Ad
Ad
Ad
Pages: « 1 [2]   Bottom of Page
Print
Author Topic: Anti-aliasing filter effects demonstrated  (Read 34613 times)
Ray
Sr. Member
****
Offline Offline

Posts: 8944


« Reply #20 on: October 09, 2007, 04:19:17 AM »
ReplyReply

Quote
As a speculation, I suggest that cameras like the 1DsIII may be reaching the point where they would function well without an AA filter for all but the best lenses.
[a href=\"index.php?act=findpost&pid=144504\"][{POST_SNAPBACK}][/a]

If the 1Ds3 might be reaching this point, then the Canon 400D and 40D, and a few Nikon models, have already reached the point.

Maybe that's the next break-through in 35mm DSLRs. Sufficient pixel density to completely dispense with AA filters. I guess the problem is, if the 35mm format is to compete with MFDBs, it needs higher resolving lenses and such lenses, at their sharpest apertures, are potentially going to create the aliasing problems, with certain subjects.

The P45 and P45+ have no greater pixel density than the 1Ds3, yet they do not have an AA filter. Why is this? Are those Digitar and Rodenstock lenses not quite what they are cracked up to be?  

Are those very professional and meticulous guys who spend $30,000 on a digital back more tolerant of moire than 35mm DSLR owners?

Beats me!
« Last Edit: October 09, 2007, 04:21:43 AM by Ray » Logged
NikosR
Sr. Member
****
Offline Offline

Posts: 622


WWW
« Reply #21 on: October 09, 2007, 04:25:24 AM »
ReplyReply

Without doubting the existense of benefits (and also cons) of the no-AA filter case, I would think that a more valid real life comparison would involve the use of appropriate sharpening and local contrast enhancement techniques for the images.

One will see then that the differences in terms of local contrast and acuteness  while still existing tend to be much less and practically not relevant for smaller output.
Logged

Nikos
jing q
Sr. Member
****
Offline Offline

Posts: 594


WWW
« Reply #22 on: October 09, 2007, 05:26:04 AM »
ReplyReply

Quote
Without doubting the existense of benefits (and also cons) of the no-AA filter case, I would think that a more valid real life comparison would involve the use of appropriate sharpening and local contrast enhancement techniques for the images.

One will see then that the differences in terms of local contrast and acuteness  while still existing tend to be much less and practically not relevant for smaller output.
[a href=\"index.php?act=findpost&pid=144791\"][{POST_SNAPBACK}][/a]

I think that would be for another thread (and I'm sure tons of people have discussed sharpening and local contrast enhancement)
Logged
nma
Full Member
***
Offline Offline

Posts: 162


« Reply #23 on: October 09, 2007, 07:30:13 AM »
ReplyReply

Quote
If the 1Ds3 might be reaching this point, then the Canon 400D and 40D, and a few Nikon models, have already reached the point.

Maybe that's the next break-through in 35mm DSLRs. Sufficient pixel density to completely dispense with AA filters. I guess the problem is, if the 35mm format is to compete with MFDBs, it needs higher resolving lenses and such lenses, at their sharpest apertures, are potentially going to create the aliasing problems, with certain subjects.

The P45 and P45+ have no greater pixel density than the 1Ds3, yet they do not have an AA filter. Why is this? Are those Digitar and Rodenstock lenses not quite what they are cracked up to be? 

Are those very professional and meticulous guys who spend $30,000 on a digital back more tolerant of moire than 35mm DSLR owners?

Beats me!
[a href=\"index.php?act=findpost&pid=144789\"][{POST_SNAPBACK}][/a]

Ray,

In theory there is a way out of this dilemma, oversample. A sensor of a given area will detect the same number of photons, provided one can neglect the effect of the packing fraction of the photosites. By having a sensor with very high spatial sampling, one can increase the Nyquist frequency and  eliminate aliasing. The price one pays is increasing photon noise level. But one can downsample the image, regaining signal to noise ratio at the expense of resolution.  This has the advantage of being completely in the digital domain; the AA filter effect would come at much higher frequency and thus be of no concern.  There could be many useful tradeoffs with this approach.  By the way, this idea is very similar to the oversampling applied to digital audio recordings: The audio data are sampled at a very high rate while recording, then resampled to 44.1 KH for production of CDs.
Logged
KAP
Full Member
***
Offline Offline

Posts: 168


WWW
« Reply #24 on: October 09, 2007, 08:00:19 AM »
ReplyReply

Quote
I did some reading online about them, and your experience seems to back that up. Moire is fixable with software, and you get to choose whether to use anti-aliasing, instead of it being forced upon you.

If only the maxmax site was navigable, as it's such a jumbled mess I can't find whether they remove AA filters from 30D/40D without any IR funkiness.
[a href=\"index.php?act=findpost&pid=144208\"][{POST_SNAPBACK}][/a]

Croma moire is easily dealt with, Luminance moira stays. I find with my Kodak SLR/n on screen artifacts once dealt with by software have the same softening or worse as shooting with the AA filter cameras. Unless you have time to hand tune every picture for output a AA filter does a good job.
Moire was the reason I had to give up with the Kodak for jobs, there just isn't the time to correct every image.

Kevin.
Logged
NikosR
Sr. Member
****
Offline Offline

Posts: 622


WWW
« Reply #25 on: October 10, 2007, 07:54:03 AM »
ReplyReply

Quote
I think that would be for another thread (and I'm sure tons of people have discussed sharpening and local contrast enhancement)
[a href=\"index.php?act=findpost&pid=144794\"][{POST_SNAPBACK}][/a]

Why would you think my comment is OT? I'm simply suggesting that appropriate sharpening and LCE significantly masks the difference between AA and non-AA images. You can agree or disagree with my comment but surely it is relevant to this thread.
Logged

Nikos
jing q
Sr. Member
****
Offline Offline

Posts: 594


WWW
« Reply #26 on: October 10, 2007, 11:56:07 AM »
ReplyReply

Quote
Why would you think my comment is OT? I'm simply suggesting that appropriate sharpening and LCE significantly masks the difference between AA and non-AA images. You can agree or disagree with my comment but surely it is relevant to this thread.
[a href=\"index.php?act=findpost&pid=145083\"][{POST_SNAPBACK}][/a]

yes but the topic is not really about sharpening in post processing,
which, btw still cannot recreate information that was lost already when softened.

very useful tool but having the initial image in all its non-AA glory is great.

perhaps it would be more useful to take an image from a non-AA filtered camera, an image from an AA filtered camera and seeing more comparisons, and even doing the whole sharpening thing

but honestly the biggest difference comes about when you see an actual print at a larger size
I've learnt that what you see on the computer screen even at 100 percent is not necessarily exactly what you'll get when you do a printout
« Last Edit: October 10, 2007, 12:03:22 PM by jing q » Logged
NikosR
Sr. Member
****
Offline Offline

Posts: 622


WWW
« Reply #27 on: October 10, 2007, 11:43:21 PM »
ReplyReply

Quote
yes but the topic is not really about sharpening in post processing,
which, btw still cannot recreate information that was lost already when softened.

very useful tool but having the initial image in all its non-AA glory is great.

perhaps it would be more useful to take an image from a non-AA filtered camera, an image from an AA filtered camera and seeing more comparisons, and even doing the whole sharpening thing

but honestly the biggest difference comes about when you see an actual print at a larger size
I've learnt that what you see on the computer screen even at 100 percent is not necessarily exactly what you'll get when you do a printout
[a href=\"index.php?act=findpost&pid=145120\"][{POST_SNAPBACK}][/a]

That's exactly my premise. You obviously fail to realize what capture sharpening and LCE exactly do and why they are in fact de-rigeuer when processing AA filtered images.

While these techniques cannot recreate any detail lost (at least theoretically, if the AA filter is perfect and correct for the given sensor there should not be any real detail lost, just artifacts) it can bring back a lot of the apparent loss in local contrast.

Now, it is my belief that the main culprit  for unprocessed AA-filtered images looking so much inferior to non-AA ones (ref. the links in the OP) is the additional (to the loss due to Bayer interpolation) loss in local contrast exactly due to the AA filter and to a much lesser extent any loss in detail due to an unavoidably non-perfect filter.

I hope you now see that it is practically useless to discuss advantages and disadvantages of AA versus non-AA if you don't consider post-processing also.

Of course I would not expect a commercial site like the one referenced to in the OP to display what the difference is after appropriate post-processing because this would weaken their marketing case.
« Last Edit: October 11, 2007, 12:12:26 AM by NikosR » Logged

Nikos
feppe
Sr. Member
****
Offline Offline

Posts: 2909

Oh this shows up in here!


WWW
« Reply #28 on: October 11, 2007, 01:51:08 AM »
ReplyReply

Quote
That's exactly my premise. You obviously fail to realize what capture sharpening and LCE exactly do and why they are in fact de-rigeuer when processing AA filtered images.

While these techniques cannot recreate any detail lost (at least theoretically, if the AA filter is perfect and correct for the given sensor there should not be any real detail lost, just artifacts) it can bring back a lot of the apparent loss in local contrast.

Now, it is my belief that the main culprit  for unprocessed AA-filtered images looking so much inferior to non-AA ones (ref. the links in the OP) is the additional (to the loss due to Bayer interpolation) loss in local contrast exactly due to the AA filter and to a much lesser extent any loss in detail due to an unavoidably non-perfect filter.

I hope you now see that it is practically useless to discuss advantages and disadvantages of AA versus non-AA if you don't consider post-processing also.

Of course I would not expect a commercial site like the one referenced to in the OP to display what the difference is after appropriate post-processing because this would weaken their marketing case.
[a href=\"index.php?act=findpost&pid=145257\"][{POST_SNAPBACK}][/a]

The MaxMax site offers RAWs from the modified and unmodified cameras, so one can apply any applicable post-processing, and see the difference for themselves.
Logged

Ray
Sr. Member
****
Offline Offline

Posts: 8944


« Reply #29 on: October 11, 2007, 02:43:13 AM »
ReplyReply

Quote
Ray,

In theory there is a way out of this dilemma, oversample. A sensor of a given area will detect the same number of photons, provided one can neglect the effect of the packing fraction of the photosites. By having a sensor with very high spatial sampling, one can increase the Nyquist frequency and  eliminate aliasing. The price one pays is increasing photon noise level. But one can downsample the image, regaining signal to noise ratio at the expense of resolution.  This has the advantage of being completely in the digital domain; the AA filter effect would come at much higher frequency and thus be of no concern.  There could be many useful tradeoffs with this approach.  By the way, this idea is very similar to the oversampling applied to digital audio recordings: The audio data are sampled at a very high rate while recording, then resampled to 44.1 KH for production of CDs.
[a href=\"index.php?act=findpost&pid=144803\"][{POST_SNAPBACK}][/a]

Yes, I'm familiar with the concept of oversampling, yet I find it very curious that every time Canon increases the pixel density of its DSLRs, which all include an AA filter without exception, there's great concern on forums like this that no purpose is being served because Canon lenses, in general, are not good enough to justify the higher pixel density.

If that's the case, why the need for an AA filter?

If a top of the range 39mp digital back like the P45+, with the pixel density of the 1Ds3 and less pixel density than the Canon 40D, does not have an AA filter, why do the latest Canon DSLRs need an AA filter?

When Sinar and Rodenstock produced their digital series lenses, there was a lot of excitement at the sheer quality of these lenses. The published MTF curves implied they were sharper than most 35mm lenses. In fact, they seemed to be similar in resolving capability to the renowned Zuiko lenses used in the Olympus 4/3rds format.

How do owners of the P45 manage with such razor sharp lenses that are almost diffraction limited at f5.6, considering the P45 does not have an AA filter?
Logged
NikosR
Sr. Member
****
Offline Offline

Posts: 622


WWW
« Reply #30 on: October 11, 2007, 11:47:49 AM »
ReplyReply

Quote
The MaxMax site offers RAWs from the modified and unmodified cameras, so one can apply any applicable post-processing, and see the difference for themselves.
[a href=\"index.php?act=findpost&pid=145273\"][{POST_SNAPBACK}][/a]

Thank you. I could find only 1 comparison (a 5D shot). My preliminary findings with opening the files in Lightroom (pls. note I'm not an expert in processing 5D files) seem to support my view. Difference is minimised when correctly post-processing. I can't imagine any differences due to the AA to be visible in any reasonably sized print. Of course there are differences in the pictures but that's because they are different shots i.e. exposure and angle variation.

Judging from this single example I wouldn't waste my money in converting a 5D. Results might be different with other images and of course other cameras, not to mention lenses and apertures used.

PS and OT. The airconditioner grills, apart from causing aliasing on the non AA image, cause Lightroom to produce 'glorious' interpolation artifacts (coloured meanders) on both images.
« Last Edit: October 11, 2007, 11:56:25 AM by NikosR » Logged

Nikos
jing q
Sr. Member
****
Offline Offline

Posts: 594


WWW
« Reply #31 on: October 11, 2007, 02:44:44 PM »
ReplyReply

Quote
That's exactly my premise. You obviously fail to realize what capture sharpening and LCE exactly do and why they are in fact de-rigeuer when processing AA filtered images.

While these techniques cannot recreate any detail lost (at least theoretically, if the AA filter is perfect and correct for the given sensor there should not be any real detail lost, just artifacts) it can bring back a lot of the apparent loss in local contrast.

Now, it is my belief that the main culprit  for unprocessed AA-filtered images looking so much inferior to non-AA ones (ref. the links in the OP) is the additional (to the loss due to Bayer interpolation) loss in local contrast exactly due to the AA filter and to a much lesser extent any loss in detail due to an unavoidably non-perfect filter.

I hope you now see that it is practically useless to discuss advantages and disadvantages of AA versus non-AA if you don't consider post-processing also.

Of course I would not expect a commercial site like the one referenced to in the OP to display what the difference is after appropriate post-processing because this would weaken their marketing case.
[a href=\"index.php?act=findpost&pid=145257\"][{POST_SNAPBACK}][/a]

NikosR, of course removing the AA filter is not the best move for everyone
For example, for people who feel that it's ok to settle for a lower quality image at the start and then trying to make it seem like it came from a higher quality camera  

trust me I am very aware of what sharpening does. I use it all the time. With my non-AA filtered digital back now I am even happier that sharpening is kept to a minimum compared to my Canon which always needed sharpening. And this does make a difference in the apparent look of the files (more detail and depth from the start)

After having seen the differences with the AA and non-AA filtered files at large print sizes I have to put my vote behind non-AA, even though it is abit more of a bit ch to deal with the problems that come with it.
I hope you can give a comparable non-AA-ed camera a try and see for yourself the effects so you understand why I would feel this way.
Logged
jing q
Sr. Member
****
Offline Offline

Posts: 594


WWW
« Reply #32 on: October 11, 2007, 02:47:48 PM »
ReplyReply

to summarize, from printing I personally feel that detail is being lost through AA filtered images.
This is very readily apparent in landscape images and gives the feeling of a photo of a photo rather than a gorgeous RA4 film print

I don't think that the cost of removing the AA filter is worthwhile if you don't print very big
I would do it for a 1dsMkII but a 5D may not be worth the effort either...
unless it was cheaper!haha.
« Last Edit: October 11, 2007, 02:53:38 PM by jing q » Logged
Christopher
Sr. Member
****
Offline Offline

Posts: 944


WWW
« Reply #33 on: October 12, 2007, 05:35:55 AM »
ReplyReply

Quote
to summarize, from printing I personally feel that detail is being lost through AA filtered images.
This is very readily apparent in landscape images and gives the feeling of a photo of a photo rather than a gorgeous RA4 film print

I don't think that the cost of removing the AA filter is worthwhile if you don't print very big
I would do it for a 1dsMkII but a 5D may not be worth the effort either...
unless it was cheaper!haha.
[a href=\"index.php?act=findpost&pid=145373\"][{POST_SNAPBACK}][/a]


Sorry but what are we talking about ? I mean this is totally senseless. Would I want to remove my AA filter from a 5D or 1DsMk2 ? certainly never.

Why ? Because the difference isn't worth it and you have a lot more problems than benefits.

Here once again are both pictures. Sharpend. ANY further sharpening on both images would create arifacts. So which one is which ? Which is better... ?

In my own oppinion there is no real difference. I reall prefer the version WITH AA Filter, but some might say the other is nicer.


My point is you can't just say without AA filter deliveres better pictures, because that just is BS....

Here are both samples.
[attachment=3547:attachment]

[attachment=3548:attachment]
Logged

Eric Myrvaagnes
Sr. Member
****
Offline Offline

Posts: 8292



WWW
« Reply #34 on: October 12, 2007, 09:21:56 AM »
ReplyReply

That's reassuring to me, since I have a 5D (with AA filter). I can't tell which picture is which (but the truck at the stop sign is certainly easier to see in one photo than in the other.     )
Logged

-Eric Myrvaagnes

http://myrvaagnes.com  Visit my website. New images each season.
jing q
Sr. Member
****
Offline Offline

Posts: 594


WWW
« Reply #35 on: October 12, 2007, 10:22:17 AM »
ReplyReply

Quote
Sorry but what are we talking about ? I mean this is totally senseless. Would I want to remove my AA filter from a 5D or 1DsMk2 ? certainly never.

Why ? Because the difference isn't worth it and you have a lot more problems than benefits.

Here once again are both pictures. Sharpend. ANY further sharpening on both images would create arifacts. So which one is which ? Which is better... ?

In my own oppinion there is no real difference. I reall prefer the version WITH AA Filter, but some might say the other is nicer.
My point is you can't just say without AA filter deliveres better pictures, because that just is BS....

Here are both samples.
[attachment=3547:attachment]

[attachment=3548:attachment]
[a href=\"index.php?act=findpost&pid=145485\"][{POST_SNAPBACK}][/a]

I like the first one in terms of details. what was the sharpening?
Logged
juicy
Sr. Member
****
Offline Offline

Posts: 254


« Reply #36 on: October 12, 2007, 12:48:42 PM »
ReplyReply

Hi!

The second image seems sharper to me in most areas but it has also more powerfull sharpening halos. Color-aliasing seems to be differently emphasized in different images which leads me to suspect there's more to do with the exact lens-position and focusing than AA-filter (or the lack of it). This is certainly pixel-peeping and kinda difficult when the scenery is not exactly the same (lighting is different). Subtle differences... Interesting topic anyway.

Cheers,
J
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8944


« Reply #37 on: October 13, 2007, 05:18:21 AM »
ReplyReply

I agree there's not much resolution or accutance difference between these two images. I downloaded the RAW images for comparison. It seems the time interval between the two shots is quite significant, judging by the shadows. There are some areas of the hotrod image which seem very marginally sharper, but there's nothing here to shout about. The f stop of f5.6 might not have provided sufficient DoF if the focussing was slightly different between the two shots.

One wonders why this company offering AA filter removal did not provide better samples. Perhaps more significantly, if there really is so little difference, with and without AA filter, why have Canon provided one on the 5D at what must be additional expense, and why are they providing one on the 1Ds3 which would surely have even less need of one?
Logged
Gary Ferguson
Sr. Member
****
Offline Offline

Posts: 540


WWW
« Reply #38 on: October 13, 2007, 01:56:32 PM »
ReplyReply

Quote
Perhaps more significantly, if there really is so little difference, with and without AA filter, why have Canon provided one on the 5D at what must be additional expense, and why are they providing one on the 1Ds3 which would surely have even less need of one?
[a href=\"index.php?act=findpost&pid=145687\"][{POST_SNAPBACK}][/a]

Good question, successful manufacturers in competitive markets don't normally spend significant amounts of money on hidden components unless they're delivering a benefit.

Incidentally, I've only recently upgraded my P25 to a P45+ so I can't speak with great authority, but I've just had my first experience of P45+ moire (caused by the pattern of tiled roofs in the middle distance), I was using a Rodenstock Digital lens at the time, which is a superb optic but not as highly resolving as their Digital H series.
Logged
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #39 on: October 13, 2007, 02:47:20 PM »
ReplyReply

Removing an AA filter is a bad thing to do.

1) Bayer pattern sensors need one to work properly. Demosaic algorithms are designed for properly optically low passed images. Remove the AA and the rest of your software is not working optimally

2) Moire and aliassing are not software removable without also removing real detail.

3) Moire and aliassing are made worse by downsampling

4) Moire and aliassing interfere negatively with compression algorithms (like making jpegs for the web) - this is even more so of a problem with motion images, which is currently my main concern

What is the solution? Oversampling with a properly optically low passed sensor. ie - downscale the image afterwards if you want it "sharper". To keep our "resolution" we need higher quality smaller photosites.

Graeme
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
Pages: « 1 [2]   Top of Page
Print
Jump to:  

Ad
Ad
Ad