Jump to content
Sign in to follow this  
Guest joconnell

Quick post on keying DV

Recommended Posts

Guest joconnell

Heya Folks,


someone was asking about shooting greenscreen stuff with dv on another site and I posted a reply about some of the issues associated with it - It may be useful for someone else here so I thought I'd stick it up here.


Make sure you use progressive scan mode on the camera - it'll make your pictures look better and be handier for effects work later.


Short answer:


Dv is bad for greenscreen because of the way ot stores colour - the colour is low quality so it gives you bad edges unless you use a keyer that can take this dv problem into account - get a copy of dv matte pro from dv garage and it'll work nicely.


Long answer:


The bad pixelated edges you're seeing isn't down to the lighting, it'd an inherent problem in dv in the first place. When you deal with digital images, the information is stored in 3 different colour channels, red, green and blue or rgb. When you shoot stuff on tape, the information is stored as brightness and two seperate colour channels - this is commonly known as yuv.


The problem with dv is what quality the information in each channel is stored at. If you've ever heard the term 4:2:2 then that's what is relevant here. The numbers are a ratio to indicate the quality that the three different video channels are stored at so it goes brightness : colour : colour in this case. The number 4 at the start means full quality which means your brightness channel is stored at full resolution (720 x 576 in pal land). The second number 2 is half of 4 so that's indicating that the colour channel is stored at half the resolution of the brightness channel which gives us an active resolution of 360 x 288. Again the last 2 is the exact same half resolution.


So this means if a video format saves in 4:2:2 you get full quality brigtness information and half res colour.


Here's our major problem. DV is 4:1:1 so it means that the colour channels are quarter resolution. It's not a problem in most cases as people dont often shoot for green screen on dv. If you do however, keying is largely based on the difference between the uniform background colour and the colours present in the foreground and since the quality of the colour information is so blocky and pixelated in dv, you get crisp detail in the footage but a really pixelated edge which is often confusing for people.


There are a few ways around this which are dependant on your compositing program. What they normally do is apply a small blur to the colour channels to remove the pixelation before trying to key the background which will make the edge of the final key a little less blocky. In a program like shake this is easy enough as you can work in yuv space quite happily but in something like after effects you can so you'll need a keyer that can do the internal conversion for you. There are three that I know of, dvmatte pro from dv garage, primatte 3 from red giant software and ultimatte advantedge. Dv matte pro is specifically for dv and the cheapest of the lot - it's fairly friendly to use and gives good results. Primatte and ultimatte are both generaly purpose keyers and give a tonne of control and really high quality results - they aren't aimed directly at dv footage but they do have a preprocessing function to sort out the dv colour problem.


interlaced vs progressive scan


Here's a boring explanation of progressive vs interlaced and why it makes a difference based on that:


If you are sizing the video down at the end of the day then it counteracts the quality problems somewhat but if you were going back out to dv after doing all of the effects at the same size you started with you have to be a lot more careful.


The progressive mode is always a benefit as it'll make the footage look a lot more filmic for one main reason. When video records footage, it can store the images in one of two ways, interlaced or progressive. Interlaced is the more common format and available on all video cameras. Pal video is 25fps pr 25 full pictures per second but that isn't necessarily the way that the pictures are stored. Old displays weren't quick enough at redrawing images to make video look smooth so to counteract that, video broke each full picture into the odd lines and the even lines. If you take a standard pal frame, it's resolution is 720 x 576 so 576 horizontal lines make up the picture. With interlacing, this full picture is split up into the odd numbered lines and even numbered lines as shown in the picture below. So in the first field, lines number 1,3,5,7,9,11 etc are saved and in the second field, lines 2 4 6 8 10 12 etc are saved. When displayed, the screen alternates between the two fields quickly and your eye blends the two together to get a smooth motion.




Progressive scan is far more modern and since newer tvs can refresh images much faster, it's able to store full images rather than pairs of odd and even lines which means the end result looks a lot more flickery but far cripser and thus more like 24fps full frame film. Another benefit of progressive scan is for effects work.


Here's an interlaced frame:




Now, if you look at the horizontal lines, that's what an interlaced picture looks like - almost like two different images ghosted across each other. For starters, the fine lines are a pain in the ass to key and if you end up using blurs etc, it'll make a mess of it. Second of all, if your footage isn't good enough to key off and you still need to seperate it, you have to resort to rotoscoping.




Above is a small frame from star wars where someone is drawing a mask around a stick to be made into a light sabre. this is fine since it's a solid shape with an obvious outline. Say for example though if you had a fielded image like the one above - you'll have to do some kind of soft falloff to take into account those lines sticking out of the guys head which will generally never look any good. When you mix in motion blur like in the stick in the image below and add fielding on top of that it makes your life very very very very very very difficult to mask out or key succesfully




Lastly, when you resize the video, factors divisible by 2 are alwasy good but if you use anything odd like 47.63% smaller, it may have conflict with the alternating lines of fields and make your edges look a little weird.


Bottom line - progressive has cripser more filmic images and are easier to post produce.


Oh also - turn off any kind of edge enhancement in the camera - it'll make your life harder when keying.

Share this post

Link to post
Share on other sites
Guest parallax

@ Joconnel:


I think your numbers are off:


chroma sampled info on a 422 image is 360 x 576. The 50% sampling is only on the horizontal part of the image.


i'm trying to devise an automated/fast chroma upsampling process.

Share this post

Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Sign in to follow this  

  • Create New...