Any suggestions to make this video better?

[chrishallowell]

Hi all,
Any suggestions to make this video better?
NTSC, 60FPS, Interlaced home video captured from analog camcorder
Camcorder ---S-Video--->DVD Recorder
http://42mb.com/v/9748809/VideoFile.m2v.html

I haven't been able to do the following successfully.... Am I expecting too much?

Deinterlace: Tried: Yadif, LeakKernelBob. Both cases: Objects are doubled in high motion areas

Destabilize: Tried: DePanStabilize. Either black around the edges or I had to zoom in yielding grainy results. Plus date/time stamp went flying all over.

Delogo: Tried: Delogo, xlogo. Results of masking the date/time stamp were poor at best.

Any suggestions on how to make this video better would be much appreciated!
My goal is to burn the edited verson back to DVD for playing on a DVD player.
Thanks,
Chris

[jagabo]

If you're going to DVD -- don't deinterlace.

You won't be able to deshake without the date and time flying around. You also won't be able to avoid black borders or zooming.

Delogo won't work well with an opaque logo.

[Soopafresh]

MVtools. Yes, it'll take forever, but your footage will look so much better. If you're using Mpeg2source, make sure to add the "Cpu=4" parameter, as in the script posted below. That will add deblocking.




Before

After


setmemorymax(512)
loadplugin("G:\benchmark\DGDecode.dll")
loadplugin("G:\benchmark\mvtools.dll")
function MVDegrain2i(clip "source", int "overlap", int "dct", int "idx")
{
overlap=default(overlap,0) # overlap value (0 to 4 for blksize=8)
dct=default(dct,0) # use dct=1 for clip with light flicker
idx=default(idx,1) # use various idx for different sources in same script
fields=source.SeparateFields() # separate by fields
backward_vec2 = fields.MVAnalyse(isb = true, delta = 2, pel = 2, overlap=overlap, idx = idx,dct=dct)
forward_vec2 = fields.MVAnalyse(isb = false, delta = 2, pel = 2, overlap=overlap, idx = idx,dct=dct)
backward_vec4 = fields.MVAnalyse(isb = true, delta = 4, pel = 2, overlap=overlap, idx = idx,dct=dct)
forward_vec4 = fields.MVAnalyse(isb = false, delta = 4, pel = 2, overlap=overlap, idx = idx,dct=dct)
fields.MVDegrain2(backward_vec2,forward_vec2,backw ard_vec4,forward_vec4,thSAD=400,idx=idx)
Weave()
}

source=mpeg2source("G:\benchmark\VideoFile.d2v",cp u=4)
mvdegrain2i(source,4,1,1)

[chrishallowell]

I get an error running the above script:
MVAnalyse does not have a named argument "dct"

Using mvtools_25_dll_20060605 from http://avisynth.org/warpenterprises/

Thanks,
Chris

[jagabo]

newer version at the bottom of this page:

http://avisynth.org.ru/mvtools/mvtools.html

[Soopafresh]

I used http://avisynth.org.ru/mvtools/mvtools-v1.8.4.zip for that example.

[chrishallowell]

Thanks! I'll try again when I get home.

[chrishallowell]

Ok, the good news: The results look great.

The bad news: It took 1.5 hours for a 20 second clip!
It will take 22.5 days to do 2 hours.

Does anyone have any clever ways to speed this up? Buying a new CPU is not an option. I am using an Intel P4 3.0. I was thinking maybe MT: http://forum.doom9.org/showthread.php?s=3413d0791caeed29a2782185e005cff8&t=94996

Anyone know if, and then how, I could incorporate MT into this script?
Thanks,
Chris

[jagabo]

Ok, the good news: The results look great.

I thought it was over filtered. Notice how all the detail in the floor and kids' hair is gone.

The bad news: It took 1.5 hours for a 20 second clip!
It will take 22.5 days to do 2 hours.

Does anyone have any clever ways to speed this up? Buying a new CPU is not an option. I am using an Intel P4 3.0. I was thinking maybe MT: http://forum.doom9.org/showthread.php?s=3413d0791caeed29a2782185e005cff8&t=94996

Anyone know if, and then how, I could incorporate MT into this script?

It might help a tiny bit with a hyperthreaded CPU. It really needs a multicore CPU.

If you want faster noise reduction you can try something like Convolution3d().

[vhelp]

Evening guys.

chrishallowell,

My first impression on your small demo left me the impression that you included a NR
setting in your dvd recorder or vhs (vcr) unit or both. As jagabo indicated also,
there is little detail (no usual tape noise) left in the capture of that demo. Thus, you
already have NR in there .. and adding more (NR/Smoothing) is removing even more.

I'm not saying that your demo sample is bad or anything. Just that I noticed it very
quickly that you had already apply a NR in their during your carpture phase .. or maybe
you didn't know it was set that way on your equipment.

Now, IMHO, I think that if you re-do your capture of that last demo sample, you might
get more detail (tape noise, etc) that might give you something work with in your next
post-process steps, after capturing.

Personally, if it were me, I would leave *all* NR (noise reduction) schemes out of the
transfer loop and just capture the source as it is, and then do the post-processing
afterwards. But, as a beginner, (and I understand this) is hard for you to understand
and utilize, because your NR knowledge (ie, avi->NR->editing->encoding) is prob limited.

Also, might help to describe your Software and Equipment, so we can get an idea of
where you might get better suggestions/tips or something.

And, for what its worth, when transfering (personal/priceless) home footage, it is
*HIGHLY* recommended that you incorporate an Analog Capture card, and capture
as close to the source type as possible -- usually as RAW yuv, or YUY2/UYVY, though
you can get by with Huffy or Lagarith if HDD space is minimal and if that your computer
is setup properly and capable of stable drop-free video captures..

** Analog Capture card
** capture codec: RAW (or, uncompressed) YUV .. ie, YUY2 or UYVY
** alternate codec: Huffy or Lagarith .. are ok
** s-video cables .. for IN <--> OUT

Unfortunately, all too many people do the wrong thing when it comes to home video
as stored on some form of tape, usually vhs/vhs-c/8mm etc., and assume that the
*grain* is the enemy. Well, let me tell ya.. it ain't

Believe it or not, its your video's detail. And you don't want your machine to remove
it with its built-in option of NR feature. Instead, leave it in there, and later on, during
your post-processing (your [..NR..] editing) and experiment with improving that area.

Try re-capturing that same demo clip and post another D/L link for us to work with,
again. Maybe it will come out better than your demo. Or maybe, not. But at least
you have some options to work with, now.

-vhelp 4396

[Soopafresh]

Try V 1.4x or 1.6x of MVtools - It is supposed to be a bit faster. MT isn't an option - it isn't compatible with Mvtools

You can reduce the strength of MVDegrain2:


fields.MVDegrain2(backward_vec2,forward_vec2,backw ard_vec4,forward_vec4,thSAD=400,idx=idx)


Try changing thSAD to 200



That's the thing about MVdegrain - Very good, but slow as molasses.

[chrishallowell]

Thanks Guys!
I will try capturing the following way: Camcorder(Hi8)----S-Video--->PC Capture card
But I can't do this until my wife gets back. She has the camcorder and is on vacation.

How would I go about capturing the video using the uncompressed option? Do I use a AviSynth script?

Software: Anything that is freeware is fair game. I'm using AvsP for AviSynth script creation and then HCGui for encoding and muxman for muxing. Then I use GUI for DVDAuthor for menus and burning to DVD.

The video clip that I posted above was captured from Hi8-----S-Video---->DVD Recorder. I have no idea what the DVD recorder did when it burnt the video to DVD....

Thanks,
Chris

[chrishallowell]

Ok, I have captured using the below without compression and in colorspace YUY2:
Camcorder(Hi8)----S-Video--->PC Capture card
http://42mb.com/v/3054230/CAPTURE2.AVI.html

Will I be able to get better results using this method and Avisynth rather than the DVD recorded version?

I noticed VirtualDubMod would only capture using 15fps. If I tried 29.8 fps it wouldn't detect the incoming video feed. So is this interlaced, Bottom field first 15 fps video? Should I be concerned with 15fps vs 29.98fps?

Thanks,
Chris

[Soopafresh]

It doesn't look right.

[jagabo]

The sample video was free of compression and denoising artifacts so it's a good place to start.

You need to figure out why you can't capture at 29.97 fps though.

Here's what the Neat Video noise filter for VirtualDub can do for it (before, after):




I left some of the noise (a little noise is your friend -- it prevents posterization artifacts) and notice you can still see the pattern in the wallpaper and the grain in the floor. I only have the demo version of Neat Video so I had to crop the frame down to 640x480.

[chrishallowell]

Thanks jagabo.

Which software would you recommend using to capture the source? VirtualDubMod doesn't seem to want to capture in 29.97fps for me.

Thanks,
Chris

[jagabo]

It's unusual that VirtualDubMod won't let you capture at 29.97fps. I don't know what capture card you're using. Didn't it come with some software?

[chrishallowell]

OK, I was able to capture in 29.97fps using Cyberlink PowerProducer. I captured uncompressed(RAW) AVI in colorspace YUY2.
http://42mb.com/v/1166404/Captured3.avi.html

Any suggestions on how to make this better using Avisynth?

Thanks,
Chris

[jagabo]

I'm of a slightly different school than most around here -- I'm not big on noise filtering. Some detail is always lost with the noise. About the most I'd do with that video is mask the head switching noise at the bottom of the frame. Then I'd encode to MPEG2 with a high bitrate (assuming DVD is your goal).

One observation: the file you uploaded is RGB24, not YUY2. Maybe you used VirtualDub to prepare the clip and forgot to switch to Direct Stream Copy mode? It looks like the YUY2 to RGB24 conversion is causing crushed whites.

[chrishallowell]

I did use VirtualDub to prepare the clip....
How do I mask the head switching noise at the bottom?
What's direct stream copy mode and how do I set that in an Avisynth script?

If I'm no going to do any color alterations, I should just capture in YV12 since I'm using HC to encode to DVD format, right?

Thanks,
Chris

[jagabo]

I did use VirtualDub to prepare the clip....
What's direct stream copy mode and how do I set that in an Avisynth script?

Direct Stream Copy mode only relates to VirtaulDub. VirtualDub's default behaviour is to decompress the incoming video (if compressed) and convert it to RGB. All of its filtering is performed in RGB then the video is passed to the ouput file or codec as RGB. Direct Stream Copy is an alternate mode of operation in which the incoming video isn't decompressed or converted to RGB, its simply copied to the output file. No filtering or recompressing is allowed in Direct Stream Copy mode.

To create your short clip you should have used Direct Stream Copy mode so that the clip would have been an exact copy of those frames from your source. Conversion to RGB crushes the darkest and lightest areas and reduces the color accuracy of the rest by a bit. RGB is also 50 percent larger than YUY2.

How do I mask the head switching noise at the bottom?

Most editors have area fill functions or cropping/padding functions. I don't know of a fill function in AVISynth (maybe someone else does) but you can Crop() and AddBorders() to replace the head switching noise with a solid black border.

Crop(0, 0, -0, -8) #crop 8 lines off the bottom
AddBorders(0, 0, 0, 8) #add 8 lines of black to the bottom

If I'm no going to do any color alterations, I should just capture in YV12 since I'm using HC to encode to DVD format, right?

If you don't plan on extensive filtering you might as well capture directly to YV12 if your card supports it. Be sure to capture interlaced and be sure the encoder knows its interlaced. Otherwise the color channels of the two fields will get blurred together.