Reinterpolate411 : Is it necessary for all mpeg encoders?

[Wile_E]

JVC 9911 VHS --> Canopus ADVC-300 --> DV --> DVD

For NTSC DV captures, is the AviSynth filter, Reinterpolate411 really necessary for all mpeg encoders? I've been trying it with Mainconcept, CCE, Procoder Express, TMPGEnc. I get different results with each encoder.

Mainconcept has the most noticeable improvement. The video looks a lot cleaner, and edges are smoother when using this filter. A LOT cleaner! The other encoders show little or no improvement. Does this mean that the other encoders are already doing this chroma upsampling to 4:2:0 internally? Or is it because other encoders handle noise better than Mainconcept? Mainconcept just looks really noisy without using this filter.

[edDV]

JVC 9911 VHS --> Canopus ADVC-300 --> DV --> DVD

For NTSC DV captures, is the AviSynth filter, Reinterpolate411 really necessary for all mpeg encoders? I've been trying it with Mainconcept, CCE, Procoder Express, TMPGEnc. I get different results with each encoder.

Mainconcept has the most noticeable improvement. The video looks a lot cleaner, and edges are smoother when using this filter. A LOT cleaner! The other encoders show little or no improvement. Does this mean that the other encoders are already doing this chroma upsampling to 4:2:0 internally? Or is it because other encoders handle noise better than Mainconcept? Mainconcept just looks really noisy without using this filter.

Interesting subject. I need to try this filter and see what it does.

All MPeg2 encoders will convert 4:1:1 to 4:2:0 because that is the MPeg2 and DVD standard.

Others may disagree, but I don't consider 4:1:1 to 4:2:0 an "upconversion", it is a rearrangement of chroma pixels in xy space through interpolation. All this is done inside the eye's chroma blur area at 1x size. In a perfect world luminance is uneffected.

DVD style 4:2:0 is always interpolated because the final chroma pixel positions fall inbetween capture samples in h and v (x and y).

[edDV]

OK I looked at the filter and this is what I found.

"ReInterpolate411 - 2003/07/31 V 0.1.1 <== new

This is a fast simple filter to correct the improper 4:1:1 => 4:2:2 conversion that seems to occur with some DV/4:1:1 codes when used with Avisynth. It assumes the odd chroma pixels are duplicates and discards them replacing them with the average of the two horizontally adjacent even chroma pixels. It doesn't matter whether the clip is interlaced though it must be in YUY2 format for Avsynth 2.5. There are no parms, and currently no readme file. "

http://mywebpages.comcast.net/trbarry/downloads.htm

What he is doing is adding an interpolation to the "missing chroma sample" when converting 4:1:1 to 4:2:2 YUY2. Instead of assuming the "missing chroma sample" is a duplicate of the even sample, he averages the current sample with the next in order to smooth the transition to 4:2:2.

This technique is well covered in this article
http://www.nattress.com/Chroma_Investigation/chromasampling.htm

In summary, this is about decompressing 4:1:1 DV into uncompressed 4:2:2 SMPTE-259 aka SDI.

It has nothing to do with 4:2:0 although you may have discovered another use.

[cosmin]

Reinterpolate411 is necessary for the encoders that do a bad 4:1:1 to 4:2:2 upconversion themselves. (It is 4:1:1 to 4:2:2, not 4:1:1 to 4:2:0; it works in YUY2 color space, and it is the MPEG encoder that does the downsampling of 4:2:2 to 4:2:0 later on.)

By the way: as far as I know, if you have the source in DV-AVI Type 1 format, MainConcept uses its own internal DV decoder, which, at least in version 1.4.2, has a very high quality interpolator. Better than even Reinterpolate411 (...as advertised by them, but not tried by me...) If you wish, try converting your AVIs to Type 1 using the free Canopus DV File Converter tool, and see if the end result is better, or worse, or similar. I will be curious about your findings.

Others may disagree, but I don't consider 4:1:1 to 4:2:0 an "upconversion", it is a rearrangement of chroma pixels in xy space through interpolation. All this is done inside the eye's chroma blur area at 1x size. In a perfect world luminance is uneffected.

4:1:1 to 4:2:2 is upconversion, because horizontal information is fabricated via interpolation. Furthermore, 4:2:2 to 4:2:0 is downconversion, because vertical information is chopped off. Of course luminance is unaffected; this is about crippling chrominance. In the end, chrominance ends up being something like "4:1:0": fake horizontal info is added, and real vertical info is removed.

Many people call this in many ways, but calling it "rearrangement" is not really the best idea. When you rearrange something, you lose nothing, it's just that things end up in a different position. Here, it's more like a glass of juice: you spill some juice, you add some water, you end up with the same amount but it will never feel the same. When you do a smarter interpolation, it's like you add spring water as opposed to tap water: the end result will taste a little better, but still, it will never feel like the original, no matter what you do.

DVD style 4:2:0 is always interpolated because the final pixel positions fall inbetween capture samples in h and v (x and y).

Not exactly. This is only true in JPEG (in JFIF-JPEG, that is), and in MPEG-1. In MPEG-2 and MPEG-4, the final chroma position falls exactly on every 2nd luma position horizontally, and it falls inbetween luma positions vertically.