Improving Hard-Telecined Film Material

[Smithee]

I have some Turner Classic Movies (TCM) MPEG encodings done with WinTV over the years. My credo is that it just isn't worth messing around trying to improve such things if the titles are available on DVD. But I have some titles that have never been released on DVD, and would like to enhance them the best I can.

I've been running the titles through the QTGMC deinterlacer. Since the encodings are all hard-telecined, I deinterlace and then decimate every 5th frame to create a 23.976fps master AVI that TMPGEnc adds pulldown flags to for the final MPEG.

Being a newbie, I've experimented with additional noise reduction and using Convolution3D.

Can anyone recommend a work flow (i.e. additional filters) to best enhance these encodings?

Although the results with QTGMC have been impressive, the biggest problem is flickering during pans or other camera movements, especially when things on-screen have certain patterns (e.g. a brick wall in the background or a shirt with fine lines). Of course this isn't a byproduct of QTGMC but the original interlaced source itself. I've done some VHS encodings using the same process but notice that flickering is almost non-existent for some reason.

I'm still very new at this and would appreciate any feedback from other film buffs who run their TCM encodings (or other such material) through a process which produces the best results.

[Didée]

A good start would be to post a representative sample of the source.

Note that QTGMC is - theoretically - not suited to "deinterlace" hard telecine. Of course it can turn out quite okay ... but as I have stated several times in the past, the employed method of temporal pre-processing does only fit to "video-type" interlacing, but is technically inadequate for telecined material. See here (http://forum.doom9.org/showthread.php?p=1478442#post1478442) for my latest try to explain the problem.

[Smithee]

A good start would be to post a representative sample of the source.

Note that QTGMC is - theoretically - not suited to "deinterlace" hard telecine. Of course it can turn out quite okay ... but as I have stated several times in the past, the employed method of temporal pre-processing does only fit to "video-type" interlacing, but is technically inadequate for telecined material. See here (http://forum.doom9.org/showthread.php?p=1478442#post1478442) for my latest try to explain the problem.

What I want to do would be applicable to any film source broadcast on TCM. That is, hard-telecined, usually black and white. So what I'm looking for is somewhat of a 'blanket' set of filters that would be used no matter the TCM material.

I've read your linked post and undertand the core of what QTGMC does, and that it is best suited for purely interlaced material. But after using it, recommended by people on this forum to do a tricky PAL to NTSC conversion, I threw it at some TCM material. This material was encoded directly to hard drive, not recorded to VHS and then encoded to MPEG. I consider myself to be picky about video quality, with a keen eye to some things the average viewer wouldn't even notice. The results with QTGMC on the few encodings I've tried have been impressive.

Below is what I used:

MPEG2Source ("P:\Film5_TCM.d2v")
QTGMC(Preset="Slow")
SelectEven()
Decimate(5)
Removegrain (mode=17)
Convolution3d (preset="movieHQ")
Greyscale()

As another example, I have a VHS that was encoded to MPEG with WinTV. Although it's a commercial studio release, the quality is quite terrible. (My VCR doesn't have a timebased corrector, which would be nice to have for such encodings.) I threw QTGMC at it--the before and after results were amazing, probably because the 3:2 pulldown was so out of pattern.

Although not suited for such material, QTGMC is performing some video magic. But if there are better ways to perform this magic, I'd like to know. It's just that I thought people out there would have standard or semi-standard ways of processing hard-telecined TV and VHS sources.

Now in reference to my original post about flickering during pans with certain patterns, below is a 4-second interlaced, unprocessed clip. Look at the vertical lines on the garage in the top-right corner:

http://tinyurl.com/3j9ymxc

I doubt there's much that can be done with such things.

[mp3dom]

While the source is quite terrible, the hardtelecine is fixed at 3:2 pulldown (at least with the provided example). QTGMC (can) alter the image to some extent (depending on the source, amount of grain etc). If you want to only have the 'original progressive source' (and add other filters later) you can simply use TIVTC+TDecimate (providing an override file if needed). If you want to deinterlace without losing too much quality you can use Tdeint with tryweave=true and full=false. In this way TDeint firstly try to 'comb' the actual frame with its previous (or next) frame. If the frame is detected as not interlaced then it is returned 'as is', otherwise it deinterlace. For telecine material this can help.
The image is already screwed up, you can't do a lot about vertical lines on the garage (there are also other details on the car that get lost during panning)

[Smithee]

While the source is quite terrible, the hardtelecine is fixed at 3:2 pulldown (at least with the provided example). QTGMC (can) alter the image to some extent (depending on the source, amount of grain etc). If you want to only have the 'original progressive source' (and add other filters later) you can simply use TIVTC+TDecimate (providing an override file if needed). If you want to deinterlace without losing too much quality you can use Tdeint with tryweave=true and full=false. In this way TDeint firstly try to 'comb' the actual frame with its previous (or next) frame. If the frame is detected as not interlaced then it is returned 'as is', otherwise it deinterlace. For telecine material this can help.
The image is already screwed up, you can't do a lot about vertical lines on the garage (there are also other details on the car that get lost during panning)

Actually the print itself is very clean for a film from 1931. The 4-second clip was shot a night, or was a day for night shot.

I tried TDeint as outlined by you, and appreciate the feedback. But the output really isn't any different than the initial tests I had done deinterlacing with Decomb--at least not in the short time spent with TDeint.

Again I go back to QTGMC and am impressed with what I see.

Here's a variation I tried:

DGDecode_mpeg2source("P:\Film5_TCM.d2v")
TDeint(tryweave=true, full=false)
TDecimate
QTGMC(Preset="Slower", InputType=1)
Removegrain (mode=17)
Convolution3d (preset="movieHQ")
Greyscale()

Here the input type is set to process a progressive source, and it still does wonders.

Now if I don't use QTGMC, what other filters could I use later? This really is the core of my question--what other filters do people generally use after inverse telecining the material mentioned in this thread?

Perhaps QTGMC could be updated to include a mode specifically after inverse telecining?

[Didée]

The usage of QTGMC in this 2nd script is OK. After IVTC the video is progressive, and QTGMC's InputType=1 (or 2, or 3) deals correctly with progressive input. The issue is only with the default interlaced mode (InputType=0), when used on a telecined (or progressive, or progressive phase-shifted) source.

[Smithee]

The usage of QTGMC in this 2nd script is OK. After IVTC the video is progressive, and QTGMC's InputType=1 (or 2, or 3) deals correctly with progressive input. The issue is only with the default interlaced mode (InputType=0), when used on a telecined (or progressive, or progressive phase-shifted) source.

Not to mention that it's 5X faster than my previous tests. Of course this is because there are less frames to process with the decimation and, more importantly, less overhead for QTGMC since it's not doing as much with a progressive source. But what it's doing is what I want to see.