To explain why many of us have had frustrating results when it comes to greenscreening, we must start with the fact that we have been working with sub-optimal source material.
What do I mean by sub-optimal? DV, HDV, and even most HD footage is sub-optimal. This is due to how DV, HDV, and HD cameras that are in the affordable range record light (luma) and color (chroma) information. (The only exception to this is the new Red camera, which is only affordable because Jim Jannard, the president of Red, is already the multi-millionaire owner of Oakley Sunglasses and doesn’t mind creating a Hollywood-quality digital camera for about 10%-15% of the cost Sony sells there competing cameras for. The $30,000 startup range is still outside most of our readers pocket books, but it’s massive stride in the right direction compared to $250,000 to $300,000 competitors.)
Color information takes up a lot of bandwidth but is not as noticeable to the human eye as light data is. As such, when DV camera manufacturers were playing lifeboat with different pieces of information for the recording and compression codecs used in these cameras, they decided to record every pixel of light their cameras sensors picked up, but only one out of four pixels of color data for NTSC DV cameras. This was called “4:1:1” color space, with:
the “4” denoting that out of every 4 pixels, every pixel's light information would be recorded,
the first “1” denoting that 1/4 pixels would have color information recorded in the first line,
the last “1” denoting that 1/4 pixels would have color information recorded in the second line.
From here, it is compressed, but, luckily, each frame is compressed separately.
PAL DV cameras (and now HDV cameras) record with a 4:2:0 color space, which, again, records four out of four light pixels, but records color information on two out of four pixels of the first line, and NO color information on the second line. In order for HDV to fit on a single tape, multiple frames must be grouped together and compressed in clusters of between 7 and 15 frames. (7 for JVC and 15 for Sony.) Obviously, when you mash groups of frames together and then must untangle them before you can even begin to key them, this makes getting good keys harder.
HD extends from HVX200 all the way to compressed Viper footage and has a 4:2:2 color space. Again, four out of four pixels have their light data recorded, while every other pixel has its color data recorded in both the first and second line. Again the footage is compressed before being recorded to the hard drive or tape, but like DV, each frame is compressed individually, which means that there is less damage done by the compression.
So what’s the color space in those high-end cameras that Robert Rodriguez used to shoot Sin City? Uncompressed 4:4:4. Which means that every pixel is recorded for both light and color and recorded without compression to a RAID array. Obviously, this gives you optimal chroma-key work, as the keying program isn’t trying to guess where pixels are due to insufficient color information. (The aforementioned Red camera generates this same color space provided it’s uncompressed Redcode RAW stream is recorded.)
For most of us, 4:4:4 is not something we will likely have access to until the Red drops further in price, we sell one of our films for a substantial profit, or we try to hijack the signal from our cameras before its color information gets decimated. (Believe it or not, a company called Real-Stream has a $2500 adapter which they will install in a DVX100 or HVX200 camera so that you can download uncompressed 4:4:4 data directly from the camera’s imagers. Unfortunately, the program that reads this data stream is only available for Macs and, reportedly, extremely unwieldy to use. Hopefully, in the future, they will come with a cross-platform package that has the kinks worked out.)
As such, we need to figure out how to overcome the problems our undersampled cameras have with chroma-keying. In this article, we will discuss how to shoot to have the best possible footage to take into post production, which should give you the best possible keys. If you do both production and post-production well, you will be surprised by how much potential greenscreening can open up to you.
