576 to 480 line conversion

[DEmberton]

I'm capturing PAL broadcasts and converting to NTSC (SVCD), so I need to resize from 720x576 to 480x480. I've generally just let TMPEng do this, which it does fine, but the results always look slightly out of focus, and I'd prefer a sharper image.

What I'm wondering is there an optimum way of doing this? These are originally US shows, so that possibly means the master was actually 480 lines and has been upsampled to 576 in the PAL conversion. So is there a process that can reverse this without blurring the image further? I was thinking maybe I'd resize to 720x480 first using one method, and then resize to 480x480 second using a different method and feed the results to TMPEng.

I'll experiment, but does anybody have any suggestions?

Dave

[flaninacupboard]

i'm afraid that when ntsc video is converted to PAL video, it becomes, in technical terms, completely shagged. there's five frames removed from the source every second, and there's no way to put them back.
to be honest, you'd do a lot better keeping it as an interlaced PAL SVCD than trying to convert to NTSC. but as ever, if you find a way that gives decent quality, let us know.

[DEmberton]

Thanks, but no . This is actualy progressive film sourced material that has been converted to PAL by speeding up from 23.976fps to 25fps, and I'm converting back to NTSCFilm format by slowing it down. So there's no issue about missing frames or interlacing, just of resizing the image in the best way possible.

Dave

[flaninacupboard]

it wasn't very clear given that you said "These are originally US shows, so that possibly means the master was actually 480 lines and has been upsampled to 576 in the PAL conversion." as US shows even if portions are 24fps film, portions will also be 29.97 video, so converting back is not an option. if you're sure it's all sped up 24fps material then that's fine, and any bicubic resize filter will be fine, it makes no difference if you do this in one stage or two.
i would more likely attribute your blurry image to either: recompressing a lossy capture (mjpeg or divx converted to mpeg) or the low bitrate of SVCD.

[DEmberton]

Yes, sorry I suppose that was kind of misleading. These are definitely film sourced sped up versions.

I'm using a DTT box to capture the actual MPEG2 Digital TV transmission (so not really a capture at all) as the source, which certainly could be better quality.

Obviously the low bit rate and resolution of SVCD won't help, but I've seen SVCDs that look better than mine so I'm just trying to optimize what I have. I guess I just need to experiment.

Thanks for the reply.

Dave

[flaninacupboard]

yeah, all the digital cable transmissions i have seen are kinda ugly...
they all seem to use real time encoders with little or no motion compensation, oh how different it would be if i worked for them....
the best thing i can suggest is a subtle noise filter, as those transmissions aren't pretty. settings of say 10 1 16 will barely be noticable to the eye, but it'll make a big difference to TMPGenc.

[DEmberton]

Did a load of experiments, and I think using the BiLinear resize in Virtual Dub then feeding that into TMPEng looks marginally better (sharper) than letting TMPEng do the resizing itself.

I've also put in a subtle noise filter as you suggested (actually using 3,1,2 as the settings which is very subtle), and that seems to improve things a bit too. I've seen SVCDs that have been de-noised, and they always look a bit artificial to me, so I don't really like that effect, but these settings seem just enough to stop the constant background noise I was getting before.

So I think I've improved things, although still not as good as I'd like. The limitations of the source is probably the problem.

Dave