October 20, 2014, 05:05:07 AM

Author Topic: Will Canon ditch the AA Filter?  (Read 21308 times)

Sporgon

  • 1D X
  • *******
  • Posts: 1977
  • 5% of gear used 95% of the time
    • View Profile
    • www.buildingpanoramics.com
Re: Will Canon ditch the AA Filter?
« Reply #150 on: January 18, 2014, 07:20:19 AM »
I think the (patient) explanations given by jrista have helped me understand why correct sharpening after capturing the data through a good AA filter results in an image that is all but indistinguishable from when there is no filter. There is enough data for the sharpening to work on as you're only taking out part of the frequency with the filter.

As anyone serious is going to do some for of pp it rather begs the question of why not keep the AA filter.

Oh I know; a bit of marketing edge for Nikon.........

 

canon rumors FORUM

Re: Will Canon ditch the AA Filter?
« Reply #150 on: January 18, 2014, 07:20:19 AM »

ajfotofilmagem

  • 1D Mark IV
  • ******
  • Posts: 972
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #151 on: January 18, 2014, 07:37:12 AM »
In Sony cameras with "nigthshot" function, the IR filter can be temporarily displaced from the front of the sensor at the touch of a button to enable night vision. Similar functionality with the AA filter gives the best of both worlds, using AA filter only when necessary. However, a large filter as 24x36mm requires additional space for temporarily be out of the way. A large camera like the Canon 1 series should have enough space for it.
« Last Edit: January 18, 2014, 07:49:42 AM by ajfotofilmagem »

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4460
  • EOL
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #152 on: January 18, 2014, 01:34:26 PM »
In Sony cameras with "nigthshot" function, the IR filter can be temporarily displaced from the front of the sensor at the touch of a button to enable night vision. Similar functionality with the AA filter gives the best of both worlds, using AA filter only when necessary. However, a large filter as 24x36mm requires additional space for temporarily be out of the way. A large camera like the Canon 1 series should have enough space for it.

There are patents now for electrically attenuated AA filters. We wouldn't need to "swing" them out of the way, simply reattenuate to effectively turn off AA filtration. With an electrically attenuated AA filter, you could adjust it strong or weak or off depending on the circumstances. Who knows if/when this technology will find it's way into DSLRs, but it would be the best way to deal with the problem, for sure.

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3917
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #153 on: January 18, 2014, 07:11:53 PM »
Your problem is that your thinking in terms of sky and frond, rather than waveform A and waveform B. If you decide to think of the problem as sky and fronds, then yes, you just have sky and fronds. You have to change your mode of thinking. It isn't sky and fronds interacting to produce...sky and fronds. It is waveform A and waveform B interacting to produce waveform AB' (which, yes, if you "imaged" waveform AB' you would have sky and fronds...but it would be sky and fronds in an entirely different and unique pattern that did not represent A or B distinctly...it would only represent aliases of A sky and fronds and B sky and fronds....does that make sense?)

I'm talking in terms of a general concept. As I said before, it depends on how you think about the problem. You can think of it as real-world objects, or you can think of it as something else...as discrete waveforms that interact. Generally speaking, of course...I am talking about moire in the abstract, not necessarily the specific. I was trying to demonstrate the concept at large.

Strictly speaking though those frond type and lines overlayed over lines moire patterns are not interfering waveforms though so the moire there really is not wave interference even in a general sense and it's something different (although there are direct similarities to what you'd get from a matched set of wave interferences, but it's different and you don't get the fringing and stuff and so on, you could 1:1 match the center point of the black parts and white parts or centers of the overlapped fronds and centers of the 'overlapped' sky parts to the very peaks and very troughs of matched interfering waveforms though).

They are a waveform, when "observed" as an image.You don't need a camera to think of things in different conceptual models...everything you see could be considered as represented in spatial waveform space. Technically speaking, rather than strictly speaking, everything in the universe exists with different representations in different conceptual models, and each one is valid. Anyway, this discussion has gone way off topic, so....

But the moire pattern isn't strictly a waveform interference pattern. I mean just go to the even simpler thick parallel black lines of slightly different spacing on transparencies and overlay them in various ways and look at the moire.  That's not classical waveform interference. The moire can be directly related in ways to waveform interference but a true waveform interference pattern of closest relation isn't the same.

The lines of your transparencies are part of a wave. The lines themselves, say black, are the "trough" of a wavelength, where as the transparency next to it is the crest. Moire occurs because you are overlaying two waveforms. Just because they are lines on a paper does not mean they cannot be modeled as a spatial frequency...a spatial frequency is exactly what they are. Thick black lines separated by thick bands of transparency is a wave of lower frequency, while thin black lines separated by thin bands of transparency are of higher frequency. This image from Norman Koren's site demonstrates:


You keep ignoring the crucial part. if you used plane waves to model the lines on the paper and set them off in the same scenario you'd get a pattern and it would look somewhat similar in many ways but it would not be the same thing as what you get overlaying the two transparencies. In the latter case (or with the fronds and sky or bird feathers) you are not truly carrying out wave interference and in the first case you are. You get not of the fringing and intermediate values and super min and super max values with the transparencies. You can directly relate important aspects of the moire you get when you overlay the two transparencies to what you get from the plane waves though.

If you made the lines approach infinite thinness the moire would become less and less.

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3917
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #154 on: January 18, 2014, 07:22:17 PM »

Same thing goes for the feathers of a bird. Each barb is a wave crest, and the space between them is a trough. Overlay multiple feathers on top of each other, and you have a complex moire pattern.


And this very often is responsible for the various colors we perceive the feathers to be...but sometimes also the feathers have pigment...and sometimes it's both...correct?

Not unless you mean the random colored dots from false color moire when you process an image taken with a camera of a certain type, processed a certain way during de-mosaic.

I think you are thinking about how some color in birds comes from pigments and in other cases from the micro-physical structure and how it refracts, diffracts,interferes. Some of the structure colors are iridescent form, getting waveform interference as some light reflects of one layer below another, and others are more constant view angle form.

Most green birds are green because the feathers have yellow pigment as well as micro-structure that produces blue.

I think many blues in nature come from structure rather than pigment, at least when it comes to animals, by internally scattering the light in some cases.

I think they've been discovering all sorts of quite remarkable micro-structures beyond simple grating or scattering types.
« Last Edit: January 18, 2014, 07:32:06 PM by LetTheRightLensIn »

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3917
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #155 on: January 18, 2014, 07:37:33 PM »
Removal of an AA filter is far from a high end feature. It is a gimmick for all but a very few niche photographer types who's work primarily involves photographing things with entirely random data that could not produce much aliasing regardless. For the vast majority of photographers, use of an AA filter is quite essential to producing BETTER image quality. Aliasing produces nonsense, noise, useless detail. ANTI-Aliasing restores that useless nonsense noise to a more accurate form.

On this I 100% agree with you though.

One side note, the Canon 7D isn't maybe the best body to use (I mean not in the equipment sense and taking pictures but in the testbed to compare detail vs other cameras in a lab sort of sense), since it has those heavily split greens in the CFA array so the de-mosaic routines have to do very tricky things which tend to leave a bit of residual loss of micro-contrast behind (it's actually surprising that they manage to not leave behind major resolution loss, they must be doing some pretty sneaky stuff to handle the split greens and preventing mazing artifacts while not hitting the resolution but a trace).


jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4460
  • EOL
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #156 on: January 20, 2014, 12:18:09 AM »
Removal of an AA filter is far from a high end feature. It is a gimmick for all but a very few niche photographer types who's work primarily involves photographing things with entirely random data that could not produce much aliasing regardless. For the vast majority of photographers, use of an AA filter is quite essential to producing BETTER image quality. Aliasing produces nonsense, noise, useless detail. ANTI-Aliasing restores that useless nonsense noise to a more accurate form.

On this I 100% agree with you though.

One side note, the Canon 7D isn't maybe the best body to use (I mean not in the equipment sense and taking pictures but in the testbed to compare detail vs other cameras in a lab sort of sense), since it has those heavily split greens in the CFA array so the de-mosaic routines have to do very tricky things which tend to leave a bit of residual loss of micro-contrast behind (it's actually surprising that they manage to not leave behind major resolution loss, they must be doing some pretty sneaky stuff to handle the split greens and preventing mazing artifacts while not hitting the resolution but a trace).

What do you mean by "heavily split greens"? The 7D has a pretty standard bayer sensor as far as I know...they shouldn't need to do any special processing (and ACR/LR seem to handle 7D files just fine with their AHDD algorithm.)

canon rumors FORUM

Re: Will Canon ditch the AA Filter?
« Reply #156 on: January 20, 2014, 12:18:09 AM »

LetTheRightLensIn

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3917
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #157 on: January 20, 2014, 12:37:13 AM »
Removal of an AA filter is far from a high end feature. It is a gimmick for all but a very few niche photographer types who's work primarily involves photographing things with entirely random data that could not produce much aliasing regardless. For the vast majority of photographers, use of an AA filter is quite essential to producing BETTER image quality. Aliasing produces nonsense, noise, useless detail. ANTI-Aliasing restores that useless nonsense noise to a more accurate form.

On this I 100% agree with you though.

One side note, the Canon 7D isn't maybe the best body to use (I mean not in the equipment sense and taking pictures but in the testbed to compare detail vs other cameras in a lab sort of sense), since it has those heavily split greens in the CFA array so the de-mosaic routines have to do very tricky things which tend to leave a bit of residual loss of micro-contrast behind (it's actually surprising that they manage to not leave behind major resolution loss, they must be doing some pretty sneaky stuff to handle the split greens and preventing mazing artifacts while not hitting the resolution but a trace).

What do you mean by "heavily split greens"? The 7D has a pretty standard bayer sensor as far as I know...they shouldn't need to do any special processing (and ACR/LR seem to handle 7D files just fine with their AHDD algorithm.)

You know how they use two greens for each red and blue so you have a G1 R G2 B well with the 7D they decided to make G1 very much not the same as G2. It seemed like they wanted to sneak just a little bit more light to the sensor by making one the greens even yet more color blind.

If you developed 7D files right when it first came out with ACR (or even DPP!) you'd notice also sorts of maze patterns appearing. It was especially bad, if I recall correctly, in orangey-yellow blocks of color. Some of were like what the heck is with the artifacts in these 7D images and then we were noticing weird things where measuring the SNR of the G1 channels in the RAW data always gave different results than measuring on the G2 channels.

A few of us brought the complaints to the converter makers and Canon's attention and some even returned their initial 7D copies thinking that maybe that had the Bayer array somehow misaligned or something. Then a few weeks later Adobe released a new ACR and I think shortly after that a new DPP came out. The early speculation would be that getting around the split greens would cause major loss of resolution or noticeably remaining artifacts, but somehow the converter makers found a way to pretty much solve all the mazing artifacts while only barely hitting the resolution at all (if you still had the early ACR beta that supported the 7D and process a file with it, you can get a touch more micro-contrast out of it with 7D files than using the later fixed beta and final releases). It's hard to say but it seemed like the fix maybe effectively knocked 1-2MP MP off the 18MP sensor, not really a big deal, perhaps it made the files a bit more filmlike looking.

And you can see references by Adobe to split green parameters in ACR. It's not just the 7D that needs them but a few other cameras as well from some of the smaller players in the digital camera world. I think Canon is not splitting the greens so much again and I;m not sure if Nikon ever did (i'm pretty unsure about this last stuff though).

If you didn't buy a 7D within the first 2-3 weeks of their very first arrival in the U.S. you probably missed the whole mazing thing with ACR (and maybe 3-4 weeks for DPP and the others).
« Last Edit: January 20, 2014, 12:39:23 AM by LetTheRightLensIn »

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4460
  • EOL
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #158 on: January 20, 2014, 01:38:44 AM »
Removal of an AA filter is far from a high end feature. It is a gimmick for all but a very few niche photographer types who's work primarily involves photographing things with entirely random data that could not produce much aliasing regardless. For the vast majority of photographers, use of an AA filter is quite essential to producing BETTER image quality. Aliasing produces nonsense, noise, useless detail. ANTI-Aliasing restores that useless nonsense noise to a more accurate form.

On this I 100% agree with you though.

One side note, the Canon 7D isn't maybe the best body to use (I mean not in the equipment sense and taking pictures but in the testbed to compare detail vs other cameras in a lab sort of sense), since it has those heavily split greens in the CFA array so the de-mosaic routines have to do very tricky things which tend to leave a bit of residual loss of micro-contrast behind (it's actually surprising that they manage to not leave behind major resolution loss, they must be doing some pretty sneaky stuff to handle the split greens and preventing mazing artifacts while not hitting the resolution but a trace).

What do you mean by "heavily split greens"? The 7D has a pretty standard bayer sensor as far as I know...they shouldn't need to do any special processing (and ACR/LR seem to handle 7D files just fine with their AHDD algorithm.)

You know how they use two greens for each red and blue so you have a G1 R G2 B well with the 7D they decided to make G1 very much not the same as G2. It seemed like they wanted to sneak just a little bit more light to the sensor by making one the greens even yet more color blind.

If you developed 7D files right when it first came out with ACR (or even DPP!) you'd notice also sorts of maze patterns appearing. It was especially bad, if I recall correctly, in orangey-yellow blocks of color. Some of were like what the heck is with the artifacts in these 7D images and then we were noticing weird things where measuring the SNR of the G1 channels in the RAW data always gave different results than measuring on the G2 channels.

A few of us brought the complaints to the converter makers and Canon's attention and some even returned their initial 7D copies thinking that maybe that had the Bayer array somehow misaligned or something. Then a few weeks later Adobe released a new ACR and I think shortly after that a new DPP came out. The early speculation would be that getting around the split greens would cause major loss of resolution or noticeably remaining artifacts, but somehow the converter makers found a way to pretty much solve all the mazing artifacts while only barely hitting the resolution at all (if you still had the early ACR beta that supported the 7D and process a file with it, you can get a touch more micro-contrast out of it with 7D files than using the later fixed beta and final releases). It's hard to say but it seemed like the fix maybe effectively knocked 1-2MP MP off the 18MP sensor, not really a big deal, perhaps it made the files a bit more filmlike looking.

And you can see references by Adobe to split green parameters in ACR. It's not just the 7D that needs them but a few other cameras as well from some of the smaller players in the digital camera world. I think Canon is not splitting the greens so much again and I;m not sure if Nikon ever did (i'm pretty unsure about this last stuff though).

If you didn't buy a 7D within the first 2-3 weeks of their very first arrival in the U.S. you probably missed the whole mazing thing with ACR (and maybe 3-4 weeks for DPP and the others).

Hmm. I can't imagine that such a thing is a huge problem. It's not all that different from Sony's "Emerald Green" CFA that they introduced many years ago (they called it RGBE). Their "Emerald" had more blue in it than the standard green. Based on all the sample images at the time, it actually produced better color accuracy...but it would be the exact same thing as your describing with the 7D.

There have been similar approaches in the past by other companies as well. Some simply do away with the second green and make it a "white". Fuji threw in an extra tiny little white pixel between all the primaries that were very widely spaced, but gathered extra luminance data. If what you say is correct, that would have caused even more problems for demosaicing, however it improved resolution a bit, and improved DR (although the DR improvement seemed minor, especially compared to what Sony did with Exmor.) Kodak, Sony, and Fuji all now have RGBW sensor designs, and Sony is even starting to patent non-square pixels (they have patents for triangular and hexagonal pixels now, and supposedly one of these pixel designs is going to be used in their forthcoming 54mp FF sensor.)

I also can't imagine that it would cause a loss in resolution. I mean, the crux of any bayer demosaicing algorithm is interpolating the intersections between every (overlapping) set of 2x2 pixels. Because there is reuse of sensor pixels for multiple output pixels, there is an inherent blur radius. But it is extremely small, and it wouldn't grow or shrink if one of the pixel colors changed. You would still be interpolating the same basic amount of information within the same radius. I remember there being a small improvement in resolution with my 7D between LR 3 and 4, and things seem a bit crisper again moving from LR 4 to 5. I suspect any supposed loss in resolution with the 7D was due to the novelty of Adobe's implementation of support for the 7D, not anything related to having two slightly different colors for the green pixels. The quality with which LR renders my 7D images only seems to get better and better with time and each subsequent version, so as Adobe optimizes their demosaicing implementation, any inherent error is clearly diminishing.

BTW, there is no way anything Adobe has ever done that could possibly "knock off 1-2mp worth of resolution" from the 7D. The most basic demosaicing algorithm will produce noisy, mazed and stair-stepped results. Better demosaicing algorithms factor in more information from a greater area of pixels as necessary (i.e. in order to avoid maze artifacts), however the amount of blur they introduce is fractional...not even close to diminishing resolution by another two megapixels over the bayer design itself. You can clearly see this when comparing DPP to ACR/LR with something like a strand of hair. DPP will produce a fairly jagged result, ACR/LR produce a very clean result. Based on the sample below, ACR is actually sharper and supports even finer detail resolution:



Additionally, the output resolution of my 7D + EF 600/4 L II is WAY, WAY, WAY better 7D + 100-400 @ 400. The difference that a good lens makes with the 7D's resolving power would completely overwhelm any perceived difference that the ACR/LR demosaicing algorithm makes. That's entirely in line with theory as well, as output resolution is approximated by the RMS of the resolutions of each component. With the 600/4 II, the 7D produces exceptionally sharp results, and is able to very finely delineate detail with a good lens that otherwise appears quite soft with the likes of the 100-400 L, 300/4 L, 70-200 L, etc. I mean, just look at the detail resolved on these birds (all 7D, all of which are processed with Lightroom):






Marsu42

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4809
  • ML-66d / 100L / 70-300L / 17-40L / 600rts
    • View Profile
    • 6D positive spec list
Re: Will Canon ditch the AA Filter?
« Reply #159 on: January 20, 2014, 01:46:54 AM »
The quality with which LR renders my 7D images only seems to get better and better with time and each subsequent version, so as Adobe optimizes their demosaicing implementation, any inherent error is clearly diminishing.

Thanks again for all your great posts on this! The problem with Adobe is that they seem to be very secretive about any improvements concerning ACR or LR, their official changlog only reflects a small part of the changes ... or is there any Adobe or 3rd party documentation on their raw converter improvements over time?

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4460
  • EOL
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #160 on: January 20, 2014, 02:18:33 AM »
The quality with which LR renders my 7D images only seems to get better and better with time and each subsequent version, so as Adobe optimizes their demosaicing implementation, any inherent error is clearly diminishing.

Thanks again for all your great posts on this! The problem with Adobe is that they seem to be very secretive about any improvements concerning ACR or LR, their official changlog only reflects a small part of the changes ... or is there any Adobe or 3rd party documentation on their raw converter improvements over time?

There isn't any official information. I wouldn't expect any, either. The minutia of the algorithmic details of Adobe's demosaicing algorithm would bore most people, and would probably be incomprehensible to anyone who didn't have a math or comp. sci. degree. The specifics aren't all that important. All I do know is that LR5, compared to LR3, produces clearer, sharper results...so Adobe certainly seems to be optimizing their algorithms. Optimization over the long term is to be expected. Demosaicing is the heart of what ACR and LR do...it is going to be the single biggest processing cost when rendering RAW images to screen (since every time you change a setting, they have to rerender, and that can be many times per second.)

I also remember a fairly significant jump in overall quality, edge definition, and sharpness between LR3 and 4. In LR3, sharp edges, hairs, bird feather barbs, etc. looked more like DPP (which as you can see clearly has a less effective algorithm, as it intrinsically blurs more, yet still produces stair-stepping.) LR4 produces really crisp, clean, smooth edges, without blurring. They never mentioned any specific changes to ACR that would cause those changes, but they were clearly there. I don't really care that Adobe keeps the specifics under wrap...sometimes it isn't a good idea to expose too many details about one's technology, as it can invite too many questions for customers when a customer decides something they are seeing in their results doesn't jive with what the company is saying about their algorithms.

Marsu42

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4809
  • ML-66d / 100L / 70-300L / 17-40L / 600rts
    • View Profile
    • 6D positive spec list
Re: Will Canon ditch the AA Filter?
« Reply #161 on: January 20, 2014, 08:35:47 AM »
They never mentioned any specific changes to ACR that would cause those changes, but they were clearly there.

As far as I read it, they don't *any* information on this, specific or not - after "camera xyz supported" that's that, they only mention improvemtents in the user front end like more denoising options.

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4460
  • EOL
    • View Profile
    • Nature Photography
Re: Will Canon ditch the AA Filter?
« Reply #162 on: January 20, 2014, 02:04:50 PM »
They never mentioned any specific changes to ACR that would cause those changes, but they were clearly there.

As far as I read it, they don't *any* information on this, specific or not - after "camera xyz supported" that's that, they only mention improvemtents in the user front end like more denoising options.

They do have more details than simply what cameras are supported at times. You might need to dig deeper for that info. A good blog to read is Julieanne Kost's blog. There is some good info there every so often. It's been years since LR4 was released, so it would take some digging to find the useful tidbits about ACR on any Adobe site, blog or not...but there usually are tidbits of information.

canon rumors FORUM

Re: Will Canon ditch the AA Filter?
« Reply #162 on: January 20, 2014, 02:04:50 PM »

CarlTN

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 2227
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #163 on: January 21, 2014, 03:36:19 PM »

Same thing goes for the feathers of a bird. Each barb is a wave crest, and the space between them is a trough. Overlay multiple feathers on top of each other, and you have a complex moire pattern.


And this very often is responsible for the various colors we perceive the feathers to be...but sometimes also the feathers have pigment...and sometimes it's both...correct?

Not unless you mean the random colored dots from false color moire when you process an image taken with a camera of a certain type, processed a certain way during de-mosaic.

I think you are thinking about how some color in birds comes from pigments and in other cases from the micro-physical structure and how it refracts, diffracts,interferes. Some of the structure colors are iridescent form, getting waveform interference as some light reflects of one layer below another, and others are more constant view angle form.

Most green birds are green because the feathers have yellow pigment as well as micro-structure that produces blue.

I think many blues in nature come from structure rather than pigment, at least when it comes to animals, by internally scattering the light in some cases.

I think they've been discovering all sorts of quite remarkable micro-structures beyond simple grating or scattering types.

Interesting...

CarlTN

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 2227
    • View Profile
Re: Will Canon ditch the AA Filter?
« Reply #164 on: January 21, 2014, 03:43:32 PM »







Regarding the demosaicing algorithm in LR...are you referring to the "moire" slider in the lens correction menu, or are you saying LR applies this algorithm automatically to all images?  Or does it only apply to RAW images that are converted to DNG?  Lastly, have you noticed a difference, or an improvement over time, from successive versions of LR?

Sorry if these questions might be beneath your brilliance...but I don't know how else to get an answer other than asking...:P...

canon rumors FORUM

Re: Will Canon ditch the AA Filter?
« Reply #164 on: January 21, 2014, 03:43:32 PM »