SVCD: does this Filesize Seem Right for 3-Hour Movie?

[easymenu]

I've played around with all the KVCD & AviSynth bits and pieces and it was time to actually go through the process of making some SKVCD's (KCVD actually ) of my broadcast captures.

I worked from a 352x480 MPEG2 capture (video at 4000Kbps, audio at 256Kbps) of an edited/trimmed-up AMC satellite-broadcast of "The Godfather". The picture was not good: banding in the dark areas (which was much of the picture, as this was a darkly lit movie); MPEG-compression artifacts; film-grain noise.

Using KWAG's recommended filter list for VHS and captures:

Quote:

LegalClip()
GripCrop( X, Y )
GripSize(resizer="BicubicResize")
STMedianFilter(10, 30, 0, 0, 10, 30)
FaeryDust() # or PixieDust()
unfilter(50,50)
mergechroma(blur(1.50))
mergeluma(blur(0.2))
# Convolution3d or DCTFilter for extreme worse cases.
GripBorders()
LegalClip()

I came up with a sample file of 36,801KB (I'm doing the calculations manually). I did't like the quality, either played directly to my TV or burned to a CDRW for a DVD-player test. Just plain noisy. So I tried this, my variation:

Quote:

# AVS 2.0.8 script
# filter MPEG2 352x480 satellite TV captures
# ----------

MPEG2Source( "TheGodfather.mpg" )
Telecide()
Decimate()

LegalClip()

Convolution3d( preset="movieLQ" )
STMedianFilter( 10, 30, 0, 0, 10, 30 )
mergechroma( blur( 1.5 ) )
mergeluma( blur( 0.2 ) ) # default=0.2, use 0.1 if too blurry
Tweak( bright=8.0, cont=0.9 )

GripCrop( width=352, height=480, overscan=1 ) # Grip - only for progressive
GripSize( resizer="BilinearResize" )
FaeryDust() # PixieDust() # Pixie for heavier dust
GripBorders()

LegalClip()

Sampler( length=24 ) # for 24 FPS material
# Sampler( length=30 ) # for 30 FPS material

I rearranged the order so that noise reduction was applied first. If the image is altered first (resizing, etc.), filters are hindered from attacking their noise type. Working from intelligent-specific to dumb-general filtering, Convolution came first and the merge-blurs came last. Unfilter was eliminated. It's positive values are for sharpening. As long as there is noise, sharpening will also make the noise more prominent, defeating our noise-removing purpose.

Then I "tweaked" the image coloring and applied Grip cropping (with a touch of dusting). The sample file came out to 34,196KB -- an improvement in file size AND the image looked better.

But ... does this filesize seem right for a just-under-3-hour movie (which I imagine must be split across 2 CD's for SKVCD)? A previous test for putting a 52 minute program (30 FPS) on one CD had the sample at 12MB with a CQ value of 26 for a full SKVCD disc. Is this right? From what I read on the forums here, my CQ should be in the 60's and I should fit closer to 2 hours rather than one.

Ideas anyone?

[Boulder]

As it's a long movie, I'd go for KVCDx3 resolution, 528x480 in NTSC and use two discs. You'll get very good quality and it will be sharp too. You could also try using more overscan (2 or 3) so you'll get more bits to the actual film pixels depending on how much your TV crops the picture.

[easymenu]

I hadn't thought of that. My main concern is good quality (as much as fixing broadcasts permit), though fitting movies on 1 or 2 disc is nice.

Many forum posters are working from much better sources than I. But as KWAG's filter recommendation didn't seem to compress as low or clean as well as expected, I thought maybe I was doing something wrong (I mean, it's not really all that difficult - KWAG has done most of the work - thanks KWAG!).

I am trying to stay within the "legal" DVD / CVD specs. That means MPEG2 and 352 / 704 / 720 x480 video and 48kHz audio.

I'm still not happy with result of these filters, if only I think they should do better. In fact, I think one filter, or some combination of filters, is introducing short, verticle, texture noise. I think I must try isolated and exhaustive comparisons ... (I'm exhausted just thinking about it).

[rendalunit]

maybe the filesize of the second sample is smaller because you've IVTC'd with telecide().decimate() Is the framecount the same in both samples

[easymenu]

The "previous test" was at 30 FPS (couldn't use 24 FPS without jerky motion) so I was expecting less time per CD. However, if 24 FPS would get approximately 2 hours per CD, then 30 FPS should get an estimated 1 1/2 hours per CD. I was 1/2 hour short of that.

Also, after some more testing, I found was that it wasn't as much the filters (or their order) that produced poor quality results. It was TMPGEnc. Apparently, it really doesn't like encoding mediocre quality satellite capture clips, even if filtered -- at least at the 2500kbps max for SKVCD. MPEG artifacts from the original capture become greatly magnified in TMPGEnc, even with filtering. I was really hoping for better.

[MrTibs]

I do mostly VHS and Analog cable TV captures so here is my method for very noisy captures. (For what its worth.)

First, my DVD player is restricted to 352x240 so all my stuff is resized down to that resolution.

1. Edit - trim out any comercials.

2. Convert to 24fps. (I have to move between Telecide() /w Decimate and IVTC() /w FieldDeinterlace() depending on the channel or VHS source.) If you are getting jumpy results, switch the reverse pulldown set and try again.

3. Black out overscan

4. Resize - resizing before temporal and special filters will produce better results on noisy souces. (I realize that logic would say otherwise but you can't argue with results.)

5. Apply noise filtering. (I use C3D("movieLQ") & PixieDust() but Kwag's combo works very well and I think will compress more.)

6. Optional - convert to pseudo-widescreen. Method 2 keeps more of the frame and produces in a slightly sharper result.
Method 1: shift frame down 16, letterbox(21,21).
Method 2: Add borders(8,12,8,12), resize(352,240), letterbox(21,21).
Don't due step 3 when using Method 2.


FYI, I usually convert FullScreen captures to a pseudo-widescreen to increase compression. In my area, a 2hr movie with comercials removed results in 90 minutes of film, which fits on a single CD with a CQ at 100%.

[vhelp]

hi easymenu,
.
.
well, here's my long explanation/experience for this evening..

>> I worked from a 352x480 MPEG2 capture (video at 4000Kbps, audio at
>> 256Kbps) of an edited/trimmed-up AMC satellite-broadcast of
>> "The Godfather".

I think I would go a little higher in the capture bitrate setting. You
did say you are (real-time) capturing to MPEG via software ??
And, I hope it's MPEG-2 (not mpeg-1, to be on the safe side)

Satellite has a tipicle 4000 or less bitrate and setting it too close
or too little might not be a good idea. I would go for MUCH higher bitrate
like 8000 becuase you don't actually know WHAT your real-time softare
MPEG capturing is compressing (that is, taking away, in order to get
best results in final capture) And as you capture at 4000 bitrate, you
could be helping to produce (more than likely) those talked about artifacts
that you are experiencing in your final encodes, and not to mention, add
to the fact that your filtering are being affected as well, and your
"manual" samples are being effected too.

All these things can and seem to be haveing a catastrafic effect on your
sample analigy thus far.

I'm a little confused though.. about what your source really is
Is it:
* Satellite records to VHS tapes, or
* Satellite captures directly
* or combination of methods ??

Tip for the day..
* If Telecined source.. always IVTC first, THEN filtering the frames afterwards
* if Interlace source.. always de-Intelrace method first, THEN filtering
the frames afterwards
* if captured source is progressive (after careful analisys) do a quick 2.BOB
in vbud to be sure you have all the remaining (or as much of it before tainting
too much of the source quality) of the leftover Interlace artifacts.

Personally, I prefer to frameserve an .AVS script into vdub to fine-tune the
source and get every bit of issue under control.

Tip 2 of the day..
when working w/ vdub..
* always swap the time-line window (to the left side) so that you can see the
effects of your filtering/debugging. If anything.. all you need to see if just
a partial of your originals' source, while still maintaining view of your
filtered source. I find that magnifying each windows (good coordination) will
help in conjunction w/ picSwitch, when analizing a frame for best filter effect.

Tip 3 of the day..
* find and use a GAUGEing (tool or method of a scene) that can be used to
analize your source.

If you haven't seen my samples yet, have a look at the Interlace one. Using
Tip 3, I was able to fine-tune the final quality.

Tip 4 of the day..
Interlace:
* not all source materials captured are exactly the same. Interlace comes in
many flavors. Note, we are not talking about IVTC (aka, Telecine) here.
Telecine ..before the IVTC step :
* same as above. However, the Telecine process for a given TV airing are much
different than a DVD source !!
The entities involved in sharing the source mateirals w/ each broadcaster(s)
do not always do a correct or proper Telecine job, when they are "making for TV"
sources to distribute to other entities !! ie, Star Trek. This show is one
big headacke!! Note, we are not talking about Interlace (aka, Interlace) here.
This can (and more than likely) another reason/cause for your poor (or artifact)
in your final encode. A poorly done Telecine job will cause poor spacial/temporal
reading in your filtering. Bottom line.. pixels will not add up correct w/ respect
to filterings logic. Hope you understood that !! Maybe Kwag can explain
otherwise ..that is, if he understood me hehe..

-vhelp

[ozjeff99]

Hi vHelp

What is a guaging tool or method?

Cheers Ozjeff99

[vhelp]

@ ozjeff99, (and others)

sorry for sounding kind of covert..

* A tool can be anything you see fit that will help you in your process.

* you have a scene that is specific in nature, has certain charactoristics
...unique and that no other "anything" can do better. THAT becomes your
...tool (in this excercise)

* another tool could be your color space that pertains to your setup only.
...in that instance, your "color space" is your tool (the one that you set
...accoring to your TV or your PC monitor or whatever, but that you are using
...as YOUR tool.

These are just some of the examples of TOOLING, as a gauge in your
never ending qulity endeavor.

You will find that tooling/gauging in your processes can be indespensible.

I hope I was better understood now
-vhelp