Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Low pass (anti-alias) filter or not?

  • Please log in to reply
10 replies to this topic

#1 sharkmelley

sharkmelley

    Vanguard

  • *****
  • topic starter
  • Posts: 2233
  • Joined: 19 Feb 2013

Posted 15 August 2019 - 04:53 AM

What are people's thoughts on the advantages/disadvantages of the AA filter built into the sensor stack of many DSLR/Mirrorless cameras?

 

I'm assuming that very good optics are being used, to form tight stars occupying just a few pixels with wide field optics.

 

With no AA filter, a single exposure will contain a mass of bizarrely coloured stars.  The stars only return to their proper colour in a stack of dithered exposures - whether the dithering is intentional or is caused by drift.  The advantage is that Bayer Drizzle stacking will give very sharp stars and excellent resolution.  But how many dithered exposures are required before we obtain acceptable star colours (clearly it depends on star width, the FWHM)?

 

With an AA filter, a single exposure has stars that are much more normally coloured and very few dithered exposures are required to give perfect colour.  The disadvantage is the slight smearing and loss of resolution caused by the AA filter, even when using Bayer Drizzle stacking.  But maybe the sharp resolution can be restored by deconvolution with the impulse response of the AA filter?

 

I'm very interested in people's thoughts on this and whether or not you consider it an important criterion when selecting a DSLR/Mirrorless camera.

 

Do you know of any studies or experiments that have been done on this issue?

 

Mark


Edited by sharkmelley, 15 August 2019 - 05:18 AM.


#2 whwang

whwang

    Mercury-Atlas

  • *****
  • Posts: 2782
  • Joined: 20 Mar 2013

Posted 15 August 2019 - 05:18 AM

All my modified DSLRs have their AA filter removed, according to the vendors who did the modification for me.  Unfortunately I never had a chance to compare the images before/after the mod, since pretty much the first thing I did to a camera is to modify it.

 

On Canon 5D2 (6.4um pixel), I can clearly see the improvement brought by Bayer drizzle comparing to normal stacking.  On D800 (4.9 um pixel), the difference between Bayer drizzled images and regular stacked images becomes quite subtle.  These are based on images taken with 1100mm focal length, so the images are almost never sharp enough to produced weird color on single-exposure images (because of seeing).

 

I guess this information is not too useful to you.



#3 mmalik

mmalik

    DSLR camera modifications

  • *****
  • Vendors
  • Posts: 9535
  • Joined: 13 Jan 2012
  • Loc: USA

Posted 15 August 2019 - 07:27 AM

Following is the order which mostly will determine what kind AA one may end up with:

 

 

1. Design of LPF filters (discrete or stacked)

 

2. Type of mod LPF design will allow

 

3. Type of aliasing one will end up with post-mod

 

 

In that order, AA ends up being the result, not a choice per se. Of course if you remove all filters, you get rid of AA. If you keep one (i.e., LPF-1 where possible), you end up with partial AA. Various design permutations (...some assumed) of AA shown here.... Regards



#4 t_image

t_image

    Gemini

  • -----
  • Posts: 3094
  • Joined: 22 Jul 2015

Posted 15 August 2019 - 10:36 AM

Great question Mark.

Maybe you can tell me if my thinking is off.

 

When you say "With no AA filter, a single exposure will contain a mass of bizarrely coloured stars."

I get you infer the situations where some star are only illuminating single pixels that are behind single color filters of the filter array so the happenstance of the CFA filter the illuminated pixels behind determine the accidental bizarre color?

Of course star drift (imprecise tracking), atmospheric conditions that don't allow best FWHM, focus issues, and undersampling created by optics/sensor mismatch also help blur even non-AA filtered setups to eliminate this artifact? [as you've hinted at being qualifications to the thought]

 

My wonder (without too much deep thinking) is might there be some physical property of the AA filter and CFA filter size/pixel size match that is calibrated for, in the camera manufacturing process,

Making the models that are intentionally made/sold with or without the AA filter may have a slight difference than

modded camera (built with AA, but had it removed, thus the camera 'thinks it still has a AA filter')?

..I guess I might add maybe more relevant to design in native lenses where the spatial resolution is predictable to a particular precision.

Maybe if there is it is negligible, or I may be guessing wrong about well-focused light and how AA v non-AA would differ in expected ways.

 

I ponder the thought (AA v no AA) as it would be the sole motivation to mod my normal a7s (AA removal for better resolution),

but my typical optics create undersampling with it anyways so I convince myself it isn't worth any benefit....

 

Maybe with narrowband IR (like CH4 filter with gas planets or Lunar) where all pixels are getting light through the CFA, that the monochrome captures would benefit immediately from better spatial resolution sans AA....

But IR sensitivity presumes a mod anyways....

 

The trend to non AA builds seems consistent with the RAW v. JPG where photogs rather have more authentic data/and more control and decide how to deal in post (i.e. 'let me deal with the moire artifact in some images, rather than deciding for me that all my images are better blurrier so I don't get moire in rare circumstances..")

But I guess it's almost similar to deciding 'sharpen' settings applied in-camera or at RAW import before taking the chance to look at the image,

v. importing with no sharpen applied and deciding on specifically where in an image it might be effectively applied to the original captured data!



#5 sharkmelley

sharkmelley

    Vanguard

  • *****
  • topic starter
  • Posts: 2233
  • Joined: 19 Feb 2013

Posted 15 August 2019 - 03:29 PM

Great question Mark.

Maybe you can tell me if my thinking is off.

 

When you say "With no AA filter, a single exposure will contain a mass of bizarrely coloured stars."

I get you infer the situations where some star are only illuminating single pixels that are behind single color filters of the filter array so the happenstance of the CFA filter the illuminated pixels behind determine the accidental bizarre color?

Of course star drift (imprecise tracking), atmospheric conditions that don't allow best FWHM, focus issues, and undersampling created by optics/sensor mismatch also help blur even non-AA filtered setups to eliminate this artifact? [as you've hinted at being qualifications to the thought]

 

That's right. Here's an actual example from when I was reviewing the Canon 200D vs Nikon D5300:

 

CanonNikonSpatialFilteringV2.jpg

 

I used a test target of white dots.  Ignore the third panel which shows how the spatial filtering on the Nikon (which kicks in for exposures 1/4sec and longer) turns the "stars" green.  See how the "star" colours in the first two panels vary from blue to red to green.  It's a kind of colour Moire effect caused by an individual pixel value being a lot brighter than the neigbours.  This pixel then dominates the star colour.

 

Interestingly, the Canon 200D has an AA filter but the Nikon D5300 does not.  I guess the AA filter on the Canon 200D is not very strong.

 

What interests me at present is that the Nikon Z6 has an AA filter but the higher resolution Nikon Z7 does not.  I wonder what difference this makes in practice.

 

Mark


Edited by sharkmelley, 15 August 2019 - 03:31 PM.


#6 sharkmelley

sharkmelley

    Vanguard

  • *****
  • topic starter
  • Posts: 2233
  • Joined: 19 Feb 2013

Posted 15 August 2019 - 06:23 PM

I have now tried the same experiment on my Nikon Z6 mirrorless camera which has a AA filter (unlike the Nikon Z7 which doesn't).  Here's the result I obtain from imaging a monochrome dot pattern at f/5.6 with a "nifty fifty" 50mm lens:

 

DotMoire.PNG

[Click on image to see it full size]

 

We clearly see red, green and blue "stars" and all colours in between.  The result is very similar to what I saw on the Nikon D5300 and Canon 200D in the previous post.

 

Conclusion:  In practice AA filters do not seem to prevent bizarre star colours.  For undersampled stars (i.e. widefield imaging) dithered acquisition and stacking is the only way to be sure of having natural star colours.

 

If you want to try this for yourself then the "dots" image can be downloaded from: https://drive.google...XR01drsSjI/view

 

Display the image at 200% scale on your monitor and adjust the distance of the camera from the screen so that the dots have a spacing of close to 10 pixels in your image.  Be careful not to let the "stars" saturate.

 

Mark


Edited by sharkmelley, 16 August 2019 - 12:45 AM.


#7 Kevin_A

Kevin_A

    Vostok 1

  • *****
  • Posts: 124
  • Joined: 22 Aug 2018
  • Loc: London, Ontario

Posted 15 August 2019 - 06:39 PM

My understanding of the AA on both the Z6 and also the D750 has ALL to do with pixel size "blockiness" compared to the smaller and finer pixel size of the Z7. 

 

Without the AA the much larger pixels are very blocky and jagged for 24MP and both the Z6 and D750 have fairly weak AA filters to overcome this blockiness without being too soft an image. 

Someone mentioned that maybe in only 1 direction as well, but not sure about that.

 

All I know is that the larger pixels of the Z6 and being only 24MP isnt for pixel peepers even with the weak AA filter unless you use a 800mm scope or higher.

 

I find that my D7200/D5300 with no AA filters and 3.9uM pixels looks better than my D750/Z6 with a AA filters and 5.9uM pixels all day/night long... but that maybe due to reach of the APS-C showing more detail.

Noise is a different kettle of fish and the reason why I keep on using both fullframe and the non AA APS-C cameras. After using both for 2 years now I would not use my Z6 for anything but widefield (14mm-135mm DSLR lens or above 800mm with a scope as in between just wasn't very nice compared to APS-C smaller pixel sizes. (too jagged n soft)

 

Just my thoughts from what I have learned (non-technical, but in reality using them)....


Edited by Kevin_A, 15 August 2019 - 06:46 PM.

  • tonyt likes this

#8 sharkmelley

sharkmelley

    Vanguard

  • *****
  • topic starter
  • Posts: 2233
  • Joined: 19 Feb 2013

Posted 16 August 2019 - 12:44 AM

For fun here is a 1:1 crop of a Nikon Z6 image I opened in Photoshop with standard settings.

 

It's a 1/5sec exposure taken with a 50mm lens at f/5.6 and the laptop computer was displaying a pattern of vertical black and white lines:

 

DSC_0848_f56_iso100.jpg

 

The camera distance from the laptop was carefully adjusted so the stripe spacing on the sensor was very close to the pixel pitch.  I used f/5.6 because for this particular lens, f/5.6 gave the strongest MoirĂ© effect.

 

Mark


Edited by sharkmelley, 16 August 2019 - 12:48 AM.


#9 t_image

t_image

    Gemini

  • -----
  • Posts: 3094
  • Joined: 22 Jul 2015

Posted 16 August 2019 - 01:30 AM

........................

I used a test target of white dots.  Ignore the third panel which shows how the spatial filtering on the Nikon (which kicks in for exposures 1/4sec and longer) turns the "stars" green.  See how the "star" colours in the first two panels vary from blue to red to

I'm curious about the test.

One my LED display, the "white dots" aren't really white dots

so I'm wondering how a CFA matrix properly sampling the source display,

matches up with the display pattern, especially when the 'white dots' are really:

whitedot.jpg



#10 t_image

t_image

    Gemini

  • -----
  • Posts: 3094
  • Joined: 22 Jul 2015

Posted 16 August 2019 - 01:41 AM

I remember my old 1990 home computer had a trick that utilized 'high resolutions' with more colors (than the computer was capable of generating) by creating a moire visually on a CRT by lighting adjacent pixels at intervals so it produced green or red or blue via the different offsets at the level of the display's pixel pitch.....(called 'artifacting')

http://www.atarimani...facting_22.html



#11 sharkmelley

sharkmelley

    Vanguard

  • *****
  • topic starter
  • Posts: 2233
  • Joined: 19 Feb 2013

Posted 16 August 2019 - 03:29 AM

I'm curious about the test.
One my LED display, the "white dots" aren't really white dots
so I'm wondering how a CFA matrix properly sampling the source display,
matches up with the display pattern, especially when the 'white dots' are really:
whitedot.jpg


That's a good question. The test image of white pixels is displayed at a scale of 200% so each one occupies a block of 2x2 screen pixels. The lens cannot resolve the individual RGB elements. Similarly for my pattern of alternating vertical lines - each line pair occupies a screen width of 4 pixels.

Mark


CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics