Is there a chromatic aberration filter?

[Malow]

hi, i will just enter in hdv world, (actualy using pro-dv) with canon XL-H1, but, this camera with "default" lens has serious chromatic aberration problems...

i will convert to cineform intermediate via avisynth, and i was wondering if exist a filter for CA filtering... is there any?

tnks!

[Mug Funky]

eeks. that'd be good, but i don't know of one.

it _might_ be possible to use something like mvtools (with a large blocksize) to compensate the red and blue channels onto the green one, then merge back in.

...but i've yet to try this, and most likely there'll still be smears of yellow and purple left behind.

[Malow]

mmm i will "try" to test... (i just used mvtools a bit)

there are some good filters for photoshop and paint shop to correct CA, whith very good results, so, i will pray for a holy man to do a "video" version...

what is the "logic" of clean CA? detect a linear color near a bright edge? smooth and shift? aparently CA is a "radial" problem, with more effect on borders of video...

[Blue_MiSfit]

Fantastic camera, the XL-H1. A coworker of mine just got his. I would second your opinion that the default lens suffers from chromatic abberation. It is an excellent camera otherwise.

I would also be interested in testing filters on some sample clips I have (nearly 5 GB of gorgeous 1080i HDV).

[Backwoods]

ChromaShift might help you.
http://www.geocities.com/siwalters_uk/chromashift26a.zip

There is a VirtualDub filter just for CA but I haven't tested it yet. I forget the name of that filter.

EDIT: Here you go, http://dsp.ucsd.edu/~wgardner/VirtualDub.htm

[Fizick]

Usually CA is dependent on zoom factor

[Malow]

mmmm i think i found a easy way... as CA for VD says, is just zoom the channel, i do a script like this:

ConvertToRGB24()
MergeRGB(Last, lanczosresize(1442,1082).crop(1,1,-1,-1), Last)

it separate , zoom and crop the channel.

a few tests show a nice result. will try to tweak...
http://images6.theimagehosting.com/ca.077.th.JPG (http://server6.theimagehosting.com/image.php?img=ca.077.JPG)
http://images6.theimagehosting.com/noca.th.JPG (http://server6.theimagehosting.com/image.php?img=noca.JPG)

[Malow]

Usually CA is dependent on zoom factor

yeah, u are right.
now "this is a problem"... :(

i gess i wil make 2 scripts, one for no zoom, and other for medium zoom...

in full zoom it show almost no CA... OMG...

[JMP]

Ehm... You surely know that you can do ...lanczosresize(1440,1080,1,1,-1,-1)...
to have it faster? OK, strictly speaking it is different, because it crops before resizing, but I don't think it makes any difference in your case :)

[Malow]

Ehm... You surely know that you can do ...lanczosresize(1440,1080,1,1,-1,-1)...
to have it faster? OK, strictly speaking it is different, because it crops before resizing, but I don't think it makes any difference in your case :)

mmm... NOW i know... ;) tnks!

[IanB]

Note!
...resize(1440,1080,1,1,-1,-1)
versus
...resize(1442,1082).crop(1,1,-1,-1)

Are not exactly the same. The boundary conditions in the resizer are different. In the first case the edge row of pixels are not used in the output image, in the second they are. A very minor point.

Also the cropping factors on the resizers are floating point numbers so you can get subpixel adjustment if required.

And as your resize percentage change is very small you might experiment with the taps=N option of lanczos.

[JMP]

Also the cropping factors on the resizers are floating point numbers so you can get subpixel adjustment if required.
...and that can be very useful in this specific case - as it is rather unlikely, that the aberration is by a whole number of pixels exactly...

[Malow]

it may be necessary subpixel, because the best settings show as 4 pixel zoom for green (2--2) ans 2 for red (1--1)

if processing interlaced content, i will need 0.5--0.5 for red, or maybe lower (0.25--0.25)

or even maybe the inverse precess (reduce other colors keping green)

wavelengths info show a proportional deviation about 1% for red and 2% for green (as blue for base)

my case looks exactly this.

i will try to do a test to see "witch one of RGB" is better to upzise/undersize"

[JMP]

Actually the lens is probably axially symmetrical, thus the aberration should be symmetrical around the center of the frame. Therefore may I suggest ...resize(1440,1080,1,0.75,-1,-0.75)... would possibly be even better.

[Fizick]

I am not sure, that abberations value is linear along radius.

[JMP]

I am not sure, that abberations value is linear along radius.
Actually, recalling my somewhat rusty knowledge of geometrical optics, it should probably be close to linear. Chromatic aberration appears because the focal length of a lens system changes with the light wavelength. And changing the focal length results, at least in the "thin lens" approximation, in simple image rescaling (and slight defocusing).

[Fizick]

The real lens in not simple 1-glass. It consist of many glasses, and projected to compensate chromatic abbearration for some zones.

Other point. I know about one program which can get per-frame exposition info from DV stream (probably zoom too).
http://forum.doom9.org/showthread.php?t=114669
maybe it is possible get such info from HDV steam too (with other program), write to some file?, and use somehow.

[Backwoods]

maybe it is possible get such info from HDV steam too (with other program), write to some file?, and use somehow.

HDVInfo
http://forum.doom9.org/showthread.php?t=110021

[Malow]

i think zoom info is not included... :(

[redfordxx]

Hi, Malow

what is your latest script you developed?

I am looking for a way for batch processing of digital camera pictures to remove CA and maybe your script could help...

Thanx

[a uio]

Some good links on Chromatic Aberration:

http://en.wikipedia.org/wiki/Chromatic_aberration
http://wiki.panotools.org/Chromatic_aberration

Cheers,
A

[a uio]

I recently tested my Sony Fx1 for Chromatic Aberration and did indeed find that it is zoom dependent, even to the point of reversing. The red channel is larger than the green channel on full wide, and the red channel is smaller than the green channel on full zoom in. The difference on either end is .1%.

Cheers,
A