DV and blue screen question

[Brian Ellis]

I'm currently editing my holiday video and, for the first time, wished to use a blue screen matte for a sequence. Having heard all sorts of dire comments about blue screen (as opposed to green) on DV footage, I was rather surprised to find that it came out perfectly, not the slightest trace of jaggies, bleeding, overlapping, fringing or whatever, even when examined very critically, frame by frame, full screen. I'm so "déçu en bien" that I'm inclined to ask what went wrong. Now, those who complain, I think, are possibly all from the NTSC side of the pond with a 4.1.1 colour depth. Is the fact that the PAL 4.2.0 depth, the larger number of vertical pixels or the colour phase alternation likely to influence the blue screen results in a positive direction? Or am I missing something with this fluke?

Just to make sure that you have all the cards on the table, the image with the blue screen background is a still of a sitting Buddha reduced to 720 x 576 and everything except the Buddha is carefully blued. This has a moving path applied so that the Buddha appears to be zoomed in from about 40% size (280 pixels) to 100% size. This is on V3. To avoid the framing at <100%, I also put a blue dummy frame, also chroma-keyed out, on V2. I applied similarity of 30% (there was nothing blue in the shot, whereas the Buddha was green jade, which was why I didn't use green) to avoid any blue spots in the background (I guess my painting wasn't quite perfect). The background (video shot) was of the temple containing said Buddha (which I wasn't allowed to video but could surreptitiously photograph). The end result is exactly what I wanted.

[Dr Mordrid]

It's easy to chromakey key a still; where keying bercomes tough is with a live subject that has things like hair and constantly changing edge details.

Also; still subjects are best done using the alpha channel (black background & saved as a *.tga w/transparancy) or lumakey, not chroma or luma keys.

That said PAL DV at 4:2:0 would be easier to key than NTSC at 4:1:1, mainly because the color sample wouldn't be spread across 4 horizontal pixels as it is with NTSC.

This is where the DV-keying jaggies mainly come into play; subject pixels bleed into the background on the right side of the key and background pixels bleed into the subject on the left side of the key.

The following should illustrate the blockiness vs. 4:2:2 analog footage;



Dr. Mordrid

[Brian Ellis]

Doc

I agree. In actual fact, I was following the same line of thought and saved in TGA, but the proggy I was using for painting didn't allow transparency in that format, which, to save time, I did in chroma key. Luma key would have been difficult as the subject had high contrasts (reflections off the polished jade to deep black shadows), which woulld have required a contrast adjustment.

I take your point about moving hair. I might try it some time when (if!) I have a few moments to spare.

"The following" did not come through

[Dr Mordrid]

Didn't come through when you d/l the page? Hmm....

Try downloading it directly: http://digitalvideo.8m.net/analog-dv.jpg

Dr. Mordrid

[Brian Ellis]

Interesting, Doc, thanks. I wouldn't mind seeing such a comparison for 4.2.0 as well, out of curiosity.

[Dr Mordrid]

Love to generate one, but don't have a PAL camcorder.

Using 4:2:0 MPEG-2 wouldn't be the same as there are quantization differences as well as a slightly different method of implementation;

4:2:0 MPEG-2:



4:2:0 PAL DV:



Dr. Mordrid