H.263 vs MPEG

[jellysandwich]

According to some guides and forums that I've looked at, I should use H.263 for anime/cartoons/low bitrate movies and MPEG for high bitrate movies.

My question is, what would you consider low/high bitrate? What do you think is the breaking point?

I'm about to back up my Godfather DVDs, and I can use around 970 kbps. Should I go with MPEG?

-jellysandwich

[Heini011]

Hi,

i don't use mpeg at all. it makes the video looking crispy and it amplified noise to much.

----

i use h263 for noisy material with lanczos resize and up to 848 x .... pixel resolution for anamorphic material.

----

for high-quality encodings i use slighty temporal + spatial noise-filtering to avoid amplifing noise, a reduced detail level within high motion-scenes for better compressability and dividee's six-of-nine hvs or andreas_78 matrix and a 800..816 x .... pixel resolution for anamorphic material.

example avi-synth script:
...
FrameEvaluate("mot_level = YDifferenceToNext/AverageLuma")

FluxSmooth(2,-1)

vid_sharp=deen("a3d",rad=1,thrY=1,thrUV=2,min=0.5,tthY=1,tthUV=2,scd=3)
vid_soft=VagueDenoiser(threshold=0.5,method=1,nsteps=6,chroma=true).deen("a2d",rad=2,thrY=2,thrUV=3,min=0.25)
ConditionalFilter(vid_sharp, vid_soft, "mot_level", "<", "0.14")

crop(0,72,718,432).BicubicResize(816,336,0,0.6)
...


greetings, Heini011.

[Sharktooth]

Originally posted by jellysandwich
According to some guides and forums that I've looked at, I should use H.263 for anime/cartoons/low bitrate movies and MPEG for high bitrate movies.

My question is, what would you consider low/high bitrate? What do you think is the breaking point?

I'm about to back up my Godfather DVDs, and I can use around 970 kbps. Should I go with MPEG?

-jellysandwich
There is no specific breakpoint but it depends on the material you're trying to compress.
However i use to apply h.263 or other soft matrices when the encode bitrate is lower than 1100-1200kbps (16:9 with full DVD resolution).
Keep in mind that a movie with lot of motion requires higher bitrates...

[Teegedeck]

Hm, personally I don't see much of a reason to use H.263 quantization at all. The HVS-best-picture custom matrix scales so nicely, perhaps you should give it a try. It depends heavily on taste, it seems, you should just decide for yourself.

[Sharktooth]

Originally posted by Teegedeck
Hm, personally I don't see much of a reason to use H.263 quantization at all. The HVS-best-picture custom matrix scales so nicely, perhaps you should give it a try. It depends heavily on taste, it seems, you should just decide for yourself.
The wise man spoken.
Everyone has is personal preferences so is always better to try and decide for yourself.

[malkion]

Just curious as I haven't come across nor possess a standalone player, but does the HVS-best picture matrice work nicely with these players?

H263 and MPEG matrice both are fully compatible I've read.

[Flipin]

I don't like the HVS custom matrix since the ringing noise and other mpeg specific noise has a distinct pattern that is much easier to spot than the seemingly random pattern of mpeg. With very high bitrates or without visible ringing noise hvs might be worth a try. Otherwise I would personnally stick with mpeg, since the h.263 has a tendency of bluring the textures and making walls in movies unrealistic looking and in anime lifeless by enhancening the "smooth gradient" feel to textures/general image.

I've tested with the french movie nikita and the japanese animation akira (both over two hours) packing both into 1cd which is quite a feat for movies that long. Nikita looked far better with mpeg and in akira I preferred the mpeg with more visible ringing than the h263 with a slight blurred touch to it.

from nikita if anyone is interested:
http://personal.inet.fi/surf/flipin/misc/clip.jpg
http://personal.inet.fi/surf/flipin/misc/clip.avi

[jellysandwich]

First, I'd like to thank everyone for their input.
I have been testing various Quantization Types,
including the ones you guys/gals mentioned, for
the past few days, but I can barely tell any
difference between them. I guess my eyes just
aren't good at distinguishing things.

However, there is something that I would like to
ask. When I look at some other people's encodes,
I notice that they do not look as blurry as mine.
For example:

http://www.freewebs.com/jellysandwich/hellsing.JPG

If you look at the faces, you can see that the one
on the right is a quite clearer than the one on the
left. What settings could I possibly use that would
help my clear away that blurriness?

In case you'd like to see my settings:

Q Type - H.263
Adaptive Quantization
B-VOP - 2/1.5/1
Closed GOV
Square (default) Pixel Aspect Ratio
First pass - Discard First Pass
Second pass - Defaults (10/1/20/5/5/5/0/0)
6 - Ultra High
4 - Wide Search
Chroma motion
0 Frame Drop Ratio
300 Min I-Frame
2/31 Quantizer for all
Trellis quantization
Debug Tab - defaults
Everything else - off

[Manao]

Well, they didn't use the same avisynth script you did ( they used antialiased font, for example ), so don't be surprised they've got different results.

From what we know, they may even not use the bitrate you did. Then, the quant matrice could be different. Furthermore, you could have picked up a B-frame in your encode, whereas in theirs it was a P-frame.

To sum up, there's no way to know what went wrong. However, I would try another avisynth script ( you seemed to have blurred too much your video )

[jellysandwich]

Originally posted by Manao
Well, they didn't use the same avisynth script you did ( they used antialiased font, for example ), so don't be surprised they've got different results.

From what we know, they may even not use the bitrate you did. Then, the quant matrice could be different. Furthermore, you could have picked up a B-frame in your encode, whereas in theirs it was a P-frame.

To sum up, there's no way to know what went wrong. However, I would try another avisynth script ( you seemed to have blurred too much your video )

The subtitles are external in both cases,
so I doubt that has anything to do with
the quality. And I used more bitrate than
them, so I don't think that is it either.

As for the Quant Matrice... That could be
it. I'll have to experiment some more with
them. I guess it could also have been a
B-frame.

And lastly, I think you're right about the
Avisynth script too. The only problem is
that I don't know much about using Avisynth...
It looks like I have something new to learn!

Thanks for the input Manao.

-jellysandwich

[K_R]

As far as hardsubbing anime, stick with h.263. I've experimented with hvs and mpeg and they produce noticable ringing around the subs, even in a full quality first pass (h.263 blurs the ringing a bit, thus making it non-noticable).

[dimzon]

Originally posted by Flipin
I don't like the HVS custom matrix since the ringing noise and other mpeg specific noise has a distinct pattern that is much easier to spot than the seemingly random pattern of mpeg. With very high bitrates or without visible ringing noise hvs might be worth a try. Otherwise I would personnally stick with mpeg, since the h.263 has a tendency of bluring the textures and making walls in movies unrealistic looking and in anime lifeless by enhancening the "smooth gradient" feel to textures/general image.

I've tested with the french movie nikita and the japanese animation akira (both over two hours) packing both into 1cd which is quite a feat for movies that long. Nikita looked far better with mpeg and in akira I preferred the mpeg with more visible ringing than the h263 with a slight blurred touch to it.

from nikita if anyone is interested:
http://personal.inet.fi/surf/flipin/misc/clip.jpg
http://personal.inet.fi/surf/flipin/misc/clip.avi

wich settings does you use?

[Sharktooth]

Originally posted by jellysandwich
If you look at the faces, you can see that the one
on the right is a quite clearer than the one on the
left. What settings could I possibly use that would
help my clear away that blurriness?
If you're using H.263 and the image is too blurry you can try to add a sharpening filter (not too strong though) in the .avs.

[Teegedeck]

It would seem a bit counterintuitive though because it lowers compressibility.

[Sharktooth]

Infact i told him to try.
But if blurriness is his major problem a little less compressibility won't hurt that much.
Some specific edge sharpening filter would do the trick without hitting to much the compressibility.
As last resort Q-pel could also help but i think it will do more bad than good.

[yaz]

if ringin of subs is the only reason for using h263 u can easily circumvent that.
on hardcoding put the subs onto a blank (black) clip & after some blurring overlay this onto the film. it helps a lot with ringing.
another point; use as low (luma) contrast as possible with subs, especially if they'r put onto the black borders. of course, keep'em visible/readable :-) say, silver, yellow or green subs produce much less ringing. if handling the subs separetely, u can decrease the contrast with coloryuv or tweak.
the bests
y

[Heini011]

Hi,

@Teegedeck:

>I don't see much of a reason to use H.263 quantization at all

i can't agree with that. h263 works as vaguedenoiser + a high-detail matrix. so it yields great results with a bigger resolution, lanczosresize and quarter pixel on most dvd-material. noise is smooth out while it maintain details, when second pass size is max 25% under first pass size.

imho only for backups of high-quality material (on hd or dvd-medium) the six-of-nine-hvs, andreas_78 or hvs-best-picture matrix is useful.

greetings, Heini011.

[jellysandwich]

Wow. Sharpness(.3) helped a lot!

http://whitevoid.com/snowman/hellsing2.JPG

Top right: Original encode
Bottom: Goal
Top left: With Sharpness(.3)

[Sharktooth]

:)

[Soulhunter]

Originally posted by jellysandwich
Wow. Sharpness(.3) helped a lot!
Try also mftoon... (http://forum.doom9.org/showthread.php?s=&threadid=53364&highlight=mftoon) ;)


Bye

[SoonUDie]

MFtoon is good, but I think a simple super-sampled XSharpen() would do the trick (resize to say 200% of the final size, do an Xsharpen(100, 150) or so, then resize back to final size). Also, instead of using the standard sharpen(), you could try using MSharpen() which only sharpens significant edges (you can get them here http://www.avisynth.org/warpenterprises/)

As far as quants go, my personal opinion is that h.263 beats out everything at EVERY bitrate except for a small slice inbetween about 1250-2250 where other matrices actually have a (small) chance. This is assuming you like h.263's denoising effect on high-bitrate material, of course. I came to this decision after testing many of the more popular matrices out there (HVS, 6of9...). h.263 simply beats them all in terms of perceived detail vs. artifacts.

[jellysandwich]

Originally posted by SoonUDie
MFtoon is good, but I think a simple super-sampled XSharpen() would do the trick (resize to say 200% of the final size, do an Xsharpen(100, 150) or so, then resize back to final size). Also, instead of using the standard sharpen(), you could try using MSharpen() which only sharpens significant edges (you can get them here http://www.avisynth.org/warpenterprises/)

As far as quants go, my personal opinion is that h.263 beats out everything at EVERY bitrate except for a small slice inbetween about 1250-2250 where other matrices actually have a (small) chance. This is assuming you like h.263's denoising effect on high-bitrate material, of course. I came to this decision after testing many of the more popular matrices out there (HVS, 6of9...). h.263 simply beats them all in terms of perceived detail vs. artifacts.

I tried mfToon, but I didn't like it. It didn't give
me what I was looking for. I'm trying out MSharpen right
now, and so far it looks good. I'll have to do some more
testing. If that doesn't do the trick, then I might try
XSharpen() and see how that goes.

I agree with your statement about H.263; it looked much
better than the other matrices at all sorts of bitrates.
But that's just my personal preference. I'll probably
stick with it for all my encodes form now on.

-jellysandwich

[Sharktooth]

A sharpening/edge sharpening filter was definatly what you needed.
Do not exagerate with the filter strength or you will get a bad overall compressibility and loose quality.
I can also confirm H.263 is a good matrix for animes/cartoons.

[lordadmira]

The absolute most amazing filter I have seen on anime is warp sharpener. I use it in VDub and I think it has an avisynth version. It turns blurry lines into crisp dark thinner and more pleasing lines.

[jellysandwich]

Originally posted by lordadmira
The absolute most amazing filter I have seen on anime is warp sharpener. I use it in VDub and I think it has an avisynth version. It turns blurry lines into crisp dark thinner and more pleasing lines.

Oooo. This looks like something worth trying.
How much depth would you suggest?

-jellysandwich

[lordadmira]

I've gotten good results on S-video capped material with depth around 76 and 2 passes. But open the preview window and play around with the settings. U get some wild effects when u start turning everything up toward maximum.

[jellysandwich]

Warpsharp is amazing.

-jellysandwich