire or selectively applying Gaussian blur to remove it.  Another option would be to produce to version one with and another without the AA filter.  On a high megapixel I could careless even if they removed it completely.  That said I think some on this forum are allergic to even the slightest moire. 

For a high megapixel mirrorless such as the Sony Ar7 it makes more sense to remove the AA filter.  This was likely done not because it improved sharpness but because the thickness of the filter caused more problems with color shift on wide angle lenses than the AA filter solved with moire. 

Will Canon ditch the AA filter well that will depend on the engineering trade-offs.  So only Canon knows.  And on this forum we will complain regardless.

As megapixel count AND lens resolving power increases (which is the actual state of things today), the need for an AA filter will remain the same. The need for an AA filter would only diminish as megapixel count increased if lenses did not improve. This assumes that one determines lens resolving power at MTF50. As pixels get smaller and quantum efficiency increases, sensors will be more capable of separating fine detail at lower levels of contrast...so one would technically need to address that by measuring lens resolving power at lower and lower levels of contrast (say, MTF20, maybe even MTF10...hell, when it comes to astrophotography, resolving power is already measured at MTF10, or the Rayleigh Criterion for overlapping airy discs, and MTF0 which is actually just above zero contrast, where airy discs have merged, but minor perturbations in the shape of the airy disc of close stars allows the number and magnitude of those stars to be determined. Not really applicable for photography, but just to point out that lens resolving power, depending on the level of contrast, can FAR, FAR outpace sensor spatial resolution...by a massive degree.)

The cost in removing AA with software is very great...at least in my experience. For color moire, you lose a lot of color fidelity...if you have strong moire to deal with, you often end up with largely monochrome detail after removing it. When it comes to actually removing monochrome moire, the cost is very great in terms of detail, which usually gets blurred or smeared, obliterating all that extra detail you eliminated the AA filter for in the first place. I don't know how your scanner software did it, but the best tools available for removing moire in photography really only work at the color level (generally ignoring the actual monochrome moire pattern), and when you dial them all the way up, the blurring is fairly considerable.

Even advanced manual techniques like this one are still only able to remove color moire...monochrome moire patterns still remain. Without an AA filter, the moire pattern is far more severe, and that much more difficult to remove. Lightroom's moire brush works on the RAW data before it is demosaiced, so it is working at the most effective level already. RawThearapy's moire removal introduces considerable blur, eliminating detail that was actually resolved, in order to remove monochrome moire...which effectively negates the point of not having an AA filter in the first place.

BTW, if you know of an actual, effective means of removing moire in post, one that is able to deconvolve it or invert the interference pattern without causing significant image degradation, I would love to hear about it. I used to mess with Lightroom's moire removal on DPR sample images and the like, and it often appears to be quite effective in those cases. It was only when I started encountering some moire with my bird photography (which is one of the consequences of having a much sharper lens like the 600/4 II), I realized how sorely ineffectual the tool is for monochrome moire. I've sometimes resorted, when I have the option, to ever so slightly defocusing to eliminate moire with the lens, as it is a vastly more effective way of solving the problem.

An optical low pass filter separates high resolution detail just enough that it no longer causes an interference pattern. It is debatable whether an AA filter has a cost in image quality...many would say it actually restructures the information the moment before it is discretized by the sensor such that it is MORE accurate, not less, and therefor preserves image quality, rather than destroying it. For all intents and purposes, fine detail resolved by a lens that falls below the nyquist frequency is nonsense. Because of aliasing, it is effectively random...noise. There is largely no reason to keep that jumbled up information...it's all fake, all "aliases", and anti-aliasing it reduces the noisy effect created by all that nonsense. Information right AT nyquist frequency might be something that can be partially preserved. This is really where "taste" comes into play. Some prefer a slightly weaker AA so nyquist data appears sharper. It is, by definition, more aliased, but if you don't actually have problems with moire (i.e. you aren't photographing any repeating patterns), I can see where aliased detail might be desirable if your goal is to have ultra-sharp results. Anti-aliasing that information doesn't necessarily destroy information though...it restructures it, and many, such as myself, find that it is more accurate and realistic and pleasing without the harsh, jagged edges and pinpoint sparkling highlights. To each his own here, and there are certainly options...you can pick and choose from a number of cameras from a variety of brands that have varying strengths of AA filters.

As for high resolution...lenses progress right along with sensors. Depending on the contrast level you determine your lens resolving power at, it is not incorrect to state that even at f/8, a lens at MTF10 (Rayleigh, low contrast) can resolve more than enough detail (186lp/mm) to still absolutely require a low pass filter for APS-C and FF sensors (and will, for quite some time). Today, at MTF50, f/5.6 still resolves more than enough detail in the context of FF sensors to absolutely require a low pass filter. At MTF50, f/4 and up all have the potential to resolve far more than enough detail to require low pass filters for APS-C. From a SPATIAL resolution standpoint, the A7r is actually not very high resolution at all...it has 4.9µm pixels, which is 102lp/mm. My Canon 7D has 4.3µm pixels (116lp/mm). Current 24mp APS-C sensors have around 3.8µm pixels (131lp/mm). From a spatial resolution standpoint, there are sensors with FAR greater spatial resolving power than the A7r, and lenses are still resolving enough detail to warrant an AA filter, and will continue to do so for quite some time (even at just MTF50, let alone at lower contrast).
My Photography
Current Gear: Canon 5D III | Canon 7D | Canon EF 600mm f/4 L IS II | EF 100-400mm f/4.5-5.6 L IS | EF 16-35mm f/2.8 L | EF 100mm f/2.8 Macro | 50mm f/1.4
New Gear List: SBIG STT-8300M | Canon EF 300mm f/2.8 L II

chromophore

  • Power Shot G16
  • **
  • Posts: 25
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #42 on: January 11, 2014, 11:18:35 PM »
At the risk of repeating myself, some people need to take a course in signal processing.  If you can define terms like time/frequency domain, Fourier and Z transforms, ringing, low-pass and high-pass filters, or better yet, know how to do the math, then MAYBE I would consider you well-informed enough to talk about aliasing and moire.

No, optical AA filters are not perfect, and yes, there is a very small tradeoff in terms of the ability to resolve spatial frequencies near the Nyquist limit.  But if you think that you can just take away the filter and get superior performance out of software processing techniques, you simply do not have any understanding of the underlying mathematics or the physical principles.  And yes, I *DID* learn the relevant subject material--my background is in pure mathematics with an emphasis on statistics.  So don't think I can't back up what I am saying.

Once the data is quantized by the sensor, you cannot possibly hope to do better than an optical AA filter to fix moire after the fact.  The information to do that is simply NOT THERE because the sensor isn't physically capable of recording it.  The simplest analogy I can give you is the use of antialiasing for on-screen displays of text.  If you turn off AA, the font outlines are "sharper," but they are also more pixelated.  I don't hear any of you complaining that the way modern computers display fonts looks "soft" and "worse" than if they were aliased.  The antialiasing step happens when the vector outline is rasterized.  Doing moire reduction in post is tantamount to taking an aliased image of text and applying a blur algorithm to it--this by definition cannot produce a superior result compared to antialiasing the text as it is rasterized, no matter how sophisticated the algorithm.

The only thing that antialiasing in post has going for it is that you can apply it selectively to parts of the image.  But as I have already implied with the text example, just because the data you received APPEARS "sharper," does not mean the data is a more faithful representation of reality.  Digital image capture is by definition a quantization process.  The artifacts of that quantization include a loss of fidelity, but in some respects it is counterintuitive in the sense that that loss doesn't always appear as random noise or blur.  Losses can manifest as spurious resolution--this is true even for purely analog systems like a pure optical system of a lens!  (One word:  diffraction.)  They can also manifest as spurious frequencies--i.e., moire.  Reducing it before it gets to the sensor is a minor penalty that gets even more minor as sensor density increases.  We all want to believe that we can extract something "pure" out of every last pixel in the sensor, but you've forgotten that the very fact that the pixel arrangement is regular and quantized means you've already lost a LOT of information, and failing to compensate for that regularity actually introduces artifacts in the signal--artifacts that may *seem* like sharpness, but is not there in real life.

canon rumors FORUM

Re: Will Canon ditch the AA Filter?
« Reply #42 on: January 11, 2014, 11:18:35 PM »

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4723
  • POTATO
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #43 on: January 11, 2014, 11:40:08 PM »
The simplest analogy I can give you is the use of antialiasing for on-screen displays of text.  If you turn off AA, the font outlines are "sharper," but they are also more pixelated.  I don't hear any of you complaining that the way modern computers display fonts looks "soft" and "worse" than if they were aliased.  The antialiasing step happens when the vector outline is rasterized.  Doing moire reduction in post is tantamount to taking an aliased image of text and applying a blur algorithm to it--this by definition cannot produce a superior result compared to antialiasing the text as it is rasterized, no matter how sophisticated the algorithm.

Great analogy with the AA fonts!
My Photography
Current Gear: Canon 5D III | Canon 7D | Canon EF 600mm f/4 L IS II | EF 100-400mm f/4.5-5.6 L IS | EF 16-35mm f/2.8 L | EF 100mm f/2.8 Macro | 50mm f/1.4
New Gear List: SBIG STT-8300M | Canon EF 300mm f/2.8 L II

alphajim

  • Power Shot G16
  • **
  • Posts: 10
  • 7D and 5DmkII
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #44 on: January 12, 2014, 12:58:38 AM »
I realize this is the CANONrumors forum, but we seem to have some people engaged here who understand AA at a fundamental level and I'd like to pose something to them.

Fujifilm's X-Trans sensor (APS-C sized, 16mp) doesn't use an AA filter.  They claim by using a different Bayer filter layout, they don't need an AA filter.

http://www.fujifilmusa.com/products/digital_cameras/x/fujifilm_x_e2/features/index.html

Blowing smoke or do they have a point?

canon rumors FORUM

Re: Will Canon ditch the AA Filter?
« Reply #44 on: January 12, 2014, 12:58:38 AM »