DVdoctor.community - View Single Post - Urgent help needed with Correct editing codec -XLH1

GlennChan · 8th June 2007, 23:00

Hi Alan,

I'm wondering... wouldn't the best scheme be one that avoids unnecessary re-sampling?

1- Suppose you shoot DV. Pixel shifting 3CCDs effectively oversamples the image (right?). The camera has to downconvert this to subsampled DV... i.e. 720x480 (for NTSC). Unfortunately, a prosumer camera won't implement high quality downsampling. But this is the only place where we need to do it.

Ingest via firewire into a 720x480 project. Any image processing doesn't have to re-sample the luma. Yes, the chroma may need to be re-sampled unless you render to a 4:4:4 intermediate codec (i.e. sheer? cineform?). Granted, right now most NLEs don't do this (or aren't good at doing so). But this approach avoids having to re-sample when you apply image processing.

1b- Your final delivery format will likely use chroma subsampling (i.e. 4:2:2 SDI to dbeta, MPEG2 / DVD / digital broadcasting, web delivery). In practice, triangle / linear downsampling produces good results. Because you are downsampling chroma, you can get away with a lot more. Chroma aliasing just isn't that visually noticeable (in my tests) for real world footage (color zone plates don't count).

An in practice, many codecs use box or point sampling on downsampling chroma (e.g. Avid DNxHD, Apple's H264 seems to do this). This looks terrible (visually noticeable) and is the worst way you can downsample chroma. But in practice not that many people complain.

2- With a RAW approach with offline/online editing, you can get good performance on desktop systems while you offline at low quality (with fast/cheesy resampling). And then do slow, good quality resampling for online.
(Though RAW isn't exactly always practical and widely supported.)

3- In other systems, going full raster and avoiding subsampling (luma or chroma) makes sense. For example, for HD staying 1920x1080 throughout the chain makes sense. The capture format should be full raster 1920x1080 4:4:4 (i.e. unlike HDCAM, HDV), editing should use 4:4:4 intermediates, and delivery should be on a full raster HD format. Broadcast is ok if it's 1440x1080... you can implement high quality filters there.

8th June 2007, 23:00	#20
GlennChan Member Join Date: Jan 2007 Location: Canada Posts: 46	Hi Alan, I'm wondering... wouldn't the best scheme be one that avoids unnecessary re-sampling? 1- Suppose you shoot DV. Pixel shifting 3CCDs effectively oversamples the image (right?). The camera has to downconvert this to subsampled DV... i.e. 720x480 (for NTSC). Unfortunately, a prosumer camera won't implement high quality downsampling. But this is the only place where we need to do it. Ingest via firewire into a 720x480 project. Any image processing doesn't have to re-sample the luma. Yes, the chroma may need to be re-sampled unless you render to a 4:4:4 intermediate codec (i.e. sheer? cineform?). Granted, right now most NLEs don't do this (or aren't good at doing so). But this approach avoids having to re-sample when you apply image processing. 1b- Your final delivery format will likely use chroma subsampling (i.e. 4:2:2 SDI to dbeta, MPEG2 / DVD / digital broadcasting, web delivery). In practice, triangle / linear downsampling produces good results. Because you are downsampling chroma, you can get away with a lot more. Chroma aliasing just isn't that visually noticeable (in my tests) for real world footage (color zone plates don't count). An in practice, many codecs use box or point sampling on downsampling chroma (e.g. Avid DNxHD, Apple's H264 seems to do this). This looks terrible (visually noticeable) and is the worst way you can downsample chroma. But in practice not that many people complain. 2- With a RAW approach with offline/online editing, you can get good performance on desktop systems while you offline at low quality (with fast/cheesy resampling). And then do slow, good quality resampling for online. (Though RAW isn't exactly always practical and widely supported.) 3- In other systems, going full raster and avoiding subsampling (luma or chroma) makes sense. For example, for HD staying 1920x1080 throughout the chain makes sense. The capture format should be full raster 1920x1080 4:4:4 (i.e. unlike HDCAM, HDV), editing should use 4:4:4 intermediates, and delivery should be on a full raster HD format. Broadcast is ok if it's 1440x1080... you can implement high quality filters there.