by Art Adams: Noise, Sampling, Codecs and Green Screen: A Potential Horror Story

[Tim Kolb]

Originally Posted by Art Adams

I’ve done a lot of testing with the Sony F55, and I’ve shot a couple of 4K projects in XAVC, but until recently I hadn’t used this camera for green screen. Sony’s implementation of XAVC uses 4:2:2 sampling, which can cause a lot of problems for green screen compositing, so when I got a call to shoot a green screen project with an F55 I decided I had to do some testing. Here’s what happened…

Read the full article at http://www.dvinfo.net/article/acquis...ror-story.html

----------------------------

I enjoy your articles, Art. Thanks for taking the time to document tests like this. Actual results are always enlightening as these codecs aren't meaningfully evaluated on spec sheets.

My first observation is that the artifacts I see along the diagonal edges on your screen grabs strike me as something I would normally attribute to more aggressive DCT compression vs color subsampling specifically.

What were the overall data rates on each format? Was the XAVC HD bitrate 90 Mbits/s vs 352 for SStP? (...obviously the camera doesn't give you infinite adjustments in this regard).

I'm curious about the assessment of SStP being the clear winner when (on your comparison pic) the edges were definitely cleaner, but I was surprised the gray block had more hue/saturation misinterpretation through the center-monochromatic area than I would think any colorist would want... However, that kind of color variation wouldn't necessarily sabotage pulling a clean key like chewed up edges do... Was that a virtue of the screen grab?

I'm certainly not questioning that 4:4:4 is better than 4:2:2 for keying... as 4:2:2 is better than 4:2:0, etc.

I find that visual artifacts from specific processes such as transforms like DCT or wavelet or color subsampling... vs a given compression process in total are sometimes difficult to unpack. Each codec can only be evaluated as a composite process obviously, but with an entire family of codec levels and profiles like MP4, we end up flipping a variety of switches to get to the various sets of "branded" parameters that constitute formats like AVC-Intra, HDcamSR, or XAVC.

Please let me know if I'm missing something fundamental about the test, but if the clear image quality winner in the comparison was working with nearly 4X the data rate and 4X the color precision (12 bit vs 10), I'm not certain how much spatial color subsampling (4:4:4 vs 4:2:2) affects the difference...

[Mike Buckhout]

Originally Posted by Tim Kolb

Please let me know if I'm missing something fundamental about the test, but if the clear image quality winner in the comparison was working with nearly 4X the data rate and 4X the color precision (12 bit vs 10), I'm not certain how much spatial color subsampling (4:4:4 vs 4:2:2) affects the difference...

I'm certainly not as experienced as Art, but my take on this is that the best results start with getting your camera to capture at the lowest-noise ISO setting possible, as that will carry through the whole capture/post process regardless of all else.

It also seems that 10 vs.12 bit is not so much of an issue here because the green background should be only one even tone all over the frame, so the 12-bit representation would only be a more precise description of the same color in 10-bit.

From there we have subsampling and compression. Again I am not the expert, but it seems to me that a better result will be had from an uncompressed 4:2:2 source than a compressed 4:4:4 source, because compression is what gives you the pixelation at the edges of differently colored areas. Of course this depends on the quality and amount of compression, so results will vary in almost every case, with the added variable of inconsistent lighting.