Sir Didymus
14th January 2006, 18:37
In the doom9.it forum I had some discussion with a couple of gentlemen about the way adopted in DVD-RB for its LB 4/3 --> 16/9 conversions.
Just to summarise, for interlaced material the method adopted in DVD-RB is to use, for each segment to convert, the following avisynth process:
1. SeparateFields
2. LanczosResize [25% of stretch; for PAL material the vertical resize is from 216 to 288 lines]
3. Weave
From my experience, I never had bad results using it, and some careful visual inspection to the resized clip left me happy about the actual method.
Nevertheless, reading this post (it is almost exactely one year old...), pointed out by the friend midnightsun:
http://forum.doom9.org/showthread.php?s=&threadid=88116
especially on page 2, where some nice comments from Mug Funky, IanB and Scharfis_brait were kindly posted, it seems that:
a. the actual method is introducing a vertical misalignement given by (OldHeight/NewHeight-1)*0.5*(NewHeight/OldHeight), resulting in total 0,16667 pixels for PAL material [if my calculation is right]. Do you consider this level of misalignement considerable ?
b. better procedure, based on bob() deintelacing exist. Are these procedures of simple introduction in DVD-RB ? I mean, I am not sure if a good avisynth script (general enough to be used to clips with both TFF and BFF field order) is applicable...
What do you think about ?
Cheers
SD
Just to summarise, for interlaced material the method adopted in DVD-RB is to use, for each segment to convert, the following avisynth process:
1. SeparateFields
2. LanczosResize [25% of stretch; for PAL material the vertical resize is from 216 to 288 lines]
3. Weave
From my experience, I never had bad results using it, and some careful visual inspection to the resized clip left me happy about the actual method.
Nevertheless, reading this post (it is almost exactely one year old...), pointed out by the friend midnightsun:
http://forum.doom9.org/showthread.php?s=&threadid=88116
especially on page 2, where some nice comments from Mug Funky, IanB and Scharfis_brait were kindly posted, it seems that:
a. the actual method is introducing a vertical misalignement given by (OldHeight/NewHeight-1)*0.5*(NewHeight/OldHeight), resulting in total 0,16667 pixels for PAL material [if my calculation is right]. Do you consider this level of misalignement considerable ?
b. better procedure, based on bob() deintelacing exist. Are these procedures of simple introduction in DVD-RB ? I mean, I am not sure if a good avisynth script (general enough to be used to clips with both TFF and BFF field order) is applicable...
What do you think about ?
Cheers
SD