Strange deinterlacing with PAL StarTrek DVD

[JimiK]

Hello,
I couldn't find an answer to my problem, so I hope this hasn't been asked before. I'm trying to compress a DVD with StarTrek episodes. The material is interlaced. Now it's the 3rd ST DVD I ripped, but the first one the problem occurs. For the other DVD's I used TomsMoComp and it worked perfect. With this DVD I tried Decomb with FieldDeinterlace(), TomsMoComp and even VerticalReduceBy2(). That's were I encounter something very strange. As far as I know, VerticalReduceBy2 should give you every second line and every interlacing effect should be gone. But at scenechanges there are still parts of both pictures visible when doing it. Then I turned to the VDub filters and the deinterlacing with Discard Field 2 worked perfect. Also Donalds SmartDeinterlacer worked fine. I'm not at home, but I could post my Avisynth script later and also a screenshot where the effect is visible or even a small part of the Vob where the problem occurs.
Best regards,
JimiK

[hakko504]

VerticalReduceBy2() will not give you every second line, for that you need Separatefields.SelectEven !VerticalReduceBy2() will give you an average of the two pixels on adjacent lines. This means that it is a sucker for creating ghosting, and should be avoided at all possible means. I use it for captures, when I capture at 352x576 and after deinterlacing when I want the final size to be 352x288.

[JimiK]

Tack så mycket! (Thank you very much!)
That explains a lot. Now I have a question about TomsMoComp. You can set the field-order (topfirst or bottomfirst), so I guess it does not just blend those fields together. As I understand it, it looks also at the previous frame and tries to put the lines of the two frames together. When I'm choosing the right field order, then I don't see two scenes in one picture, but there seem to be wrong details and the image is very blurred before a scene change. When watching the movie you can't see the blurring in the last frame before the change, but it's still not right and I would guess the codec has to use more bits, because there is a change to the previous pictures, even though it's before the scene change. And: why does smart deinterlace work, but decomb is giving me strange results, no matter what settings I use. Don't get me wrong, I appreciate the work of all those graphic gurus. And I don't have the slightest idea how all that works. It's just that Decomb is a later piece of software than the smart deinterlacer and still fails where smart deinterlacer works. So I wonder why.
Sincerly,
JimiK

[hakko504]

First, have you checked what the original picture looked like before any deinterlacing? There might be strange things encoded from the beginning and that you can't do anything about.
Other than that I think the difference between fielddeinterlace and smartdeinterlace is that field tries to determine if a frame needs deinterlacing at all before going to work, whereas smart will deinterlace all pictures.
Quite frankly, since I usually go for quite low resolutions I almost always use Greedy for deinterlacing. That way I don't get any ghosting, and it looks good on both TV and monitor.

[JimiK]

Thank you again,
I watched my source more closely and you're absolutely right. In this special scene, where the effects are obvious, the picture gets worse before the scene change. There is already a bluring and something I would call aliasing in the mpeg2 source. The bluring effect seems to be increased by the deinterlacing filters. Greedy does the best job here, followed by TomsMoComp. I also watched the scene with smart deinterlacer and now there are artifacts with the exact setting I used before. But that's not so important if Greedy is doing the same job. It may be slow, but VDub filters are also not the fastest (o.k., not the filters are to blame, but the colorspace conversion).
Btw, is there a newer version of Greedy than 0.4.0.0? I just downloaded it from Tom's page, but maybe he keeps newer versions in a secret place ;)
Thank you,
JimiK

[trbarry]

Sorry, no newer version of GreedyHMA. I started working on one but then got involved with some other things and never got back to it. I haven't touched it in awhile.

- Tom

[JimiK]

Thank you for the answer Tom. It's not that your filter didn't work well, I'm just one of those guys that always want to play with the newest toys :)
Best regards,
JimiK

[hakko504]

Considering the current development of YV12 support in AviSynth 2.5a, is there a possibility of a Greedy version that works directly in this colorspace?