Ad
Ad
Ad
Pages: [1] 2 3 ... 11 »   Bottom of Page
Print
Author Topic: 1Ds MKIII and Optical Low Pass filtering  (Read 119828 times)
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« on: October 18, 2007, 05:50:49 PM »
ReplyReply

Optical low pass filters (OLPF's) or Anti Alias filter don't come in different "strengths". The filter works in two passes, each layer splits the light in two either horizontally or vertically, so by combing them together, you get vertical and horizontal filtering. The distance of seperation of the two rays of light is governed by the thickness of that layer, so if you want, (or need to as you don't have square  photosites) you can adjust the filter accordingly. You choose the thickness of the filter in relation to the spacing of the photosites on the sensor.

Aliasing is what occurs when too high a frequency (too detailed information) enters a sampling system. A CMOS or CCD sensor has a regular array of photosites, and is a sampled system. If you put in too much detail, any detail beyond which the sensor can handle is "folded back" as an "alias" into recordable and visible frequencies, producing aliassing artifacts known as moire or "jaggies".

Once aliasing gets into a sampled system, it is very hard to remove. That is because the aliasses occur in the same levels of detail as real detail in the image. The more detailed the information that caused the alias, the lower the frequency it folds back to, corrupting more and more real image information. That means you cannot remove aliases without also removing real image data.

The thickness of the OLPF is usually chosen to split the light so that it matches the pitch of the green photosites on the bayer array. Green is by far and a way the largest component of luma, and it has the closest spacing on the bayer array, so by setting the thickness for that, you're using the least optical low pass filtering you can get away with. If you were to set the thickness for that of the red or blue, you'd be reducing the resolution of the final image too much, but you would avoid chroma aliasing. As it stands, the best compromise (and all engineering is a series of educted compromises) is to filter the green correctly, and hope that chroma moire doesn't intrude too much.

One thing that does not remove moire or jaggies, but actually makes things worse, is downsampling. Because downsampling is a filter followed by decimation process where all frequencies not allowed in the small image are removed in the large image, then pixels thrown away to create the small image, if you have aliasing you have frequencies that will not get removed by the downsampling filter as they're folded back into frequencies where detail exists that you wish to keep. Often poor image downsampling filters are used, and these actually create more aliasing on the small image as they're not strong enough to filter out the too high frequencies, and this only makes matters worse.

Now, most of what I've written also applies to three chip systems (like in video cameras) or single chip depth based systems like Foveon. With bayer pattern sensors there's an extra complexity caused by the bayer pattern itself and how it works, and that is when you get aliassing artifacts, you don't just get luma aliasing, but you get chroma aliasing too, and that appears as funky coloured edges to sharp objects. This is quite objectionable - even more so than the pure luma aliasing that you get with the other approaches. Bayer demosaic algorithms that reconstruct the RGB rely on analysis of the surrounding image content to determine the best possible guess of what the colour should be. If there are aliases in there, the algorithm cannot detect which is real detail and which is aliassed, thus causing results that are not as good as they could be.

Aliasing doesn't just add artifacts to an image that should not be there. There are other implications, that perhaps are not as critical for stills cameras, but they do effect my area of speciality, which is moving images. Compression systems work on frequencies, and aliases add extra high frequencies that don't correlate to image content, making the image harder to compress. Also, as I deal with very high definition moving images, anything you might want to do to a still, like "paint" out a problem, is not applicable or appropriate.

The major problem with OLPFs is not that they're in a camera - to me, they're a necessary part of the design to make sampling theory work without producing artifacts - but that because of their physical nature, they're not a very "steep" filter so it's hard to remove unwanted frequencies without effecting wanted frequencies. Of course, the lens' MTF acts as a low pass filter, so we don't have vast amounts of high frequency detail entering the system, but we do have enough to be an issue, especially on the size of photosites we're using to get low noise. A steeper OLPF would obviously solve this, but they don't exist. Similarly and OLPF that would filter the red and blue stronger than the green would help, but they don't exist either.

The strange thing is, in my mind, the more resolution in terms of pixel-count the more you should be using a OLPF as you're more likely to be scaling that image down. They best way around the issues of an OLPF, is to oversample, capturing more pixel resolution than you need, and using a proper downsampling filter to reduce the resolution, but increasing pixel-sharpness in the process. I  personally don't think that aliasing helps algorithms to intelligently upscale images aliassed images, on edges, tend to have pixels that are less correlated with their surroundings, and hence there's no local information to properly infer edge direction, and you can get into the situation where there is no "good" direction through that pixel as it becomes like a saddle-point.

To remove the necessity of an OLPF, you either need to shrink the photosite size (probably leading to noisier images) or go to larger sensors at the same time (and increase your lens cost and size) or go to poorer MTF lenses. Or put up with aliasing artifacts. I think I'm sensitive to them - I don't like them one little bit, or what they do to algorithms for working with images.

Graeme
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
joedecker
Full Member
***
Offline Offline

Posts: 142


WWW
« Reply #1 on: October 18, 2007, 08:25:39 PM »
ReplyReply

Excellent post, right on.    I'd add one thing:

Quote
Optical low pass filters (OLPF's) or Anti Alias filter don't come in different "strengths". .... You choose the thickness of the filter in relation to the spacing of the photosites on the sensor.

I find no discomfort in the idea of referring to the "spread" (you call it thickness) of the OLPFs as a "strength", and it's pretty clear that that's what's meant.  As you note, this can be set to different pitches, e.g., the pitch of the green pixels is a common pitch for it to be set at, you go further to talk about this being an engineering tradeoff between blurring and moire in the red/blue channels.  I don't see any a priori reason to believe that different cameras don't tweak the "thickness", "spread" or "strength" of these filters based on fluctuations over and above tweaks necessary to compensate for changes in pixel size.

But... yeah, I twigged on the same line in the 1Ds3 article you did.  In the end, it's a great article, and none of our tech geeking invalidates anything else MR said about the 1Ds3.

--Joe
Logged

Joe Decker
Rock Slide Photography
http://www.rockslidephoto.com/
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #2 on: October 18, 2007, 08:37:50 PM »
ReplyReply

Given it's a filter, if you look at it, the horizontal would look like (if you set the pixel pitch to that of the sensor photosite) [1/4 1/2 1/4], and similarly vertical. That would make the 3x3 that it generates:

[1/16 1/8 1/16]
[1/8 1/4 1/8]
[1/16 1/8 1/16]

or something like that. Now it's possible by changing the thickness of the layers to alter the "width" of that filter, but the "strength" of the filter is fixed, if you get my thinking. Yes, it's arguing semantics to an extent...

Now, if you don't make the pitch so that it aligns with the photosites, I'm not sure how it would work if make it wider. I can sortof see it working if you made it thinner though.

Indeed, great article, but I hate aliases more than I hate the resolution loss. The lack of sharpness can be quickly and easily fixed with some edge contrast or other sharpening tools, or downsampling and alliasing can not. To me, it's as simple as that.

Graeme
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #3 on: October 18, 2007, 08:42:34 PM »
ReplyReply

And I forgot to add that fill-factor and microlenses also contribute to the overall effect of the OLPF. They cannot provide an anti-aliasing effect, but they can minimise to an extent the appearance of what aliasses do arise.

Graeme
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
Tim Gray
Sr. Member
****
Offline Offline

Posts: 2002



WWW
« Reply #4 on: October 18, 2007, 09:19:32 PM »
ReplyReply

But, at least as I read Michaels report - MFDB's "typically" don't have an AA filter?   I don't really understand how the absolute size of the sensor is relevant to the issue of moire.

... and I don't know how Graeme's first post relates to this quote from the article:

"Here's the Achilles heal of almost all DSLRs. They have an AA filter primarily to save battery power and processing time when computing a JPG. Medium format backs, for example, typically don't have AA filters because they don't need to produce in-camera JPGs, and therefore have no need to demosaic the Bayer pattern in-camera."

I do confess that this is the first time I've heard the reason for having an AA filter expressed this way.

I also wonder at "The reason that anti-aliasing is even mentioned in this review, though the whole AA filter issue applies to almost every DSLR on the market, is that the IIIs is the __first DSLR to offer resolution in what has, until now, been medium format territory.___"  I would have thought the 1ds2 had, at least at one time (ie the Kodak DCS), medium format resolution.

The kind of landscape shooting I do just doesn't raise the circumstances eg: fabric or as per the generic example here: http://en.wikipedia.org/wiki/Anti-aliasing that I would think would cause problems.

As for down sampling, why wouldn't someting like up rez 200% gaussian blur of even 1 pixel, or .5 and downsample from there work?  In any event if I'm downsizing it's most likely for the web and I really don't care as much as I would for a printed image.
Logged
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #5 on: October 18, 2007, 09:27:56 PM »
ReplyReply

Probably because MF backs are large and a large OLPF without defects is probably quite expensive indeed.

I don't see how not having to demosaic in camera to produce a JPEG has any effect on the engineering decision to use an OLPF or not. Because moire and aliasing cannot be removed without removing real image detail, no amount of funky demosaic algorithm will get you out of jail free unless you slap a blur on the image of significant size.

Because aliases mirror around the maximum recorded frequency, their effects can be at quite a low frequency if the input frequency is high enough. At some point the fold back is at a low enough resolution that you'd have to blur the image to the point of destruction to get rid of the aliases. They're insidious once they get inside the system, which is why, normally, people using sampled systems take great care to avoid letting them in the first place.

Graeme
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
Ray
Sr. Member
****
Offline Offline

Posts: 8845


« Reply #6 on: October 18, 2007, 09:28:53 PM »
ReplyReply

Well, the article certainly provides an insight into why it is thought AA filters are needed, but doesn't really explain why Bayer type cameras without either AA filters or microlenses, those ridiculously expensive MFDBs that knock your socks off, get by without them. (I think I'm right in saying that MFDBs do not have microlenses, but if I'm wrong I'm sure you'll correct me   ).

It's interesting to note also that Sigma's Foveon cameras are typically low resolution (the SD14 just 4.3mp) yet the clarity and 3-dimensionality is much admired by some. The cameras are also criticised for producing lots of aliasing and artificial detail, due to lack of an AA filter, which cause the images to appear sharper than they actually are, and no doubt some of that extra clarity is due to the fact that all the pixels are real, ie. non-interpolated.

Nevertheless, if aliasing is not such a big disadvantage with such a low-pixel-density sensor as the SD14, one wonders if it is really necessary with higher pixel density cameras such as the 1Ds3 and particularly the 400D and 40D.
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8845


« Reply #7 on: October 18, 2007, 09:39:17 PM »
ReplyReply

Quote
Because moire and aliasing cannot be removed without removing real image detail, no amount of funky demosaic algorithm will get you out of jail free unless you slap a blur on the image of significant size.
[a href=\"index.php?act=findpost&pid=147085\"][{POST_SNAPBACK}][/a]

I can't quite see the logic of this. Because the Foveon type sensors can actually resolve beyond the Nyquist limit, the false detail is of a higher frequency than the highest real detail, in which case it should be possible to apply a low pass filter in post processing which is just sufficient to remove those highest frequencies which are false, but not the next highest frequencies which are real, if you get my point.

By way of clarification, although I understand that the real high frequency detail that is mixed up with the aliased artifacts, and which is of the same frequency as the false detail, cannot be separated by software, we are, after all, comparing images with and without AA filters.

The image without the AA filter would presumably not contain such real, high frequency detail. At best it would be attenuated.

As already explained, these low pass AA filters are not 'brick wall' filters that are able to cut off all frequencies immediately above the Nyquist limit whilst allowing all frequencies immediately below the Nyquist limit to pass. Frequencies just below the Nyquist limit will be significantly attenuated. If not, there would be no point and no interest in removing the AA filter.

Removal of the AA filter should allow those highest frequencies near the Nyquist limit, including real detail, to be recorded. Software which attempts to remove aliasing artifacts cannot distinguish between the high frequency artifacts and real detail of the same frequency which is however only present due to the lack of an AA filter.

Removal of such real detail during the process of removing the artifacts simply brings us back to square one. We have an image no more detailed than what we would have got if we hadn't removed the AA filter.

However, not all scenes have fine, evenly structured detail which is the subject of aliasing. Are you able to follow my convoluted argument?  
« Last Edit: October 18, 2007, 11:40:15 PM by Ray » Logged
Graeme Nattress
Sr. Member
****
Offline Offline

Posts: 582



WWW
« Reply #8 on: October 18, 2007, 09:45:19 PM »
ReplyReply

ALthough the Sigma doesn't interpolate the luma, the chroma is pretty poor because silicon being a rather poor colour filter, so heavy noise reduction, which by definition in the case of a single image, must use surrounding pixels - ie interpolation - is used. I clearly see aliases on practically every in-focus Sigma shot I've seen. But when you've got such a low pixel count, you need all the help you can get, even if it does lead to nasty artifacts.

If you'd like that level of pixel sharpness, but without the artifacts or need for heavy NR, just properly downsample your bayer DLSR image. Works a treat.

I'm not sure about micro-lens status on MFDBs - Perhaps it's because of the large sensor that people have to stop down a lot shooting with them, and then the aperture diffraction blurs the image enough that an extra OLPF is not really necessary? Perhaps a filter-wheel then that plops in the right OLPF for the aperture you're shooting at then??

As for knocking socks off, yes the mirror slapback on the Hassy did indeed knock my socks off :-)

Graeme

Quote
Well, the article certainly provides an insight into why it is thought AA filters are needed, but doesn't really explain why Bayer type cameras without either AA filters or microlenses, those ridiculously expensive MFDBs that knock your socks off, get by without them. (I think I'm right in saying that MFDBs do not have microlenses, but if I'm wrong I'm sure you'll correct me   ).

It's interesting to note also that Sigma's Foveon cameras are typically low resolution (the SD14 just 4.3mp) yet the clarity and 3-dimensionality is much admired by some. The cameras are also criticised for producing lots of aliasing and artificial detail, due to lack of an AA filter, which cause the images to appear sharper than they actually are, and no doubt some of that extra clarity is due to the fact that all the pixels are real, ie. non-interpolated.

Nevertheless, if aliasing is not such a big disadvantage with such a low-pixel-density sensor as the SD14, one wonders if it is really necessary with higher pixel density cameras such as the 1Ds3 and particularly the 400D and 40D.
[a href=\"index.php?act=findpost&pid=147086\"][{POST_SNAPBACK}][/a]
Logged

www.nattress.com - Plugins for Final Cut Pro and Color
www.red.com - Digital Cinema Cameras
juicy
Sr. Member
****
Offline Offline

Posts: 254


« Reply #9 on: October 19, 2007, 04:26:35 AM »
ReplyReply

Hi!


Quote
I think I'm right in saying that MFDBs do not have microlenses, but if I'm wrong I'm sure you'll correct me smile.gif

At least P30 and P21 have micro lenses.

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

Posts: 7725



WWW
« Reply #10 on: October 19, 2007, 04:46:46 AM »
ReplyReply

The way I read Michael's post is "don't expect a significant resolution difference compared to the 1ds2", and he is not even talking about wide lenses.

Considering that the d2x was IMHO in the same league as the 1ds2 already, it might that the actual gap in resolution between the 1ds3 and the d3 is much smaller than the pixel count might lead one to think.

TBC with testing of course.

Cheers,
Bernard
Logged

A few images online here!
mahleu
Sr. Member
****
Offline Offline

Posts: 564


WWW
« Reply #11 on: October 19, 2007, 05:03:51 AM »
ReplyReply

Quote
Considering that the d2x was IMHO in the same league as the 1ds2 already, it might that the actual gap in resolution between the 1ds3 and the d3 is much smaller than the pixel count might lead one to think.
[a href=\"index.php?act=findpost&pid=147132\"][{POST_SNAPBACK}][/a]

Most of the world still seems to be convinced that more megapixels are better though.
Logged

______________________________________________________________________
Anyone selling a 1DSIII or 6D cheap?
BernardLanguillier
Sr. Member
****
Offline Offline

Posts: 7725



WWW
« Reply #12 on: October 19, 2007, 05:41:05 AM »
ReplyReply

Quote
Most of the world still seems to be convinced that more megapixels are better though.
[a href=\"index.php?act=findpost&pid=147133\"][{POST_SNAPBACK}][/a]

Yep, but too bad for them.

I tend to be fact based. I'll probably rent a 1ds3 with a 16-35 f2.8 and will compare it to the D3 and Mamiya ZD to see to what extend "most of the world" is right to believe that more megapixels actually help.

Cheers,
Bernard
Logged

A few images online here!
francois
Sr. Member
****
Offline Offline

Posts: 6727


« Reply #13 on: October 19, 2007, 06:01:51 AM »
ReplyReply

Quote
Yep, but too bad for them.
...
I would say too bad for those who believe that more pixels is not always a good thing! I'd trade a couple of MP for better dynamic range.
« Last Edit: October 19, 2007, 06:02:42 AM by francois » Logged

Francois
BernardLanguillier
Sr. Member
****
Offline Offline

Posts: 7725



WWW
« Reply #14 on: October 19, 2007, 06:18:06 AM »
ReplyReply

Quote
I would say too bad for those who believe that more pixels is not always a good thing!
[a href=\"index.php?act=findpost&pid=147138\"][{POST_SNAPBACK}][/a]

More pixels are always a good thing all other things being equal. I wouldn't be interested in having to use a 30+MP AA filtered sensor to get less detail that I am already getting with my 22MP ZD.

Quote
I'd trade a couple of MP for better dynamic range.
[a href=\"index.php?act=findpost&pid=147138\"][{POST_SNAPBACK}][/a]

Same thing here.

cheers,
Bernard
Logged

A few images online here!
frankric
Jr. Member
**
Offline Offline

Posts: 52


WWW
« Reply #15 on: October 19, 2007, 07:08:54 AM »
ReplyReply

Quote
I tend to be fact based. I'll probably rent a 1ds3 with a 16-35 f2.8 and will compare it to the D3 and Mamiya ZD to see to what extend "most of the world" is right to believe that more megapixels actually help.

Cheers,
Bernard
[a href=\"index.php?act=findpost&pid=147135\"][{POST_SNAPBACK}][/a]

Bernard

Unless you want to handicap the 1Ds lll (and I'm sure you don't) I wouldn't think the 16-35 would be the best lens to try it with. My thinking is that only decent primes or the best zooms - e.g. 70-200/2.8 - would be up to the task. The 24-105/4 is pretty sharp too. Pity about the distortion and vignetting, but at least they're easily fixable in software. Hopefully Canon will release some decent ultrawide glass before too long.....

I may get the chance to compare 'apples to oranges' and shoot a 1Ds lll alongside my ZD shortly. I'll be interested to see how they compare.

Regards

Frank
Logged

BernardLanguillier
Sr. Member
****
Offline Offline

Posts: 7725



WWW
« Reply #16 on: October 19, 2007, 07:12:34 AM »
ReplyReply

Quote
Bernard

Unless you want to handicap the 1Ds lll (and I'm sure you don't) I wouldn't think the 16-35 would be the best lens to try it with. My thinking is that only decent primes or the best zooms - e.g. 70-200/2.8 - would be up to the task. The 24-105/4 is pretty sharp too. Pity about the distortion and vignetting, but at least they're easily fixable in software. Hopefully Canon will release some decent ultrawide glass before too long.....
[a href=\"index.php?act=findpost&pid=147146\"][{POST_SNAPBACK}][/a]

As a landscape guy, I use wides very often, the 16-35 f2.8 II is the lens I would be using and it is the reason why I would test the 1ds3 with this lens.

I am not really interested in doing a generic comparison, I would focus on my applications.

Cheers,
Bernard
Logged

A few images online here!
Tim Gray
Sr. Member
****
Offline Offline

Posts: 2002



WWW
« Reply #17 on: October 19, 2007, 09:19:52 AM »
ReplyReply

Quote
As a landscape guy, I use wides very often, the 16-35 f2.8 II is the lens I would be using and it is the reason why I would test the 1ds3 with this lens.

I am not really interested in doing a generic comparison, I would focus on my applications.

Cheers,
Bernard
[a href=\"index.php?act=findpost&pid=147147\"][{POST_SNAPBACK}][/a]

Are you happy with the 16II?
Logged
Christopher
Sr. Member
****
Offline Offline

Posts: 944


WWW
« Reply #18 on: October 19, 2007, 10:00:14 AM »
ReplyReply

I tested two copies of the 16-35 f2.8 II and gave them all back. Sorry but the lens is still crap, perhaps they should have dones something like 16-28 or so. I also use wides, but I only use Leica below 20mm than zeis and canon starts at 35 ;-)
Logged

NikosR
Sr. Member
****
Offline Offline

Posts: 622


WWW
« Reply #19 on: October 19, 2007, 12:00:38 PM »
ReplyReply

Quote
Removal of such real detail during the process of removing the artifacts simply brings us back to square one. We have an image no more detailed than what we would have got if we hadn't removed the AA filter.

No, in  fact it might bring us back to square -3 since aliasing will be apparent (and mixed with the signal) at frequencies lower than the cutoff frequency.
« Last Edit: October 19, 2007, 12:01:19 PM by NikosR » Logged

Nikos
Pages: [1] 2 3 ... 11 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad