Pages: [1]   Go Down

Author Topic: false information "AA filter or not"  (Read 3507 times)

marcmccalmont

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1780
false information "AA filter or not"
« on: March 16, 2012, 01:51:51 am »

I know this has been discussed many times but a lot of us still don't understand completely.
If a camera has no AA filter is false information present when there is a repeating high frequency pattern ONLY or is false information always present when resolving details the size of a pixel? To my eyes (4 cameras without AA filters) I only see it when there is a high frequency repeating pattern and very rarely.
Marc
Logged
Marc McCalmont

DaveCurtis

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 508
    • http://www.magiclight.co.nz
Re: false information "AA filter or not"
« Reply #1 on: March 16, 2012, 02:23:34 am »

I know from a photographers point of view that  non 'AA' images looks better at @ 100% than images taken with an 'AA' filter.  e.g. a M9 image. The smooth blobby look which I sometimes see with my 1DS3 doesnt look that appealing.

 
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: false information "AA filter or not"
« Reply #2 on: March 16, 2012, 02:27:45 am »

Hi,

Here is a sample from Imaging Resource. Left Pentax 645D no AA-filter, right Nikon D800 with AA-filter.

But I essentially agree, if you have used high quality MF and had no issues with AA-filterless design it is obviously not a problem for your kind of photography.

Best regards
Erik


I know this has been discussed many times but a lot of us still don't understand completely.
If a camera has no AA filter is false information present when there is a repeating high frequency pattern ONLY or is false information always present when resolving details the size of a pixel? To my eyes (4 cameras without AA filters) I only see it when there is a high frequency repeating pattern and very rarely.
Marc
« Last Edit: March 16, 2012, 02:29:55 am by ErikKaffehr »
Logged
Erik Kaffehr
 

HarperPhotos

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1309
    • http://www.harperphoto.com
Re: false information "AA filter or not"
« Reply #3 on: March 16, 2012, 02:58:15 am »

Hello,

As a advert photographer I use a Leaf Aptus 75 and Nikon D3x. The main reason I purchased the D3x was because it has an AA filter I just got so feed up with colour moiré in my shots. Now the D3x is used for about 80% of my work with the Aptus 75 for the rest and mostly in my studio.

I down loaded the sample image Erik posted and added an unsharp mask of 100% @ 0.9 pixels to the Nikon D800 image and to my eye the resolution is the same with no colour moiré in the label compared to the 645D.

As a working photographer I can’t afford to have problems like colour moiré holding me up an annoying my clients.

Cheers

Simon
Logged
Simon Harper
Harper Photographics Ltd
http://www.harperphoto.com
http://www.facebook.com/harper.photographics

Auckland, New Zealand

marcmccalmont

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1780
Re: false information "AA filter or not"
« Reply #4 on: March 16, 2012, 03:18:02 am »

Is the quilted pattern in the 645 because the green is made up of duo tone dots not solid green?
proving my premise that a high frequency pattern must be present? The D800 looks plenty sharp w/the AA filter to my eye.
Marc
« Last Edit: March 16, 2012, 03:21:03 am by marcmccalmont »
Logged
Marc McCalmont

Jeremy Roussak

  • Administrator
  • Sr. Member
  • *****
  • Offline Offline
  • Posts: 8961
    • site
Re: false information "AA filter or not"
« Reply #5 on: March 16, 2012, 04:41:47 am »

Here is a sample from Imaging Resource. Left Pentax 645D no AA-filter, right Nikon D800 with AA-filter.
Of course, after you've drunk a few pints of the excellent content of the bottles, you cease to care about any subtle differences between the shots.

Jeremy
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: false information "AA filter or not"
« Reply #6 on: March 16, 2012, 04:52:18 am »

I know this has been discussed many times but a lot of us still don't understand completely.
If a camera has no AA filter is false information present when there is a repeating high frequency pattern ONLY or is false information always present when resolving details the size of a pixel?

Hi Marc,

The simple answer is that lower frequency aliases will always be present, not only with repetitive patterns. However, the aliases will be more visible when they disrupt a repetitive (=predictable) pattern. The human visual system is very sensitive to recognizing patterns and edges.

The only way to avoid aliasing is by pre-filtering the original input signal before it is sampled. Low contrast signals already benefit from a mild low-pass filtering (e.g. diffraction), higher contrast signals require stronger low-pass filtering to reduce the amplitude of the aliases. When the amplitude of the aliases is relatively low compared to the real signal at that same position it may be hard to notice the disruption.

Cheers,
Bart
« Last Edit: March 17, 2012, 09:48:11 pm by BartvanderWolf »
Logged
== If you do what you did, you'll get what you got. ==

marcmccalmont

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1780
Re: false information "AA filter or not"
« Reply #7 on: March 16, 2012, 05:16:29 am »

Hi Marc,

The simple answer is that lower frequency aliases will always be present, not only with repetitive pattern. However, the aliases will be more visible when they disrupt a repetitive (=predictable) pattern. The human visual system is very sensitive to recognizing patterns and edges.

The only way to avoid aliasing is by pre-filtering the original input signal before it is sampled. Low contrast signals already benefit from a mild low-pass filtering (e.g. diffraction), higher contrast signals require stronger low-pass filtering to reduce the amplitude of the aliases. When the amplitude of the aliases is relatively low compared to the real signal at that same position it may be hard to notice the disruption.

Cheers,
Bart
Bart
I love you smart guys, clear precise answers! I'm guessing you would order the D800 over the D800E?
Marc
Logged
Marc McCalmont

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: false information "AA filter or not"
« Reply #8 on: March 16, 2012, 06:18:37 am »

Adding to Barts reply:
My best example of aliasing is rotating wagon wheels in western movies. That is temporal aliasing, but the principle extends to spatial aliasing that we are concerned with here. The film is shot at 24 frames per second. If the wheel is rotating at 24 times/second, it may seem to "stand still" because each movie frame capture a wheel that seems identical to the other*). If the wheel is rotating at slightly more that 24 times/second, it might appear to move backwards. Clearly, this is a "false" impression, caused by an imaging system that is not able to capture the true scene, and instead captures/renders a untrue scene. If the camera had had a long exposure time ("aa filter"), it would smear the wheel into a fuzzy thing that may or may not appear more realistic - at least there would be less room for confusion as to what really went on.

There will always be aliasing in your image (excepting some corner cases that are probably less interesting for photographers).

An AA-filter will only _reduce_ aliasing, not completely remove it. It will do so at the cost of some real image detail. Those details can to some degree be brought back by sharpening/deconvolution. If the product of scene detail, scene/camera motion, lens/diffraction blur etc is sufficient to suppress aliasing to the point where it does no harm, adding an aa-filter may be unwanted, as it will still cause a slight attenuation of image sharpness.

The visibility and degree of annoyance depends on the scene, the viewer, the print and recording technique. Scenes with highly periodic fine detail (e.g. feathers) are known to cause more annoyance than scenes with less periodic fine detail (e.g. leaves).

The use of Bayer color filtering (and debayer in raw development) makes the analysis more complex, but the basic principle is true even for Foveon-type sensors (contrary to popular internet belief).

-h
*)Assuming that the camera has a very short shutter time (generally untrue) and that the wheel has no rotational symmetry (generally not true).
« Last Edit: March 16, 2012, 07:26:43 am by hjulenissen »
Logged

Tony Jay

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2965
Re: false information "AA filter or not"
« Reply #9 on: March 16, 2012, 06:53:43 am »

What the current information in this thread is yet to shed light on is why "35 mm" DSLR's apparently need AA-filters (because manufacturers put them in their cameras) when medium format digital back apparently do not need AA-filters (because their manufacturers do not insert them) when by all the physics I understand regarding sensel density on the sensor and the characteristics of the lenses indicate that sensors with similar or identical sensel density are at similar or identical risk of moire. Better lenses that can resolve more detail (possibly used more frequently by those with MFDB systems) will make the problem worse not better.

On a practical level since the sensel density on MFDB's is very similar to many modern "35 mm" DSLR's there must be an inconsistency here. Either both types of system require AA-filters or they don't.

Again, on a practical level, there must be sort of threshold (as the sensels get smaller) beyond which moire ceases to be a practical issue. It appears MFDB manufacturers feel they are beyond the threshold but DSLR manufacturers do not.
I would really appreciate some clarity explaining the apparent inconsistency regarding the utility of AA-filters and their presence or abscence in different camera systems despite very similar sensor characteristics.

Kind Regards

Tony Jay
Logged

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: false information "AA filter or not"
« Reply #10 on: March 16, 2012, 07:38:18 am »

I would really appreciate some clarity explaining the apparent inconsistency regarding the utility of AA-filters and their presence or abscence in different camera systems despite very similar sensor characteristics.
I don't know. Possible reasons in no particular order (not claiming this to be true):
*Designing an AA-filter costs money. With very small production runs, this cost is unacceptable to MF manufacturers
*Manufacturing an AA-filter costs money. With very large surfaces, this cost is unacceptable to MF manufacturers
*MF users tends to operate in the diffraction-limited regime where aliasing is less of an issue
*MF users tends to be expert operators willing to spend more time shooting/processing each image
*MF users tends to shoot landscape, where repeated structures is less common

-h
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: false information "AA filter or not"
« Reply #11 on: March 16, 2012, 08:48:04 am »

Hi,

It seems that color moiré is often a problem in MF images. The label on the beer bottle was for instance the first image I have analyzed from the P645D, and it was showing it very obviously. But, would I not have a moiré less image to compare with I may not even have noticed.

Very clearly, aliasing is showing up in high quality work. Stopping down to f/11 reduces moiré and so does defocusing. Moiré and other aliasing artifacts may be less obvious in landscape photography. Different demosaic alogrithms may be less or more smart. If you think about tree branch about one pixel wide it should show up shifting color, because it will fall on either R, G, B pixels or in between, but most demosaic algorithms hide this.

Another factor is thickness an cost. An AA-filter is constructed from two slices birefringent cristalls. I think they use unicristals of Lithium Niobate normally. These cristalls are expensive to make and they seem to be quite thick.

The thickess may affect astigmatic correction on traditional wide angle lenses. DSLR optics may take the AA-filter into account. The Nikon D800E is said to have one of it's filter slices inverted so the AA-filtering is eliminated but the thickness of the package is still the same.

Best regards
Erik

 

What the current information in this thread is yet to shed light on is why "35 mm" DSLR's apparently need AA-filters (because manufacturers put them in their cameras) when medium format digital back apparently do not need AA-filters (because their manufacturers do not insert them) when by all the physics I understand regarding sensel density on the sensor and the characteristics of the lenses indicate that sensors with similar or identical sensel density are at similar or identical risk of moire. Better lenses that can resolve more detail (possibly used more frequently by those with MFDB systems) will make the problem worse not better.

On a practical level since the sensel density on MFDB's is very similar to many modern "35 mm" DSLR's there must be an inconsistency here. Either both types of system require AA-filters or they don't.

Again, on a practical level, there must be sort of threshold (as the sensels get smaller) beyond which moire ceases to be a practical issue. It appears MFDB manufacturers feel they are beyond the threshold but DSLR manufacturers do not.
I would really appreciate some clarity explaining the apparent inconsistency regarding the utility of AA-filters and their presence or abscence in different camera systems despite very similar sensor characteristics.

Kind Regards

Tony Jay
« Last Edit: March 17, 2012, 01:21:52 pm by ErikKaffehr »
Logged
Erik Kaffehr
 

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: false information "AA filter or not"
« Reply #12 on: March 16, 2012, 10:01:32 am »

What the current information in this thread is yet to shed light on is why "35 mm" DSLR's apparently need AA-filters (because manufacturers put them in their cameras) when medium format digital back apparently do not need AA-filters (because their manufacturers do not insert them) when by all the physics I understand regarding sensel density on the sensor and the characteristics of the lenses indicate that sensors with similar or identical sensel density are at similar or identical risk of moire. Better lenses that can resolve more detail (possibly used more frequently by those with MFDB systems) will make the problem worse not better.

On a practical level since the sensel density on MFDB's is very similar to many modern "35 mm" DSLR's there must be an inconsistency here. Either both types of system require AA-filters or they don't.

Hi Tony,

The laws of physics do not care about camera models or sizes, therefore the both need AA-filters to reduce aliasing.

The reasons for omitting such filters on MFDBs are multifold.

  • Cost is an isue for the size of OLPF filters needed. I've tried finding sources for large enough optical grade Lithium Niobate wafers of specific thicknesses, and there are few, and the prices even for stock items is steep.
  • The angle of incidence of the edge rays may prohibit the use of some wider angle lenses, especially those without a retrofocus design. Also the use of tilt and shift may add to the variation in filtering efficiency, and color cast may be affected.
  • For an identical Field of View, the projected image on an MFDB has a higher magnification due to the longer focal length. Therefore the detail that would cause aliasing may be partially resolved, and the remaining aliasing thus reduced.
  • MFDBs either require narrower apertures to achieve the sam DOF, thus causing more diffraction which can somewhat functionas an AA-filter, or has more Out-of-Focus zones (due to shallower DOF) which also functions as a low pass filter.


Quote
Again, on a practical level, there must be sort of threshold (as the sensels get smaller) beyond which moire ceases to be a practical issue. It appears MFDB manufacturers feel they are beyond the threshold but DSLR manufacturers do not.

It depends on which issues one encounters in one's practice. For some the false color artifacts are problematic, for others the jaggy edges are an issue when upsampling for large output, and for others the moiré in fabric is the main issue. When sensels reach a 1 micron sensel pitch, then many aliasing issues will stop being issues, but other challenges (e.g. on the practical side, filesize and readout speed) will be the next hurdle.
 
Quote
I would really appreciate some clarity explaining the apparent inconsistency regarding the utility of AA-filters and their presence or abscence in different camera systems despite very similar sensor characteristics.

I hope the above info clarifies that it's not the technique itself, but the differences due to larger image size and magnification.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: false information "AA filter or not"
« Reply #13 on: March 16, 2012, 10:16:20 am »

I love you smart guys, clear precise answers!

Hi Marc,
 
You're welcome, but it's not always easy or possible to give concise answers due to the many variables which require consideration. Your question was clear enough to give a direct answer.

Quote
I'm guessing you would order the D800 over the D800E?

Probably, also based on my experience with 35mm full frame sensors with AA-filter that occasionally also still cause moiré trouble. Jaggies, false color, and such, are not issues I experience. The unknown factor is the effect of the smaller sensel size in combination with a somewhat disabled AA-filter, together with the lenses one uses. My feeling sofar is that the D800 + deconvolution sharpening will not have worse resolution than the D800E which will not respond as gracefully to sharpening. Time will tell.

See, the answer got longer because of the variables involved ... ;)

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==
Pages: [1]   Go Up