fredmiranda.com Login
This topic is in the 2015 archive. You can read it and like its posts here. It is read only.

Archive 2015Moire? 5Dsr.

  
 
melcat
Offline
Details
p.4 #1Moire? 5Dsr.
I just want to point out that there are other forms of aliasing than moiré. It is a common misconception that the Bayer filter somehow causes aliasing. It would still be there if the Bayer filter wasn't - the Bayer filter just causes aliasing to manifest as moiré a lot of the time. What causes it is the array of sensels ("pixels").

BTW, I use only Process 2010 in ACR, so there is no moiré removal in the adjustment brush (not that I've ever used the adjustment brush). I have not, and won't, buy a camera without an antialiasing filter.
Jun 26, 2015 at 03:53 AM
EB-1
Offline
Details
p.4 #2Moire? 5Dsr.
melcat wrote:
I have not, and won't, buy a camera without an antialiasing filter.


You may need to stock up soon if the trend of non-AA-filter bodies are any indication of the future.

EBH
Jun 26, 2015 at 06:12 AM
AJSJones
Offline
Details
p.4 #3Moire? 5Dsr.
Jeff Fenske wrote:
I also agree that TDP is a *fantastic* resource. But it could be even more fantastic if he allowed comments, which could have refuted the Cambridge in Colour myth that TDP spreads, that smaller pixels cause diffraction sooner.

The diffraction we see is from the lens.


The myth being spread is that either of those sites actually says anything like "smaller pixels cause diffraction sooner". At least I couldn't find anything resembling that sentence. The discussions are careful to point out, as others have here, that with smaller and smaller pixels, the ability to detect the effect of (the always present) diffraction increase and the "limit" (however it might be defined or assessed) moves to bigger and bigger apertures. Such is the internet and the result of either casual or less critical readers not understanding what is written. The phenomenon has been around as long as (tiny pixel) compact cameras have been compared with (large pixel) DSLRs and why the former are typically physically limited to f/8. Norm Koren specifically discussesdiffraction and digital camera pixel size here and gives a (subjectively cast, based on "quality") handy rule of thumb (best f/stop is one stop wider than 3.2*pixel size)

Norm Koren wroteThe highest spatial frequency a sensor can resolve is its Nyquist frequency, equal to 0.5/(pixel spacing). When a lens is stopped down so its Rayleigh limit is below the Nyquist frequency, the camera is limited by the lens rather than the sensor. For optimum quality (when extreme depth of field is not required), the aperture should be set at least one stop larger than the aperture where the Rayleigh limit equals the Nyquist frequency: NR=N = 3.2 * pixel spacing (µm). For a pixel spacing of 3.4 microns, typical of compact digital cameras, NR=N = 3.2 * 3.4...Show more →
He also gives a table showing the narrower aperture available as pixel size increases, so the TDP and CiC sites are in agreement with Koren. The TDP and CiC tables seem to describe the aperture value at which the effect might detectably decrease the available resolution and both are pretty clear on the consequences of that.

TDP wrote:
Diffraction at the DLA is only barely visible when viewed at full-size (100%, 1 pixel = 1 pixel) on a monitor or when output to a very large print. As sensor pixel density increases, the narrowest aperture we can use to get perfectly pixel-sharp images gets wider.


CiC wroteDiffraction thus sets a fundamental resolution limit that is independent of the number of megapixels, or the size of the film format. It depends only on the f-number of your lens, and on the wavelength of light being imaged. One can think of it as the smallest theoretical "pixel" of detail in photography. Furthermore, the onset of diffraction is gradual; prior to limiting resolution, it can still reduce small-scale contrast by causing airy disks to partially overlap.

..
One can also come at this from another metric: at a given aperture, how many megapixels' worth of infomration can a FF camera capture? Table 3 of this article from 2011 puts the number at 29 MP for f/8 and 60 MP for f/5.6. Of course, the details of their calculations differ a little from everyone else's but the concept remains the same.

If people are striving for the finest detail/maximum shaprness in the captured image (and the many discussions on lens sharpness alone support that), they are well advised to do their own testing to find what their limit is for sensor to print and aperture limitations for "acceptable sharpness". They are also well advised to repeat that if they move from a sensor with large pixels to one with small pixels because of the (same) diffraction effect. For web images versus huge prints, those tests are likely to have quite different "limits".

These sites all agree that diffraction at any given aperture is a constant and that as pixel size goes down, the maximum aperture diameter for optimal detail capture needs to increase. The disagreement comes only in the numerical value of the limit, not in the concept behind it (they agree that this pin exists and there are indeed angels on it, but less so on the number of angels ).
Jun 26, 2015 at 09:47 AM
jcolwell
Online
Details
p.4 #4Moire? 5Dsr.
AJSJones wrote:
(they agree that this pin exists and there are indeed angels on it, but less so on the number of angels ).


Reminds me of something...

"Rabbit and Owl count megapixels", http://tinyurl.com/cyt5uy

Jun 26, 2015 at 12:54 PM
gdanmitchell
Offline
Details
p.4 #5Moire? 5Dsr.
The "myth being spread" is most often (though not exclusively) spread by those who misinterpret or misunderstand the term "diffraction limited aperture" to mean something negative about higher MP sensor cameras and who misconstrue technically correct writing about the subject.

The misinterpretations by those who do not understand all too often lead to false conclusions such as:

1. Higher MP sensor cameras will be softer at a given aperture. (They will not.)

2. Lower MP cameras will produce sharper images at smaller apertures. (They will not.)

3. One should be careful to shoot at the diffraction limited aperture in order to get sharp photographs. (Not necessarily.)

Frankly, although I understand what it really means, the words "diffraction-limited" have been the cause of many of the incorrect interpretations. No one likes being "limited." Sometimes I wish that it were "diffraction matched" or something else. ;-)

Take care,

Dan

AJSJones wrote:
Jeff Fenske wroteI also agree that TDP is a *fantastic* resource. But it could be even more fantastic if he allowed comments, which could have refuted the Cambridge in Colour myth that TDP spreads, that smaller pixels cause diffraction sooner.

The diffraction we see is from the lens.
The myth being spread is that either of those sites actually says anything like "smaller pixels cause diffraction sooner". At least I couldn't find anything resembling that sentence. The discussions are careful to point out, as others have here, that with smaller and smaller pixels, the ability to detect the effect of (the always present) diffraction increase and the "limit" (however it might be defined or assessed) moves to bigger and bigger apertures. Such is the internet and the result of either casual or less critical readers not understanding what is written. The phenomenon has been around as long as
...Show more →
Norm Koren wroteThe highest spatial frequency a sensor can resolve is its Nyquist frequency, equal to 0.5/(pixel spacing). When a lens is stopped down so its Rayleigh limit is below the Nyquist frequency, the camera is limited by the lens rather than the sensor. For optimum quality (when extreme depth of field is not required), the aperture should be set at least one stop larger than the aperture where the Rayleigh limit equals the Nyquist frequency: NR=N = 3.2 * pixel spacing (µm). For a pixel spacing of 3.4 microns, typical of compact digital cameras, NR=N = 3.2 * 3.4 = f/11, so the aperture should be set at f/8 or larger. The corresponding aperture range for digital SLRs, which have pixel spacings between about 6.8 and 9 microns, is f/16-f/22.
Earlier quoteTDP wrote:
Diffraction at the DLA is only barely visible when viewed at full-size (100%, 1 pixel = 1 pixel) on a monitor or when output to a very large print. As sensor pixel density increases, the narrowest aperture we can use to get perfectly pixel-sharp images gets wider.
CiC wroteDiffraction thus sets a fundamental resolution limit that is independent of the number of megapixels, or the size of the film format. It depends only on the f-number of your lens, and on the wavelength of light being imaged. One can think of it as the smallest theoretical "pixel" of detail in photography. Furthermore, the onset of diffraction is gradual; prior to limiting resolution, it can still reduce small-scale contrast by causing airy disks to partially overlap.

Jun 26, 2015 at 01:51 PM
Jeff Fenske
Offline
Details
p.4 #6Moire? 5Dsr.
AJSJones wrote:
Jeff Fenske wroteI also agree that TDP is a *fantastic* resource. But it could be even more fantastic if he allowed comments, which could have refuted the Cambridge in Colour myth that TDP spreads, that smaller pixels cause diffraction sooner.

The diffraction we see is from the lens.
The myth being spread is that either of those sites actually says anything like "smaller pixels cause diffraction sooner". At least I couldn't find anything resembling that sentence. The discussions are careful to point out, as others have here, that with smaller and smaller pixels, the ability to detect the effect of (the always present) diffraction increase and the "limit" (however it might be defined or assessed) moves to bigger and bigger apertures. Such is the internet and the result of either casual or less critical readers not understanding what is written. The phenomenon has been around as long as
...Show more →
Norm Koren wroteThe highest spatial frequency a sensor can resolve is its Nyquist frequency, equal to 0.5/(pixel spacing). When a lens is stopped down so its Rayleigh limit is below the Nyquist frequency, the camera is limited by the lens rather than the sensor. For optimum quality (when extreme depth of field is not required), the aperture should be set at least one stop larger than the aperture where the Rayleigh limit equals the Nyquist frequency: NR=N = 3.2 * pixel spacing (µm). For a pixel spacing of 3.4 microns, typical of compact digital cameras, NR=N = 3.2 * 3.4 = f/11, so the aperture should be set at f/8 or larger. The corresponding aperture range for digital SLRs, which have pixel spacings between about 6.8 and 9 microns, is f/16-f/22.
Earlier quoteTDP wrote:
Diffraction at the DLA is only barely visible when viewed at full-size (100%, 1 pixel = 1 pixel) on a monitor or when output to a very large print. As sensor pixel density increases, the narrowest aperture we can use to get perfectly pixel-sharp images gets wider.
CiC wroteDiffraction thus sets a fundamental resolution limit that is independent of the number of megapixels, or the size of the film format. It depends only on the f-number of your lens, and on the wavelength of light being imaged. One can think of it as the smallest theoretical "pixel" of detail in photography. Furthermore, the onset of diffraction is gradual; prior to limiting resolution, it can still reduce small-scale contrast by causing airy disks to partially overlap.


It does seem that compact cameras go into diffraction sooner, especially at the shortest focal lengths, but not as much when zoomed out. They also often have variable aperture lenses.

Is there another factor going on there, affecting where diffraction sets in, besides pixel size?

You say: "they are well advised to do their own testing."

Here is a guy who did just that, comparing a 12MP full frame sensor with a 36MP full frame sensor, and didn't find the smaller pixels to cause diffraction sooner.

http://www.scantips.com/lights/diffraction.html

I haven't seen anyone show actual, real-world photos that demonstrate the mathematical theory that smaller pixel full frame cameras go into diffraction sooner.

Cambridge in Colour still has the calculator that claims that diffraction occurs sooner with smaller pixels, and they state exactly where "diffraction limiting" takes place.

I wrote this post, showing what Cambridge is still doing; though, they've apparently changed their site quite a bit: http://www.dpreview.com/forums/post/55969318
Jun 26, 2015 at 06:04 PM
RustyBug
Offline
Details
p.4 #7Moire? 5Dsr.
Jeff Fenske wrote:
It does seem that compact cameras go into diffraction sooner, especially at the shortest focal lengths, but not as much when zoomed out.


This is because they have a smaller aperture for the light to pass through (not to be confused with f-stop). The smaller the hole, the more the diffraction being induced by the optical projection. The more we "squeeze" the photons through a tiny hole, the more they bounce off the edges of the hole and go scrambling about.

Kinda like dropping a pebble in a glass of water vs. a calm lake. The rippling effects of the smaller surface area are much more noticeable from the proximity of the wave amplitude interactions.

Diffraction is an optical projection issue ... not a capture issue. Just because you slice it into smaller pieces, sure you slice it small enough and your diffraction will "encompass" your tiny slice, while it may not encompass larger pieces. But, whether you put together a 100 piece jigsaw puzzle or a 10,000 jigsaw puzzle ... the overall picture remains a product of what was projected onto the entire puzzle, not how many slices it gets cut into.
Jul 18, 2015 at 10:36 AM
gdanmitchell
Offline
Details
p.4 #8Moire? 5Dsr.
Jeff Fenske wrote:
You say: "they are well advised to do their own testing."


I strongly agree. To that I would add, beyond considering "perfect aperture," also consider the real world effect of departing from that aperture.

I understand quite well the idea of diffraction-limited aperture and how this plays out against different formats and different sensor resolutions. But I now am being to have enough photography under my belt with the 5Ds R to start to draw some conclusions based on what I see in photographs. To generalize a bit...

- if aperture choice is not significant for other reasons, to ahead and shoot at the so-called diffraction-limited aperture for your lens/camera, unless...

- the "personality" of the lens you are using (lets say it vignettes a bit more that some others at that aperture) may allow you improve things at a slightly different aperture with insignificant negative effect on resolution, unless...

- you need narrower depth of field, in which case you open up the aperture to create the visual effect you need, unless...

- you need larger depth of field, in which case you stop down to whatever aperture you need, understanding that trade-off you are making.

As to stopping down, I was very interested to see what photographs on the "R" would look like at apertures smaller than the so-called diffraction limited aperture. Using high quality L series lenses, a tripod, live view, remote release, and careful technique... f/8 looks excellent with all of my lenses. There is very little difference in resolution at f/11, especially after applying typical post-processing sharpening. At 100% magnification I can see some softening of finest detail at f/16, but I certainly can use this aperture effectively when I nee more depth of field. I generally steer clear of f/22 and smaller unless something other than resolution is much more important — and that can happen in some cases.

Dan
Jul 18, 2015 at 11:56 AM
rbf_
Offline
Details
p.4 #9Moire? 5Dsr.

I wish they had licensed the X-trans CFA from Fuji for use on the R. Their color filter array was specifically designed to reduce moire, allowing them to delete the anti aliasing filter without that penalty. I have no problems with moire on X-trans.
Jul 18, 2015 at 12:15 PM
gdanmitchell
Offline
Details
p.4 #10Moire? 5Dsr.
rbf_ wrote:
I wish they had licensed the X-trans CFA from Fuji for use on the R. Their color filter array was specifically designed to reduce moire, allowing them to delete the anti aliasing filter without that penalty. I have no problems with moire on X-trans.


I have the "R" and I have an X-trans sensor Fujifilm camera. I love the Fujifilm and its image quality, but the "R" doesn't give up anything to the Fujifilm sensors.

Dan
Jul 18, 2015 at 12:23 PM
rbf_
Offline
Details
p.4 #11Moire? 5Dsr.
gdanmitchell wrote:
rbf_ wroteI wish they had licensed the X-trans CFA from Fuji for use on the R. Their color filter array was specifically designed to reduce moire, allowing them to delete the anti aliasing filter without that penalty. I have no problems with moire on X-trans.
I have the "R" and I have an X-trans sensor Fujifilm camera. I love the Fujifilm and its image quality, but the "R" doesn't give up anything to the Fujifilm sensors.

Dan


Didn't mean to imply that it does. I was specifically referring to moire, CFA's and the demosaicing algorithm's that produce moire. As far as I understand, the Fujifilm sensor is actually a Sony sensor with their own CFA(X-trans) and the AA filter removed. The moire is a artifact of the CFA and demosaicing algorithms so 'in theory' any sensor(corrected: any sensor using a CFA) could benefit from an improvement there.

Edited on Jul 18, 2015 at 02:05 PM · View previous versions
Jul 18, 2015 at 12:45 PM
Monito
Offline
Details
p.4 #12Moire? 5Dsr.
Moiré is not caused by the Colour Filter Array (CFA) and demosaicing algorithms. The CFA & demosaicing complicate the moiré and aliasing. But using single sensels to sense all the colours does not eliminate aliasing. It does eliminate a class of false colour aliasing, but aliasing (and hence moiré) including some colour aliasing are still possible because it is inescapable when you use a sampling system without an anti-aliasing filter.
Jul 18, 2015 at 01:00 PM
rbf_
Offline
Details
p.4 #13Moire? 5Dsr.
Monito wrote:
Moiré is not caused by the Colour Filter Array (CFA) and demosaicing algorithms. The CFA & demosaicing complicate the moiré and aliasing. But using single sensels to sense all the colours does not eliminate aliasing. It does eliminate a class of false colour aliasing, but aliasing (and hence moiré) including some colour aliasing are still possible because it is inescapable when you use a sampling system without an anti-aliasing filter.


t most certainly is.
Jul 18, 2015 at 01:07 PM
Monito
Offline
Details
p.4 #14Moire? 5Dsr.
rbf_ wrote:
t most certainly is.


Wrong. You'll have to do better than schoolyard " 'tis so!" rejoinders if you would like to be convincing.

Here for example is moiré from a Foveon sensor with no CFA, and the moiré is obvious:

http://img.photobucket.com/albums/v467/billb6/Photo%20sites/sigma-aliasing-artifact.jpg
Jul 18, 2015 at 01:18 PM
gdanmitchell
Offline
Details
p.4 #15Moire? 5Dsr.
rbf_ wrote:
gdanmitchell wrote
rbf_ wroteI wish they had licensed the X-trans CFA from Fuji for use on the R. Their color filter array was specifically designed to reduce moire, allowing them to delete the anti aliasing filter without that penalty. I have no problems with moire on X-trans.
I have the "R" and I have an X-trans sensor Fujifilm camera. I love the Fujifilm and its image quality, but the "R" doesn't give up anything to the Fujifilm sensors.

Dan
Didn't mean to imply that it does. I was specifically referring to moire, CFA's and the demosaicing algorithm's that produce moire. As far as I understand, the Fujifilm sensor is actually a Sony sensor with their own CFA(X-trans) and the AA filter removed. The moire is a artifact of the CFA and demosaicing algorithms so 'in theory' any sensor could benefit from an improvement there.


I could have been more explicit. I'm not so sure that the real effect of the X-trans sensor is that significant when it comes to aliasing, and there is a compensatory (and probably equally minor) downside related to color separation along boundaries between intense colors.

Based on my uses of the two systems, I don't think that there would be any real overall benefits from using the X-trans type sensor in the Canon cameras — it wouldn't be a bad thing, but I don't think it would be so much of a good thing either.

By the way, and related to some other earlier posts, aliasing is a product of digital sampling. All of the other stuff constitutes ways of trying to deal with the fact of aliasing in systems that record data digitally. Every digital camera that I've ever used was capable of generating, for example, moire. The nature of the "problem" and the situations in which this might happen will vary, but it is always an issue at some level and in some circumstances.

Dan
Jul 18, 2015 at 01:23 PM
Monito
Offline
Details
p.4 #16Moire? 5Dsr.
By the way, rbf, Welcome to FM. (Sorry I had to contradict you on your first post.)
Jul 18, 2015 at 01:46 PM
rbf_
Offline
Details
p.4 #17Moire? 5Dsr.
Monito wrote:
By the way, rbf, Welcome to FM. (Sorry I had to contradict you on your first post.)


Thanks and no problem!
Jul 18, 2015 at 01:49 PM
rbf_
Offline
Details
p.4 #18Moire? 5Dsr.
Monito wrote:
rbf_ wrotet most certainly is.
Wrong. You'll have to do better than schoolyard " 'tis so!" rejoinders if you would like to be convincing.

Here for example is moiré from a Foveon sensor with no CFA, and the moiré is obvious:

http://img.photobucket.com/albums/v467/billb6/Photo%20sites/sigma-aliasing-artifact.jpg


Well I was specifically referring to standard sensor designs using a CFA like Kodak's bayer CFA which have been around for decades. With that said the artifacting in these sensors is produced by the interpolating of the CFA output referred to as demosaicing algorithms. I don't know what the artifacting you are seeing with the Foveon sensor is caused by since it doesn't use the same algorithms due to a different sensor architecture. I have heard it's immune to these specific types of artifacts. But that doesn't mean it's immune to other types of artifacts.

I was really just trying to point out that there's nothing stopping someone like Canon from using a different CFA and algorithms when deleting the AA filter to reduce this specific type of artifact. Like Fuji did.
Jul 18, 2015 at 01:57 PM
Monito
Offline
Details
p.4 #19Moire? 5Dsr.
Monito wrote:
Here for example is moiré from a Foveon sensor with no CFA, and the moiré is obvious:

http://img.photobucket.com/albums/v467/billb6/Photo%20sites/sigma-aliasing-artifact.jpg


rbf_ wrote:
With that said the artifacting in these sensors is produced by the interpolating of the CFA output referred to as demosaicing algorithms.


Nope. Some colour aliasing is attributable to the CFA, especially colour flecks in black hair and green flecks and wavy tinges in blond hair. A different CFA (like Fuji) might reduce that but not eliminate it. A Foveon will not have false colours but it will have moiré.

rbf_ wrote:
I don't know what the artifacting you are seeing


Same aliasing (we can be more specific than "artifacting") you are seeing, in the example.

rbf_ wrote:
with the Foveon sensor is caused by since it doesn't use the same algorithms due to a different sensor architecture. I have heard it's immune to these specific types of artifacts.


It's a myth that Foveon does not alias and the example proves it is a myth.

The Foveon senses RGB in each sensel, so there is no de-mosaicing. Thus there are no false colours. But there is aliasing. If you have a gray fabric at the right range, you will get gray moiré. And similarly, if you have a green fabric at the right range, you will get green moiré. And red with red and blue with blue.

Aliasing occurs when spatial frequencies are overlapped because a digital sampling system (photographic sensor) does not use an anti-aliasing filter (Foveon, Fuji, Nikon D810, Canon 5DsR). An ideal aliasing filter is low-pass and cuts off at 1/2 the sampling frequency (2 pixels on Foveon if it had it) with a sharp drop-off and zero ripple in the passband (Nyquist Sampling Theorem; only a theory in name only, like gravity and evolution are "theories").

In audio, aliasing is very obvious because speech and music depends tones and harmonics that continue over many samples (many wavelengths). Such data is much rarer in images. That is why (regarding it the other way around) the audio sampling rate is 44.1 KHz which is more than twice the highest audio frequency commonly heard (20 KHz). If we sampled audio at 20 KHz, then the upper half of the frequencies (10 - 20 KHz) would reflect around 1/2 the sampling frequency (10 KHz). Then a 12 KHz tone (10 + 2 KHz) would sound like an 8 KHz tone (10 - 2 KHz).

Many examples here: http://en.wikipedia.org/wiki/Moir%C3%A9_pattern and here: http://en.wikipedia.org/wiki/Aliasing

rbf_ wrote:
I was really just trying to point out that there's nothing stopping someone like Canon from using a different CFA and algorithms when deleting the AA filter to reduce this specific type of artifact. Like Fuji did.


Patents.
Jul 18, 2015 at 02:23 PM
rbf_
Offline
Details
p.4 #20Moire? 5Dsr.
Monito wrote:
Monito wroteHere for example is moiré from a Foveon sensor with no CFA, and the moiré is obvious:

http://img.photobucket.com/albums/v467/billb6/Photo%20sites/sigma-aliasing-artifact.jpg
rbf_ wroteWith that said the artifacting in these sensors is produced by the interpolating of the CFA output referred to as demosaicing algorithms.
Nope. Some colour aliasing is attributable to the CFA, especially colour flecks in black hair and green flecks and wavy tinges in blond hair. A different CFA (like Fuji) might reduce that but not eliminate it. A Foveon will not have false colours but it will have moiré.

rbf_ wroteI don't know what the artifacting you are seeing
Same aliasing (we can be more specific than "artifacting") you are seeing, in the example.

rbf_ wrotewith the Foveon sensor is caused by since it doesn't use the same algorithms due to a different sensor architecture. I have heard it's immune to these specific types of artifacts.
It's a myth that Foveon does not alias and the example proves it is a myth.

The Foveon senses RGB in each sensel, so there is no de-mosaicing. Thus there are no false colours. But there is aliasing. If you
...Show more →
rbf_ wroteI was really just trying to point out that there's nothing stopping someone like Canon from using a different CFA and algorithms when deleting the AA filter to reduce this specific type of artifact. Like Fuji did.


Very interesting, didn't really know much about the Foveon sensor. Googling around I have seen some other reports of people being surprised at what apparently is moire in images from Foveon's. I think we are in agreement that a different CFA may reduce but not eliminate these types of artifacts. Either way if I were to purchase a larger sensor, higher resolution camera(like the 5Ds R) without an AA/low-pass filter I would prefer they improve upon the bayer CFA, even if it means licensing some patents. I do really enjoy being able to shoot in infrared with my Fuji though I find that a much more compelling reason than just ~10% more resolution, I guess I just don't love AA/low-pass filters much

Edited on Jul 18, 2015 at 02:55 PM · View previous versions
Jul 18, 2015 at 02:43 PM
FM ForumsCanon ForumJoin Upload & Sell