PDA

View Full Version : Quality Improvement From Video Capture


EpheMeroN
29th April 2008, 01:20
I have a digital stream from the digital cable box captured via firewire. It appears to already be dvd-compliant 704x480 MPEG-2.

Do you guys think any avisynth filters would potentially yield a nicer result if I re-encoded it?

Here's a sample of the video: http://www.filefactory.com/file/6d5e09

And a couple stills (ImageShack doesn't support PNG anymore):
http://rapidshare.de/files/39259894/sample-stills.zip.html

thetoof
29th April 2008, 03:39
MCbob rm_logo dfttest crop+spline36resize

This is the bare minimum.

edit: just had a doubt in my mind... is this rule 6 violation?
I have a digital stream from the digital cable box captured via firewire.

EpheMeroN
29th April 2008, 05:38
It's from regular television... no need for rule violations!

And MCbob would be to double framerate yes? Why would I want to do this? The stream is 704x480 29.976fps wouldn't I only want to bob this video if I was to filter it with a denoiser or something?

2Bdecided
29th April 2008, 14:54
It appears the closed caption data is on-screen in the very top line, so maybe crop that - but is it worth it? I'd leave it as it is.

Cheers,
David.

thetoof
29th April 2008, 19:45
MCBob().Selecteven() would keep the original framerate. To my eyes, deinterlaced video always looks better (if you play it on a TV, you won't notice the difference, but when you play it on a computer, the program has to deinterlace on playback, which is always worse that what a great mocomped bobber like MCbob can do)

EpheMeroN
29th April 2008, 20:12
It appears the closed caption data is on-screen in the very top line, so maybe crop that - but is it worth it? I'd leave it as it is.

Cheers,
David.
Yea I noticed that too. I was going to crop it out, but only if I was going to re-encode it because I'm putting this on a dvd and a tv's overscan would hide that top line anyways.


MCBob().Selecteven() would keep the original framerate. To my eyes, deinterlaced video always looks better (if you play it on a TV, you won't notice the difference, but when you play it on a computer, the program has to deinterlace on playback, which is always worse that what a great mocomped bobber like MCbob can do)
This is going just to dvd so I'm going to keep it interlaced as it is. I would bob it first if some type of denoising/sharpening might yield a nicer result since I know most desnoisers can't be run on interlaced material. So far, from what I've tried (which isn't much) I haven't found a filter combo to improve the quality yet.

thetoof
29th April 2008, 20:28
Well, what I suggested before (dfttest for denoise + rm_logo to remove those annoying logos) is enough imo to improve your source. If you can't see the visual difference, it'll at least improve the compressibility of your source. For the bottom, I'd suggest that you crop it (unless it's important for you) and use Addborders. Yes, it will be lost anyways in the overscan, but it'll make your source more compressible, again.

EpheMeroN
29th April 2008, 21:38
Well, what I suggested before (dfttest for denoise + rm_logo to remove those annoying logos) is enough imo to improve your source. If you can't see the visual difference, it'll at least improve the compressibility of your source. For the bottom, I'd suggest that you crop it (unless it's important for you) and use Addborders. Yes, it will be lost anyways in the overscan, but it'll make your source more compressible, again.

Got any recommended settings for this dfftest? I'll search it after this post. As for compressibility, the entire video, which is an hour and 25 minutes, is already only 2.10gb in size so we won't have to worry about compressibility with this one.

EpheMeroN
29th April 2008, 22:58
Is "dfftest" that "very slow FFT Denoiser"? 'Cuz that's all that search brings up.

thetoof
30th April 2008, 00:12
Yeah, it's slow, but it's good.

You can download it here (http://forum.doom9.org/showthread.php?t=132194). Make sure to put libfftw3f-3.dll in your system32 folder.

For SD sources like yours, I usually use dfttest(sigma=x,smode=0,sbsize=5,tbsize=5)Where x is a value that needs ajustment on every source.

But if you don't care about compressibility, don't want to remove the logos and if the video looks good to your eyes, I'd encode it straight to DVD without processing.

Blue_MiSfit
30th April 2008, 01:31
MCBob().Selecteven() would keep the original framerate. To my eyes, deinterlaced video always looks better (if you play it on a TV, you won't notice the difference, but when you play it on a computer, the program has to deinterlace on playback, which is always worse that what a great mocomped bobber like MCbob can do)

That's true, but you absolutely will notice the difference in motion.

Interlaced video lets you cram 60 pictures per second into 30fps, which is fantastic for sports and other "lively" things. If you throw away half that temporal resolution you will most likely notice it!

Of course, if you're talking a flatpanel TV that just throws away half the fields... then yes you won't notice a difference and in fact the MCBobbed version will look a _lot_ better.

I try to backup my interlaced stuff to full rate progressive - that way I can spend all the billions of CPU cycles to MCBob it, then filter as necessary, and encode to 480p60 - keeping the full temporal resolution. Win.

~Misfit

EpheMeroN
30th April 2008, 08:04
Will that dfftest filter help combat this? I'm noticing it at certain spots during the video... appears to be artifacts from compression in the source?

http://img255.imageshack.us/img255/4034/10hk7.th.png (http://img255.imageshack.us/my.php?image=10hk7.png)

And is it just my eyes or do the colors seem a bit off? The faces on the two hosts look a bit too orange'ish.

http://img209.imageshack.us/img209/5466/10to4.th.png (http://img209.imageshack.us/my.php?image=10to4.png)

thetoof
30th April 2008, 08:11
Check this (http://forum.doom9.org/showthread.php?t=137324) thread (especially post #7) vinverse + deblock_qed will help you.

Also read the part where you can filter only a part of the video if the problem is not present throughout your clip.

thetoof
30th April 2008, 08:15
For the colors, play a bit with tweak and coloryuv. The "color correction" section of this website (http://www.animemusicvideos.org/guides/avtech/avspostqual.html) will help you.

2Bdecided
30th April 2008, 12:29
Of course, if you're talking a flatpanel TV that just throws away half the fields... then yes you won't notice a difference and in fact the MCBobbed version will look a _lot_ better.I haven't seen many flat panel TVs that are so stupid - the vast majority bob deinterlace to 60p.

If such "field dropping" TVs exist, then the mcbobbed version will either look identical to, or worse than, the original. Why? Because the TV is throwing half the fields away! It's either keeping the fields from mcbob that match those exactly from the original, or it's keeping the "created" fields from mcbob that don't. As it is not using both, it cannot benefit from mcbob.


Of course mcbob deinterlacing looks better on a PC than most on-the-fly deinterlacing - but dropping 60i to 30p to put it on a DVD for watching on a TV is an abomination - unless you like that fake film stuttery motion look I don't, and I think it would look terrible on the posted clip.

Cheers,
David.

talen9
30th April 2008, 14:28
And a couple stills (ImageShack doesn't support PNG anymore):

That's definitely not true: i tried to upload a PNG image right now and it work. Maybe the image you tried to upload was simply too big ( >= 1.53 MBytes which is the current limit).

Sorry about the OT.

EpheMeroN
2nd May 2008, 21:50
That's definitely not true: i tried to upload a PNG image right now and it work. Maybe the image you tried to upload was simply too big ( >= 1.53 MBytes which is the current limit).

Sorry about the OT.

Yes you were right! My original images were just over the size limit.

EpheMeroN
2nd May 2008, 21:51
Is there a way to setup deblocking so it ONLY deblocks when blocks actually appear? I've used deblockers in the past when only probably 4-5% of the actual video had blocking present, and in return the parts with no blocking get overly smoothed or flattened.