shifted blue channel due to conversion?

[WolleWu]

hi guys,

i’m having serious problems keying this kind of footage.

as you can see there is a green halo on the right side of her coat. i was wondering where this colour shift comes from? this is f****** annoying because the halo on the right is in every shot we did. we shot this one with a jvc-hd100 but didn’t record on tape but routed the signal through a YUV-SDI converter and recorded the SDI signal directly onto the inferno (PAL, 4:2:2, 8 Bit). i’m just curious how this shift or halo could have happend? any ideas about how to get rid of the halo? i can post a full res still if you’re keen to play around with it.

cheers and have nice weekend

ps: this image is zoomed by 300% – just to avoid confusion 😆

[loops]

Oooh dear… could be an analogue fault. This might sound silly or obvious but did you use radically different lengths of cable for the U and V connections? You could try converting back to seperate Y, U and V channels, and see if shifting U or V left a bit helps. Can you still see the halo if you view the footage in black and white?

[WolleWu]

hi loops and the rest of you,

gnaaarrrr, i had a look at the yuv channel via keyer/range/yuv. now i’m 100% sure that the b-y channel is shifted to the right side, damn it. trust me you don’t wanna know long the cord was 🙂 but i’m quite sure all of them had the same lenght. somehow the b-y channel must have lost some information on it’s last 5000m… 😆 just kidding. now for you cracks out there, i’m sitting in front of an inferno (4.7) without any sparks or fancy stuff on it. i’m quite lost in figuring out a way to seperate yuv without using sparks…i know there is a tinder spark for that 😉 but i don’t have one on this machine. i read that you are quite fit in hacking some sparks and stuff for the ffi systems, any ideas loops? anybody out there heeelp! a shout on me for the whole night for the one who can help me out.

gonna hit the sack ….an start crying 😯

[loops]

Always wondered how people went straight into Inferno from a greenscreen stage… so… with stupid long wires, then? 🙂 If one cable was slightly different or older or broken or something it could affect the transmission speed down that wire and so do this.

Anyway you don’t need any fancypants sparks – two of the example sparks that come with Inferno do RGB -> YUV and the reverse. You should find them in /usr/discreet/inferno_4.7/sparks or similar. Use sparkYUV on your footage, then Seperate, shift the blue or green (could be either), Recombine, and put the result through sparkRGB.

Sadly the two sparks don’t work in Batch, I could fix that but it’s midnight here. Good luck 🙂 I guess you could take both chroma channels into Action to line them up, if you’ve got something sharp in frame, maybe if you can see the edge of active picture on the delayed one?

[WolleWu]

aaalright, i seperated the channels via Seperate. am i right that the green channel seems to be the Y one? why is the separate/combine node not included in batch, would make things way easier 🙂 thx for your help gonna let you know if everything turned out well, in case you wann a know.

cheers mate

[loops]

Seperate and Combine are in Batch in the newer versions. I suppose that spark gives you Y on R, U on G and V on B but I can’t remember off the top of me head. Well if it was obvious in the keyer which one was wrong I guess it should be obvious if you take them all into Action 🙂 Do let us know how it all pans out.

[WolleWu]

okay, it’s done. to make a long story short. i converted the RGB footage to YUV with the menitioned YUVSpark and seperated Y U and V

after that i figured out the evil channel and shifted it via action 3px to the left..recombined the YUV footage..and converted it back to RGB.
here are some images to compare the different results, on the left side is the evil footage and on the right the corrected one.

the YUV footage, left uncrorrected, right corrected

the converted RGB footage, left uncrorrected, right corrected

after pulling a key with the MK

yay…..a crisp matte and a lot more detail in the hair and no stupid halo around the brown coat 🙂 this whole process takes 4 times the space and i can’t get it to work via batch processing because i don’t have a Seperate and Combine Node in batch (Inferno 4.7) and the Sparks don’t work in Batch either. can’t blame discreet, only my old blue greyish baby 🙂

thx for that loops

ps: this post is quite detailed just in case someone else is having the same kind of problem and needs some advise

[Anonymous]

hi,

nice work!

you can do the separate/combine in batch by linking the image onto 3 layers & cc L1 with G+B gain to 0, L2 R+B gain to 0, L3 R+G gain to 0
then add 1 image from each & set it to simple add; action should have back off
this should give you the desired result & enable the shifting too in one pass ..
just the sparks have to stay outside batch

ciao
robert

[Anonymous]

robc wrote:

you can do the separate/combine in batch by linking the image onto 3 layers & cc L1 with G+B gain to 0, L2 R+B gain to 0, L3 R+G gain to 0
then add 1 image from each & set it to simple add; action should have back off

This is illustrated in a tip by Henry Birdseye: http://www.fxguide.com/fxtips-51.html

[WolleWu]

first of all thx for your help. unfortunately the cc batch stuff doesn’t work for me because i’m temporarely working in YUV color. it would work perfectly if i would have a RGB image, but i have to shift a chroma channel not a RGB color channel. Seperate is giving me complete different results compared to your batch pipe…did i miss something?

cheers wollewu

[Anonymous]

hi,

actually i tried it with a yuv sequence (converted by the spark)
the cc procedure in batch/action does exactly the same as the separate/combine modules, the difference beein, that the modules separate monochrome channel images & add the color components (which are actually RGB) in combine & the action setup gives you the RGB components in their respective color.

if you want 1:1 identical workflow results as with separate/combine you can cc > reenter: R>R, R>G, R>B for red + G & B accordingly (this is separate) and set the action > cc to RGB gain as explained. be sure to switch action > surface(image) > filtering off

ciao
robert

[loops]

If you really really wanted you could do the colourspace conversion with CC nodes too – it’s linear, so each of Y, U and V are just weighted sums of R, G and B. If you really wanted to do this you could get the coefficients from Charles Poynton’s excellent article about the conversion: http://images.autodesk.com/adsk/files/Merging_RGB_and_422.pdf

You could probably do the RGB to YUV using the keyer in YUV mode outputting mattes with 0 tolerance and max softness but this is just getting silly 🙂

[Anonymous]

hi loops,

good idea!!
once you happen to have a setup going inthis direction, i´d be really curious to see (if you wouldnt mind to share)
i tried once, but gave up – my math is too weak … (i learned latin & old greek at school, instead ;-))

ciao
robert

[loops]

RGB to YCbCr using CC and Actions nodes is easy enough but I’d forgotten what a pain in the arse doing the inverse is… you run into all kinds of nasty issues with clamping and negative numbers and quantisation. Last time I did this it was in a pixel shader in floating point which was a lot easier.

Anyway here’s the setups to do the RGB->YUV part: http://lewissaunders.com/ycbcr/ycbcrBatch.tar.gz. Hopefully it makes some kind of sense… CC nodes do the Seperate part, the matrix multiplication (basically) is done by Action nodes with Simple Add and Subtract transfer modes and transparency. The half_grey node in there is for adding 0.5.

And since I failed to complete the second half of the colourspace conversion playground laboratory OF DOOM, I wrote RGB->YUV and YUV->RGB sparks which work in Batch, you can get them here: http://lewissaunders.com/ycbcr/colourspaceSparks.tar.gz. They work fine but the usual arse-covering disclaimer applies 🙂 There’s 32- and 64-bit binaries for Irix and the source too. Enjoy!

[Anonymous]

Hey loops,

great, thanks a lot !
i´ll definitely check the setups out tomorrow (hope i´ll at least understand something ;-))

cheers
robert

[Author]

Posts