get rid of CA aberration using a HVX-200 and the Brevis35

[Biel Bestue]

the big problem when solving CA aberration when using a HVX-200 is that the CCD is moved to the left

if you try to get rid of that aberration (with Virtual dub and a correctCA plugin (you can find it somewhere here in the forum)) you end up with a part of the image corrected and another uncorrected

so i've tried to find a way to solve this

first i set the camera to z 72 and mf 8/9

than i include the achromat from Brevis to the camera (so i get the aberration, it's not necessary to include the whole 35mm adapter)

now i get a nice CA aberration

i load the footage in Combustion and move the whole thing 60 pixels to the left

i keep the whole thing at 1440*1080 with a pixel aspect of 1.33 but now i render it to 1560*1080

that keeps the whole original image plus a black line to the right side (that is 60 pixels wide)

then i load it to VirtualDub and make the CA aberration corection process with a 1*blue 1.003*red 1.002*green

render it, open it back at Combustion and move the footage now 60 pixels back to the right

render it now at usual 1440*1080 1.33 and thats it

hope it helps :)

i estimated the center position of the CCD's... i aproximated it, that's not 100% truthful
i hope someone with enough time will find the real center acurately

the images at the end still have a little bit of CA but i guess we'll have to live with it, i don't think we'll ever get rid of it completely

you can use Combustion or whatever you want to render the files, just make shure you have complete control of waht the program catches as render target and how it renders it

[Dennis Wood]

Biel, we're seeing some interesting things with our optical flip module. The optics changes combined with path elongation when using the flip are reducing CA levels dramatically in our test rigs. Footage to follow with the latest setup :-) Hopefully we can reduce your work flow in post!

[Biel Bestue]

oops i found a problem with the method

since the CA correction is made to a digitally anamorphic image, the correction then suffers from anamorphic aberration

so in order to avoid such stupid situation the solution is to work only with 1920*1080 footage

that means

- export the footage up to 1920*1080 square pixel

- import it into the composition program (Combustion / AE..)

- make the composition as big a s 2080*1080 (1920 + 80 + 80 = 2080)

- move the footge -80 pixels (to the left) (80 = 60*1.33333333, 60 was the average pixel position i had found the ccd was moved to the right, and that was when the image was not square pixel but 1.333333 )

- the image should end up touching the left side of the composition and should have in his right a black stirpe 160 pixels wide touching the right side of both the footage and the composition

- render it

- CA corect it with VirtualDub and the pluggin with the same values that are: 1*blue 1.003*red 1.002*green

- render it 2

- create a new composition back at the comp program, make it 1920*1080

- and move the footage back at it's original position +80 pixels (to the right)

- render it 3

- smile


if you want also to eliminate the barrel aberration of the HVX-200 (although in Z72 I guess it's unoticiable) you can use 2d3's Boujou (a 3d tracking program)

i'll sugest skip the last repositioning (+80 back at right) and keep working in a lens-centred version of the footage since every camara-dependant effect/aberration will be lens-centered

the same if you wanna make some 3d tracking

the end result should then be recentred to fit the 1920*1080 standard just at the end of the whole process

Originally Posted by Dennis Wood

Biel, we're seeing some interesting things with our optical flip module. The optics changes combined with path elongation when using the flip are reducing CA levels dramatically in our test rigs. Footage to follow with the latest setup :-) Hopefully we can reduce your work flow in post!

it's great to hear such news, but out of curiosity... elongation of the path light have to go don't mean a bit loss of light?

anyway Dennis your adapter works wonderfully!

i would like to make a request to you, since you have those matteboxes (that are lens-centred!) would it be much of an anoyance to try something for the comunnity? put a sheet of plastic with a small mark in the center in the mattebox, center it well within the mattebox and post here the resulting image

that way we should have a real position of the lens center (even we would know if the ccd is also "Y" offsetted)

can you do it? thanks in advance :D

[Dennis Wood]

The most accurate way to do this would be with a step down ring. An 82-55 step down ring screwed onto your HVX threads would introduce vignetting as you zoomed out, and this would be very easy to see and measure.

[Biel Bestue]

but since you zoom out you don't know the exact pixels of the position, the idea would be to keep it focused to Z72 at MF 8/9 an use an adapter to make it just as equal as when using it

if you zoom out you loss the visual cue of that position!

[Jon Wolding]

You can probably figure out how off center your CCD is like this:

Lock the HVX off on a tripod. Set up a target that is centered on the frame. Perform a slow zoom out (max telephoto to max wide) while recording (720 or 1080).

Then, in your NLE, put a small square over the target center (hopefully it's at the XY center point). On playback, you'll notice the square "drift" left/right. On the max wide shot, move the square to the target center and check out the new XY coordinates. This is your pixel drift. Let's say it's 35 pixels over the entire zoom range.

Can someone else help me out with the math here?
If we know the size of the CCD, how many millimeters across 35 pixels translates to, and lens info at max telephoto and max wide, then we should be able to come up with a simple formula to determine off-centeredness for any camera sensor.

[Biel Bestue]

the question is we don't know the size of the CCD for sure (since there are features built in like OIS or STEADYSHOT, the number of active pixels of the CCD could change from a company to the other)

HVX-200 is 55mm focal distance at Z0

http://upload.wikimedia.org/wikipedi...ryTriangle.svg

In this right triangle: sin(A) = a/c; cos(A) = b/c; tan(A) = a/b.

a would be half the ccd (counting it is centred...)
A would be half the FOV
C is 90º always!

("c" square) equals to ("a" square) plus ("b" square)

but anyway since we don't know 55mm is for sure (it could be 55.3 or 55.2 or .whatever, it could be rouded to look cool printed in the camera lenses!)
and since the ccd is milimeters big, a little error in the math would represent a lot of error in pixels since like 6 milimteres wide represent 1920 pixels !

since this could be difficult to establish right i would recommend to approximate the problem visually

[Biel Bestue]

i got the final number footage should be shift

60 pixels at square pixels or 45 at 1.3333

and export it at 1530 at 1.3333 or 2040 at square

if you zoom out/in quickly you get an area that it's barely blurred, that area is the center of the lens, and it's 60 pixels to the left/right (depending if you're doing the 180º flipflop)

i hope it helps some folks :P