View Full Version : 4:2:0 to 4:4:4???
g3nov3s3
07-03-2008, 02:17 AM
Can you convert footage shot in AVCHD at 1920*1080 in 4:2:0 to 960*540 at 4:4:4??????How do you do it?
Would the image look better?
Wayne D
07-03-2008, 04:53 AM
hi - you probably can but it will be 4:2:0 data into 4:4:4....so it will still be 4:2:0 if you know what i mean.....you are going the wrong way up the resolution tree.
it will still look good - what's you delivery format and waht are cutting on?
Peter Majtan
07-03-2008, 04:55 AM
If You are asking if it looked better the downscaled 960x540 as compared to a 960x540 crop from the original AVCHD - of course YES. Otherwise it is a tricky question and very subjective one. What I would do is to convert the 1920x1080 4:2:0 AVCHD to uncompressed 1920x1080 4:4:4 as soon as I have the footage off the original media (not gaining any quality at this point, but also not loosing any) and then keep my DI fully uncompressed. This will guaranty the best possible results, but it will be "little" heavy on resources...
Peter Majtan
07-03-2008, 04:59 AM
Wayno - this is not entirely so. If You are having half the chroma samples at 1920x1080 and You have scaled the footage ti half of its size it will essentially become 8:4:2 (if there was such a thing) which would transpire to perceivable 4:4:4 - of course You will have to get out of the AVCHD codec...
amrrahmy
07-03-2008, 10:50 AM
without calculations, i think u have to downsample twice and even then and even in a really high bit u still wont get perfect colors,(maybe 1.65-1.85) of what the 4:4:4 image would be (if u had 4:4:4 and downsamlpled to 4:4:4)(not adding the compression from the codec).
the image will alter a little from the original when changed to 4:4:4, but as mentioned before, it would make a significant improvement if done before the start of the post.
g3nov3s3
07-03-2008, 10:51 AM
So how do you do it?
Do you convert the 1920*1080/4:2:0 footage to 4:4:4 and then you simply resize it to 960*540?
4:2:0 means that blocks of 2x2 pixels contains average chroma information.
So if you convert this 2x2 pixels to one pixel you get 4:4:4 but half resolution.
But couldn't you let's say apply an improved algorithm when resizing the luma information(the bicubic algorithm for example) and another one for chroma to pull all the info from it?
Chris Kenny
07-03-2008, 12:33 PM
Why would you want to do this? Sure, you end up with 4:4:4 video, but you're getting there by throwing away luma resolution, not by adding chroma resolution. At a given viewing size, the 1920x1080 4:2:0 you started with is going to look better.
amrrahmy
07-03-2008, 12:49 PM
it wont be accurate, it's not about the #s, it's about what would be stored in them.(if the image had already an average assumption to a 4:2:2 and then another assumption into 4:4:4, the image wont be the same as the 4:4:4 image re-sized, so it wont be accurate or real 4:4:4), sometimes u cant even do it in a colorful scene, it would mess it up be changing some colors in parts of the screen and not others(or diff degrees), other than the fact that the whole image would shift colors.
u cant get info that u don't have, but u can improve info that would be applied in post.
amrrahmy
07-03-2008, 12:50 PM
Why would you want to do this? Sure, you end up with 4:4:4 video, but you're getting there by throwing away luma resolution, not by adding chroma resolution. At a given viewing size, the 1920x1080 4:2:0 you started with is going to look better.
exactly
g3nov3s3
07-04-2008, 10:48 PM
Isn't color correction,chroma keying, digital composing,etc. better done at 4:4:4:????
GlennChan
07-05-2008, 01:46 PM
It's done in 4:4:4 regardless. Usually the filter architecture won't let you work on chroma subsampled data... everything is converted to 4:4:4 before being sent to the filter. (Otherwise filter developers would need code to handle 4:4:4, 4:2:2, 4:2:0, etc. etc.)
The 4:whatever --> 4:4:4 conversion makes a difference. Most NLEs default to using box reconstruction... which looks boxy and won't work well for keying. So "chroma smoothing" type filters can help. Some keyers have that built-in.
2- If you want a good clean key, key your footage in its original resolution... so if your AVC HD is 1920x1080, then stick with a 1920x1080 project. Depending on your keyer + codecs, chroma smoothing might help.
Low noise is also very important. Unfortunately you are limited by AVC HD, but if you can do things like keeping gain at 0 and making sure there aren't shadows on the screen then that will help.
There's different keying techniques to fix bad keys. It depends somewhat on what software you're using.