NTSC DV problem - 4.1.1 color to 4.2.0

[Apulo]

Whatever route I seem to follow (direct hardware MPEG2 capture, or DV capture and software encoding) I keep bumping into problems

I was considering buying a Canopus ADVC50, but now it seems there's a problem with NTSC DV, as it is capturing at 4.1.1 color resolution. When encoding this to MPEG2, which uses 4.2.0, it seems this introduces a great deal of color resolution loss.

I read it was possible to somewhat improve on this using a program called Avisynth and a filter, but this would only minimize the impact, not fix it, and more importantly, Avisynth seems to have an incredibly steep learning curve. It's script based rather than have a userfriendly interface, and all help files and manuals I found on the net seem to take for granted one has experience with scripting, programming and all technical aspects of video.

So, should I abandon the DV-then-software-encode route (not worth the trouble & filtering no good), or do such filters also exist for VirtualDubMod, or one of the Ulead editing programs (don't own one yet but Videostudio seems nice and has a good MPEG2 encoder built-in as well as an AC3 plugin available)?

Other alternatives for not-so-high-end PCs? The Canopus DVstorm or equivalent Matrox cards sound nice but are beyond my means.

Thanks!

Apulo

[Brian Ellis]

It is true that NTSC DV uses a colour space of 4:1:1, co-sited, and PAL DV 4:2:0, co-sited. It is also true that DVD MPEG-2 uses 4:2:0, non-co-sited. In both cases, therefore, the YCbCr signal is re-sampled and the losses are identical to the eye. I've used a Canopus ADVC-100 for nearly two years and I have no problems with this. The resampling is done very well in both standards. The only problems that may appear would be a slightly worse fringeing when using an alpha channel, especially blue screen. For this reason, many DV users in NTSC land tend towards green-screening, where it is less problematic.

What you are implying is that the DV-25 format is no use to man nor beast. Yet literally millions use it well every day, from the veriest amateurs to full-blown professionals. Are they all wrong? No! Because the eye's chroma resolution is actually far worse than the DV's. When I was a student, my dissertation was on colour TV, in 1951. I demonstrated this by taking three photos of a bowl of apples, with a black background, two on Kodachrome, one on Ilford FP2. The first was an ordinary monochrome shot with 2 lights at ~45°, showing the relief. The second was a straightforward colour shot and the third used flat lighting with a half stop underexposure. I projected the 1st and 2nd shots side by side on a screen and then projected the 3rd one to superimpose over the first one. The appearance between the superimposed one and the plain transparancy was minimal, after adjusting the light levels with a Variac. So, I had kinda sorta separated the chrominance and luminance. I then deliberately put the chroma projector well out-of-focus. What happened? Nothing visible, nothing at all. Switching off the luma projector, all that was left was a vague mass of colour patches showing the red and green of the apples but without any form. Switching on the luma again, the apples sprang back to life. I used this as a demonstration that the eye cannot resolve colours in the same way as it can the luma, actually in the ratio of about 1:6. So, two sets of chroma data for every 4 luma signals is well within the tolerance of the eye and one would observe perfect (within the limitations of the rest of the video chain) colour reproduction with either 4:1:1 or 4:2:0 except with artificial manipulations, such as blue screen.

Do not be bamboozled by theoretical science: it is what the eye sees on the TV screen that counts. IOW, pragmatism should reign, especially if you do not have access to fully pro equipment with which you would possibly work in 4:2:2 colour space if you had to do some fancy effects.

[Apulo]

Thanks for the explanation Brian. But you are missing part of the question. I agree that for effects & multigeneration copies 4.1.1 is good, I read on many sites it's better than 4.2.0. I'm absolutely not trying to imply 4.1.1 is useless. But I'm not talking about effects or multigeneration editing - the usual advantages of using DV over direct MPEG2 captures - I'm talking about capturing in DV4.1.1 and then encoding to MPEG2 4.2.0.

From the many sites I've seen about this they all without exception say one _must_ use a special filter for going from 4.1.1 to 4.2.0 (Avisynth filter called ReInterpolate411). The pictures they posted showed a very clear difference (big jaggies on color edges; filter removed those near perfectly). I'll try and find the appropriate links but I'm sure a verteran like yourself already knows the Doom9.org website and its excellent (but not easy to get into as a newbie) forum.

The laserdisc backups I'm going to make should be as good as I can afford - there may never be another chance to make backups. So if a simple filter can visibly improve quality, I guess I'll try it, unless other options are found.

Do videoediting applications like Video Studio have such a filter built-in, or does anyone know if it can be purchased separately? I've been googling a lot but it seems there's no VirtualDub version of ReInterpolate411.

Apulo

[Brian Ellis]

For me, the criterion is how it looks on a TV, not a theoretical discourse of how an image looks at 500% magnification. I maintain that 4:1:1 colour space DV, encoded with a good encoder, to DVD-compliant 4:2:0 MPEG-2 at a reasonable bit rate (say, 6000 CBR) will look just fine, jaggies or not. Thousands are doing it. By definition, ANY filtering, will lose you data: you cannot generate data that is lost so that, even if you convert DV 4:1:1 to RGB 4:4:4, you can not gain the slightest iota of data. In fact, the weakest link of the chain is not the DV, but the MPEG-2, if it is DVD-compliant. So, as I said before, think pragmatism and believe your eyes viewing the end result.

There are always those who think that a chain becomes stronger by adding steel to the strongest link and ignoring the weakest. Think holistically.

[Apulo]

My fault I think, not explaing well enough. The Redistribution411 filter does not filter as such, it redistributes the colordata more evenly so that the vertical color resolution is increased, removing jagged edges. It's plainly visible on normal screens, but perhaps not as much on interlaced as on progressive screens. Not sure if you actually have a loss in this "filter" type.

A friend will drop by with a ADVC100 box this weekend. I'll try it out and see what unfiltered quality looks like. Filtered... not this weekend... Avisynth's manuals still seem mostly chinese to me

Apulo