Anti-aliasing filter in 1DX
- Thread starter mrmarks
- Start date
- Status
Yikes! That's no good! I find the AA filter of the mkiii stronger then the mkii to the point it can impact IQ. Am I imagine things or seeing things that are not there?
I wish there was a way to remove some of the impact of a AA filter in post processing?
I wish there was a way to remove some of the impact of a AA filter in post processing?
Upvote
0
Hope this isn't taking things too OT, I'm familiar with anti-aliasing in general and the use of low-pass filters when it comes to sampling traditional analog sources (such as audio) but I'm having trouble wrapping my head around how it works in optical systems.
You have different wavelengths of light, which presumably aren't low-pass filtered because you'd lose colour information, so the only other thing I can think of is that you'd be filtering is more like a 2D domain to average out the effect of sharp edges etc, but I thought that would be fairly trivial to do in the digital domain.
Anyway I'm sure there's a good reason, I just need someone to enlighten me
.
You have different wavelengths of light, which presumably aren't low-pass filtered because you'd lose colour information, so the only other thing I can think of is that you'd be filtering is more like a 2D domain to average out the effect of sharp edges etc, but I thought that would be fairly trivial to do in the digital domain.
Anyway I'm sure there's a good reason, I just need someone to enlighten me
.
Upvote
0
L
Loswr
Guest
In an optical system, the AA filter basically blurs the image slightly by passing the light through two layers of a birefringent material (lithium niobate). A birefringent material separates incoming light into two rays, essentially polarizing it. The two layers are oriented orthogonally to one another. There's a 1/4-wave plate in between the two layers to circularly polarize the light after it passes through the first one (else it would be blocked by the 90° polarizers). The thickness of the layers determines the degree of separation of the light rays, and that thickness is varied according to the pixel density of the sensor, to match the Nyquist frequency of that sensor.
Upvote
0
A well-tuned low pass filter does not really lose any detail, whereas a camera without a low pass filter will generate false detail. A filterless camera is destructive to detail.
Upvote
0
I had thought that Canon had put on a thinner, not thicker, AA (low-pass) filter in the 5D III. I remember it being explicitly called out in this blog:
http://blog.jeffascough.com/photographers/2012/03/canon-eos-5d-mark-iii-review.html
As I understand it, Jeff Ascough is pretty tight with Canon, and would likely have fairly in-depth knowledge about the 5D III, so I'm not sure his claim about a thinner AA filter can simply be written off as hearsay or some such.
http://blog.jeffascough.com/photographers/2012/03/canon-eos-5d-mark-iii-review.html
As I understand it, Jeff Ascough is pretty tight with Canon, and would likely have fairly in-depth knowledge about the 5D III, so I'm not sure his claim about a thinner AA filter can simply be written off as hearsay or some such.
Upvote
0
Stephen Melvin said:
Now I'm confused
What would be the advantage of having a different version of the D800 without a low pass/AA filter then? I somehow gained the impression that it was for e.g. landscape photographers who want more detail in their images. If absence of a filter generates false detail/is destructive to detail, what would be the advantage of leaving the filter off?
Upvote
0
Thanks for all the background info provided here. A few more questions:
Is this problem less pronounced in CCD sensors? As far as I know the Leica M9 doesn't have such a filter and I still think that it's one of the cameras that delivers the best results of them all (I don't have one and there may be other factors at play including the lenses).
Also, wouldn't it be possible to leave out the filter and deal with any moire effects in post? Or would it be possible to arrange the photosites more randomly? I'd like to see something that was more along the lines of film grain anyway...
Is this problem less pronounced in CCD sensors? As far as I know the Leica M9 doesn't have such a filter and I still think that it's one of the cameras that delivers the best results of them all (I don't have one and there may be other factors at play including the lenses).
Also, wouldn't it be possible to leave out the filter and deal with any moire effects in post? Or would it be possible to arrange the photosites more randomly? I'd like to see something that was more along the lines of film grain anyway...
Upvote
0
ereka said:Stephen Melvin said:
Now I'm confused
What would be the advantage of having a different version of the D800 without a low pass/AA filter then? I somehow gained the impression that it was for e.g. landscape photographers who want more detail in their images. If absence of a filter generates false detail/is destructive to detail, what would be the advantage of leaving the filter off?
Low-pass filtration is a bit of a controversial topic. It boils down to spatial frequencies (wavelets in two dimensions), and how they are recorded. When you have spatial frequencies smaller than the nyquist frequency of the sensor (that really just means three rows of pixels are necessary to resolve one "cycle" of spatial frequencies, a ratio of 1.5 pixel rows to line pairs), the effect those frequencies have on image detail is largely indeterministic. The general result is that at very fine levels of detail, noise and aliasing increases. The intent of adding a low-pass filter is to blur spatial frequencies that are smaller than the size of a pixel, while simultaneously leaving spatial frequencies as large or larger than a pixel unaffected. Designing a perfect low-pass filter can be difficult. You don't want to design it too strong such that it blurs useful frequencies, however you don't want to design it too weak such that it doesn't blur enough non-useful frequencies that you still get noise and aliasing.
When it comes to landscape photography, the random nature of detail mitigates the impact these "useless" spatial frequencies impose when they are recorded by the sensor, which is why its not really a big deal to eliminate the low-pass filter for such photography. However when you have fine detail in repeated patterns...such as with fabric, the impact those "useless" spatial frequencies impose tends to result in semi-random coloration...color moire and concentric color banding or color sparkle. That is as much a consequence of using bayer-type sensors as it is due to the removal of the low-pass filter, though. With a Foveon-type or monochrome sensor, the effects of not having a low-pass filter are FAR less. The original Foveon sensors did not actually use a low-pass filter, if memory serves, and it was only in the more recent generations were low-pass filters added.
Upvote
0
ereka said:Now I'm confused
What would be the advantage of having a different version of the D800 without a low pass/AA filter then? I somehow gained the impression that it was for e.g. landscape photographers who want more detail in their images. If absence of a filter generates false detail/is destructive to detail, what would be the advantage of leaving the filter off?
slight improvement in sharpness but run into the effects of moire. Nikon site has a blurb on it.
http://imaging.nikon.com/lineup/dslr/d800/features01.htm
also had a good comparison of D800 vs D800E and the moire effect in things like fabrics vs landscape
http://en.nikon.ca/en_CA/o/CBTZ8ZQ3AzZ4Pm9E2GAU8EKdTa4/PDF/Moire_and_False_Colour.pdf
New Fuji X-1 pro (?) is using non-Bayer type sensor to supposedly eliminate moire without using an AA filter. "inspired by film."
www.fujifilm.ca/products/digital_cameras/x/fujifilm_x_pro1/features/page_03.html
.. which might be an interesting alternative to non AA-filtered Bayer sensors or Foveon.
Upvote
0
7enderbender said:
A lot of that is psychological...there is a hard-core subculture of Leica fans, similar to the hard-core subculture of say Apple fans. Apple stuff is nice, but it is not necessarily "the best", and a significant factor of its success is the psychological impact of Apple's advertising and presentation (they really do excel at making people want their products, regardless what the product is.) There is also plenty of information on the net about professional photographers who have complained about moire and aliasing with their incredibly expensive Leica cameras. A couple google searches should bring all that to the surface. Don't get me wrong, Leica cameras are certainly excellent...however they are by no means perfect, there are issues with the lack of an AA filter, and the gap between Leica and regular old DSLR's has certainly shrunk with the advent of the D800 (and even the 5D III, which is still a very high resolution camera.)
7enderbender said:
You can always try to fix things in post, but no amount of post-process work can ever really entirely restore what you may have lost initially. Lightroom 4's new Moire removal tool is pretty nice, but it is not perfect, and can only do so much.
As for arranging photosites randomly, certainly! Thats essentially how the retina's of our eyes work, with an entirely random distribution of luminance and chrominance sensing structures. Trying to manufacture something similar, however, would be much harder than manufacturing something with a simple, repeated pattern. The design of such a CMOS decide would be incredibly complex, and one would have to wonder exactly how it would be done at all (How would you handle "row" and "column" activate? Could you even do row and column activation? How do you parallelize readout if you can't? How would random distribution affect current flow and electronic noise? What is the smallest density you could achieve? etc.) There may be some ways to achieve such a thing while still using a grid-based pixel layout, called Digital Film System. There are plenty of challenged with that approach as well.
Randomizing distribution should certainly resolve the issue though, as moire is the interference pattern generated when similar repeating patterns interact in a non-ideal way. Eliminate one of the repeating patterns, and the effect should disappear.
Upvote
0
- Status
