Applying filters before encoding to SVCD

[LordKariya]

I'm using tmpgenc to convert live-action video from .avi (huffyuv) to SVCD .mpg. The results are great, but not *the best* I've seen from live-action capture. It's my understanding that some experts use sharpening filter combinations to enhance the overall picture crispness (probably similar to the stuff divx uses). I'm wondering, for those of you who have *fantastic* quality SVCDs, what filters are you using ? Specifically, filters NOT internal to tmpgenc or CCE. My divx encodes of the same caps look better than SVCD - and that just shouldn't be.

Thanks for your help.

-LK

[Poplar]

It all depends on sources and bitrates. If your source is very clean, you don't need any filter at all. If it is noisy, temporal and spatial softening filters are applied, after that a sharpening filter. SVCD and DivX serve different purposes. If you think DivX is better, show me a good DiVX cap from a noisy VHS tape.

[LordKariya]

Poplar,

Thanks for the reply. The source is flawless, it's from Digital Satellite Through Monster S-Video. If I compress to divx (2-pass) without any additional filters, the result is superior to an SVCD without additional filters. They both look really, really good, but the SVCD has some mild blockiness to it, while the divx has the clean, crisp sharpness I'm striving for. It's my understanding that there are some external sharpening filters (not the standard Vdub ones) that will have similar effects to divx's natural filtration; this is what I'm looking for, to use pre-SVCD.

Thanks again.

[snowmoon]

Personally I would start with a clean 480x480 capture to huffyuv.

I use vdub and the following filters.

ViewFields ( split the interlacing so that progressive filters work right )
If necessary levels
2d optimized cleaner or Smart Smooth HQ either with reduced settings, this will blend out high frequency noise. If you feel the cap is good then drop this.
kNRC You might want to turn down the NR levels some to taste
xsharpen or msharpen ( depnds on the video - to taste )
UnViewFields restores any interlacing to the video

Encode with CCE make sure that progressive in unchecked and correct field order is used.

You also have to be careful, as divx playback is often heavly filterd based on bitrate. You can't compare a DIvx that has had massive post processing on it to help diminish mpeg artifacting to a poor mpeg-2 decompression without assistance.

[cdog]

It might be the way you encode.

I do the same as you... i cap digital satellite, but i use composite instead of svid. I personally found better results with my composite... dunno why, but i always thought svideo was better, but my composite proved me wrong (both rca gold stuff).

I use picvid quality setting @ 19, sometimes 20.

I get really good results using cce, with higher noise filter and better quality settings... 4-7 quality, 14-20 noise. vbr 5 passes @ 1600-2500 average, minimal 500 max 3500. Settings seem weird, but i get really good results.

Also, its a good idea to clean the avi source with virtualdub.

[ronnylov]

Instead of sharpening filters you can use softening and denoising filters to imporove the SVCD quality. Personally I prefer avisynth and the Convolution3D filter. You can user bilinearresize to get more softening.

Another thing you can try is increasing the values of the "soften block noise" setting in TMPGenc. There also is a denoising filter option built in TMPGEnc that is good but slow, you might want to do some tests with that also.

If you sharpen the picture then you'll need more bitrate to avoid the blockiness. So if your problem is blocky picture it's better to use softening to make it easier to compress at SVCD bitrates.

[DJRumpy]

All of this is mute if we don't know what your encoding with, and what settings are used. Everyone is just guessing.

Give us more information, specifically, what encoder your using, resolutions, and bitrates (min/avg/max if your using VBR).

[LordKariya]

Guys,

Thanks for the outstanding tips. As for more info,

ATI AIW Radeon, normally I cap to 640x480 huffy. I was under the impression that it was better to cap at 640x480 (no frames are dropped), apply filters/and then let TMPGenc resize to 480x480 - Am I off the mark here ?

Normally with TMPGenc, I'll use the Deinterlace (top-first, seems to be the one that works best) and noise reduction filters. Noise reduction setting of 100/4/100 and high quality mode. 2-pass VBR, SVCD compliance. I think the term 'blockiness' is definitely accurate to describe the end result. It's GOOD, but not Amazing. In this case, I should use softening filters as opposed to sharpening filters ?

Thanks again,

LK

[DJRumpy]

You should capture in the resolution your encoding to if possible. Your just adding in additional time to resize your source. You gain nothing by stretching, and then squashing your horizontal resolution. Note: this assumes your source is from broadcast tv, and not HDTV or some other high resolution source. It also assumes you are actually receiving your source at that resolution (i.e. SVHS will do you no good if your capturing with the composite jacks). If the source resolution is higher, then I would capture at the resolution of your source.

Your Top First/Bottom First should always be the same. I think it's related to your capture device (assuming your working on material you've captured..not some other source). Determine which it should be and stick with it. (someone back me up on this one..i'm not 100% sure)

What minimum/Average/Maximum settings are you using for VBR? Also, what Motion Search setting are you using?

[LordKariya]

DJ,

I think digital satellite is broadcast at 640x480. Actually, I made a mistake, I was thinking of IVTC when I said top-first. I used the even-odd adaptive deinterlace filter in tmpgenc, Highest quality(slowest) motion search, and I don't think I configure the VBR bitrate settings ? Doesn't 2-pass (SVCD compliant) do that for you, max of 2520 ?

Funny you should mention HDTV, I was going to get the decoder this weekend. Is it possible to capture that stream ? I had assumed it would not be a possibility.

Thanks again,
LK

[DJRumpy]

It would depend on the outputs it had available. To my knowledge, the S-Video cable will support up to 352x480. Component Outputs will support DVD, so I'm guessing at least 720x480 for component connections. I have an HDTV, but I don't care about the broadcast HDTV, and I use cable, not satellite. I never opted for the decoder, so I don't know what outputs it has. I would think that it would have to have at least component. The problem here is that your Capture Device probably has only S-Video, or Component Inputs, meaning your still only getting 352x240, or 352x480. If you have component IN on your capture device, then I would match your sources resolution (640x480 or whatever it happens to be).

[snowmoon]

Broadcast HDTV CAN be captured. You can even capture the raw DVB mpeg-2 and save yourself the hassle. After adjusting the stream to comply with mpeg-2 you should be able to use standard ( ok, not quite standard ) DVD convertion tools.

[LordKariya]

snow,

any chance you could elaborate on the HDTV capping ? This is unfamiliar territory to me. Do any caps cards actually have a component input ?

(edit: any cards that cost less than say, $1700 ? )

Thanks,
LK

[snowmoon]

none that I know of.

OTA = Over the air

Terestrial OTA HDTV/DTV in the US is in MPEG-2. With the proper Hauppaude ( WinTV-D? ) you can actually CAPTURE the mpeg-2 stream rather than some analog represnetation of it. If you are receiving digital TV/HDTV through any sort of cable or satellite provider this would be impossible without ( as you mentioned ) a companient HDTV capture board. I doubt they would even be that cheap.

Cheers.

[tonyp12]

Terestrial OTA HDTV/DTV in the US is in MPEG-2.

What part of USA have over the air Digital TV signal?

It seems that is only taking off in Europe for now.

http://www.digitag.org/

[snowmoon]

Over 500 digital TV stations are running at least part time in the US. Most in the top 20 major markets.

http://www.hdpictures.com/stations.htm

[vhelp]

afternoon all.

snowmoon,
inquery... you mentioned above on ViewFields Am I to assume that
you mean in AVS scripts there's a ViewFields funtion or something??
I've searched and tried typing in ViewFields() but bails out w/ an
error message.

so, I'm guess, if AVS script, is their a *.DLL file(s) that I need to insert into
my AVS scripts??

Thanks for any replies.
-vhelp

[snowmoon]

http://www.geocities.com/siwalters_uk/fnews.html

I only use avisynth for IVTC ( if necessary ) all other filters are Vdub

[vhelp]

Update to viewFields. . .

...case anyone was following the ViewFields filter issue.

- - -snip - - - snip

Ok, correction, link NOW works. . .
You can Download below:
* ViewFields v1.2
* UnViewFields v1.2

Note:
Remember to do your other filtering in between the viewFields.
-vhelp

[snowmoon]

(un)viewfields is great since it allows almost any frame/progressive based filter to operation correctly on field based materal.

[vhelp]

snowmoon,

ahaahh, didn't realize that you have to perform w/ BOTH.

Am I correct to assume that the proper syntax use for this filter (below)
is to be use this way (other ways, i'm sure). . .

So, an EXAMPLE Avisynth script Syntax would be:


LoadPlugin("UnViewFields.dll")
ViewFields()
.
. . . other AVS filters here
.
UnViewFields()

VDUB is a little different, but basically the same principle applies. . .
START:
* insert your video (AVI) clip here or after filter chain.
* Filters
* Add
* .. select (to open) viewFields v1.2
* .. add your other filters here ie, sharp, blurs, etc.
* .. select (to close) UnViewFields v1.2
* bring in your video (AVI) herer, if you haven't at the start of your filtre chain

Ok, so, basically, this is what I was doing in AVS script but w/ other
elaborate AVS functions - but it's another way for me to do, and give a
try and see if any differences can be seen.

So, (snowmoon) you're saying that this filter combo can
or may help to provide a sharper quality in some of our encodes
due to the video frames seporation and HOW we use other filters ie the
sharp/unsharpmask filters in vdub? I think I understand where you
are getting at, with this filter combo. So, people like myslef and LoredKariya,
may benefit from this filter combo - great! I've ben looking for something
like this. I know it can be done w/ AVS script, and I have done something
similar, but I was looking for something else different to try. I've done
various filter chains for sources like VHS mostly, as Satalite doesn't really
need it so much, except in extream cases.

See above thread for the two:
* vdub filters
* AVS script plug-in LoadPlugin("UnViewFields.dll")

Anyways, I learned of a new filter and will be giving it some good testings.
-vhelp

[snowmoon]

I perfer to use vdub and other tools like CCE that use standalone .avi files for encoding.

Most shapening filters are not interlaced aware and will give weird results around interlaced secions of the video. (un)viewFields fixes that problem for the most part allowing non-interlaced aware filters to be used on interlaced materal. I have found filters like xsharper ( like warp sharp ) and msharpen to be great at giving the video a little punch since often analog sources are soft to begin with.

[vhelp]

.
.
for me, when I pre-filter for my encodes, (CVD/SVCD etc) I will often use vdub
for most of my frameserving needs. I like using it because I can SEE what
the final result (close to it) will look before it gets to the encoder.
After the filtering, it's the encoder that will make or break your final
encoded quality. And, that will depend on the level of experience of the
user w/ his/her encoder. . .
* how are filters utilzed
* how are settings utilized in the encoder
* authoring settings
* type of CDR/CDRW media (as that also plays a part in final quality)
* DVD player, and it's ability to play YOUR encode projects
* size of your TV and in comob w/ all the above.

Any one of the above can through off the balance of final quality.
That's why it important to do small encodes, then burn and test on
your given DVD/TV setup. However, if you are destend to do mutliple of these,
I would suggest that you do a number of encode projects/settings like 5 to 10
and then burn those 5 or 10 test clips. This way, you don't:
* ware out your CDR burner,
* spend GREATER amounts of time (filtering, editing and encoding)

-vhelp

[DJRumpy]

vhelp, you can 'see' the same thing with AVISynth. You can play/preview your scripts directly in Media Player (or the player of your choice). It will treat the script just like an AVI, MPEG, VOB, or whatever your script is pointing to.

[snowmoon]

I have just finished up a pure AVISynth script to do most of what I was doing in VDub. Their are two scripts one for 3:2 materal ( movies ) and the other for interlaced materal. Note that I made up the latter on the spot it may have errors.

#Cap files are 704x480 for reference going to CVD.

#This line only for movies decomb is v4.0 right now. Get it!
LoadPlugin("c:\avisynth\decomb.dll")

SegmentedAVISource("g:\ntcap_31.avi","e:\ntcap_31. avi")

#These two lines for movies use post=true for hybrid where only a very small portion is hybrid
Telecide(guide=1,post=false)
Decimate()

#These are from playing in vdub, numbers are 1:1 with vdub
levels(25,1.0,228,16,235)
SpatialSoften(1,8,16)
TemporalSoften(1,3,6)
BicubicResize(352,480)
sharpen(.15)
#DONE#

# For interlaced materal the following would be similar to my vdub sctips ( making this up as I type may have errors )

master = SegmentedAVISource("test.avi").SeperateFields() # double frame rate half vertical resolution
even = master.SelectEven()
odd = master.SelectOdd()
filteredEven = even.levels(25,1.0,228,16,235).SpatialSoften(1,8,1 6).TemporalSoften(1,3,6).BicubicResize(352,240)
filteredOdd = odd.levels(25,1.0,228,16,235).SpatialSoften(1,8,16 ).TemporalSoften(1,3,6).BicubicResize(352,240)

# mix two streams back together ( UnViewFields )
Interleave(filteredEven,filteredOdd)
#DONE#

As always adjust to taste, I would turn down the Soften paramaters if you are not using 704/720 width as you will loose much more detail.

[DJRumpy]

Snow, shouldn't you be using Biliner instead of Bicubic, or did you use it to retain some sharpness after the temporal?