Deinterlace video

[simdavid]

Hi to all,

I just discovered a way to make briliant deinterlacing video. Well, I'm sharing this method of mine which is IMHO a good way for me. It might not suit everybody of course. However, it's the spirit of sharing this with video capturing pals. Hopefully, you find this useful enough for your video! I would really love anybody who had more better way of deinterlacing or improve methods over mine would share. Well, I also have follow and tried what is suggested in the www.100fps.com site.

Firstly..some background, I convert all my video capture to SVCD mainly. I have a Chinese series of video tapes that I want to put on SVCD. This is a martial art chinese series, so the actions and movement is very high. At first, I convert to SVCD using TMPG and CCE 2.64 without deinterlacing..I wanted to output what is close to original without applying filter. However due to the low bitrate of SVCD

(I was using 2616 kps for video and 128 kps for audio) for 37 mins per CD (650mb). The bitrate wasn't simply high enough to output the video without MPEG artifacts. So the square blocks is very apparent during the fighting scenes. However, in low motion scenes, the output was brilliant. This is not good enough for me becos the
fighting scenes was where it makes the show interesting. So I'm searching for a way to preserve the quality.

I found out that I can't do it in SVCD format. Unfortunately, I dun stay in the U.S and DVD burners and DVD-R is a luxury item in my country. I can only make do with SVCD at this moment. So begin my quest for achieving the quality I wanted.

After experimenting for 3 months. I had many failures and had burn alot of CD-R just for testing becos I want to watch on standalone DVD player.This is what i do now.

I capture at 480*480 resolution 29.970fps for NTSC. This is for SVCD requirements. However, by doing so, we are capturing two field of the video and all information from the video is capture. Using virtualdub and it internal filters. I use the filters in order of sequence:

1. Deinterlace (Blend both field best)- This will preserve the 2 fields in the video.
2. Resize (352*240) (352*288)

Now, this is crucial for me. The reson I resize was that after much thinking about the relationship between BITRATE and RESOLUTION. I understand that the higher the resolution the more bitrate is needed to cover per MPEG pixels to preserve quality.So for SVCD resolution of 480*480..2616 kps wasn't simply enough to produce the quality at that resolution. So the trick I used for to downgrade the resolution and solve the MPEG artifacts but at the same time keeping SVCD bitrates which my standalone DVD player can accept. IMHO that DVD are fault tolerances nowadays..It sort of can accept XVCD format. So I decided to go 352*240, 2616 bitrates. I did get what I want..the MPEG artifacts is like 3% IMO. I mean, I can watch the movement of the legs and hands moving during the fighting scenes more clearly than what I did before. The Drawback of this method was that VCD wasn't as sharp, colour wasn't as vibrant as SVCD becos of lower resoution. For that matter, I use virtualdub fitlers.

1. Sharpen (20) values which I used - You can play around to adjust to your liking
2. Smoother (10)
3. Brightness/Contrast

The reason why I deinterlace was becos since I'm converting to VCD format. The encode is progressive frames in TMPG and CCE. Furthermore, encoding interlace lines without enough kps..the interlace lines IMHO interfere with picture information in the MPEG block pixels. Thus giving more square blocks in the end. I did experiments with interlace and deinterlace from same source and
bitrates in SVCD format. If I give high enough bitrates like 6000kps...the SVCD output is very solid like a rock with interlace!

The deinterlace filter in virtualdub is absolutely fabulous. The video doesn't lose any frame information at all..since both frames are still being displayed.

The result I get was near SVCD quality. At least it does suit me now. I belive this method is also suitable for conversion to DivX and DVD rips. Hope it helps!

So this is my experience :-)

[Jehovantodistaja]

At first it was promising, but after the resize the video quality got really bad. I captured from an analog source...

[Valky]

Hmm..actually you are not doing proper svcd because your audio is something non-standars for that..at least for my opinion.

And secondly isn't this de-interlade filter really necessary..if you capture x480 and after that resixe it to x240 you are doing true de-interlace there anyway..?

[dilly]

By discarding one field.

[dilly]

@poster

Doesn't SVCD natively support interlaced video? And if it does, and your destination is a set-top box, why would you deinterlace?

[simdavid]

The reason for deinterlacing.


encoding interlace lines without enough kps..the interlace lines IMHO interfere with picture information in the MPEG block pixels. Thus giving more square blocks in the end. I did experiments with interlace and deinterlace from same source and bitrates in SVCD format. If I give high enough bitrates like 6000kps...the SVCD output is very solid like a rock with interlace!

Well, I found from my experience that If u encode an interlace video to a SVCD and if the bitrates is not high enough. The end result is a lot of square block on that particular scene which bitrate is shortfall. I deinterlace using blending in VirtualDub..I manage to reduce the squares blocks without altering any changes of my encode setting in CCE. So the only variable that change was that I Blend the 2 field together.

Conclusion, blending the 2 field together I thinks make the encode easier for CCE to derive picture information per 8x8 blocks.

For SVCD audio..you can use from 32kps to 354kps go to http://shelob.mordor.net/dgraft/ for more information.

[lemon]

But when deinterlacing, you are going from a 60fps video to a 30fps one.
I these martial arts movies where originally film, they should be 24fps, and you should inverse telecine them.
If they were originally made for TV, then you will lose half the temporal resolution when blending fields, Also, as the two fields are in different time moments, you will have some blurring in the images.
Try to encode them in half D1 resolution (352x480). Perhaps your player is able to play them, then you'd have more bits per pixel than in SVCD while conserving full 50fps.

[bb]

Originally posted by dilly
By discarding one field.
No, discarding a complete field would happen if you captured at the lower resolution (352x288 or 340x240). Capturing at full (at least full vertical) resolution, then resizing to a lower resolution results in better quality, because the two fields get averaged, and the downsizing reduces noise.

The suggested CVD type encoding (352x480 interlaced) will be blockier than the progressive low res (352x240 (why not 340x240?), 352x288), because there are twice as many pixels to encode, and up to 30% quality loss due to interlaced encoding. But it's worth a try. I wouldn't recommend interlaced encoding for SVCDs, but with the lower res CVD format you can get pleasing results.

bb