MPEG2 quality [Archive] - Doom9's Forum

eng3

6th June 2003, 22:42

I am trying to get my Xvid encoding to look as good as a MPEG2 encoding. I have tried changing around many settings in different combinations (Motion Search, VHQ, Qpel, Chrom motion, (MPEG, H.263)) and even when I make it so the filesize ends up the same as my MPEG2 encoded clip, the MPEG2 one still looks better.
The mpeg2 one is smooth, almost no blocks can be seen, yet it does not look washed out.
The Xvid one looks good too, but many block can still be seen.

Shouldn't a MPEG4 video look better than a mpeg2 video of equal size?

Am I at the limit of what I can do with the settings I mentioned above? Do I need to try some custom quantization matricies or try changeing the Alt Curve settings?

I've read several guides that explain what each setting does, but I guess I lack the experience to get really good results.

Sigmatador

6th June 2003, 22:55

the only way to have the same quality for the same filesize, is of course to don't transcode the source :D

however, for very very high bitrate, try to reduce the max i-frames interval and use a high quality matrix (andrea's one maybe)

eng3

6th June 2003, 22:59

Originally posted by Sigmatador
the only way to have the same quality for the same filesize, is of course to don't transcode the source :D

however, for very very high bitrate, try to reduce the max i-frames interval and use a high quality matrix (andrea's one maybe)

what do you mean by "transcode the source"?

I have one source video file.
I create one compressed using MPEG2 and one using XviD
Both the same filesize.
Shouldn't the MPEG4 one look better than the MPEG2 one?
Ofcourse I my goal is not to make the XviD file so large, I just want to beable to make it look at least as good.

I'll start trying different quality matricies

Teegedeck

6th June 2003, 23:43

I have one source video file.
And would you mind telling us what kind of source file? Is it that thing in your sig? Huffyuv, MJPEG... Or do you capture directly to XviD? Interlaced, progressive? Low or high quality?

Etc., etc.

Shouldn't a MPEG4 video look better than a mpeg2 video of equal size?
Not necessarily. If you use settings that are optimized for speed (=deactivate many features that give MPEG4 an advantage over MPEG2), you won't get the best possible quality of course. And if you use 'CBR' you won't get anything watchable, anyway. It really isn't that easy. So you can't possibly expect to get any meaningful ideas from other users if you describe what you're aiming to do so vaguely.

Sigmatador

7th June 2003, 00:40

ok i didn't have understood that you make the mpeg2 and the mpeg4 from the same source.

yop, can you telling us what kind of source is it. in the case of a capture, you can use a lossless codec (vble ^^) and reencode it with 2pass (for both mpeg2 and mpeg4 format)

imho, mpeg2 could be more efficient on very high bitrate.

eng3

7th June 2003, 01:29

Huffyuv, MJPEG... Or do you capture directly to XviD? Interlaced, progressive? Low
I've tried Huffyuv, MJPEG, Divx5 (1pass, 150000, fast, interlaced) and XviD (everything off, quant=3, interlaced), MPEG1, MPEG2
My results do not seem to be as good when using Huffyuv and MJPEG.

I then recompress using Xvid 2pass
I've tried a many different combinations of Motion Search=6, VHQ, B-frames or noB, chroma motion, MPEG or H.263, ...
But in each case, when comparing to MPEG2 (or recompressing to MPEG2), it looks worse.

My goal is to simply make it look better than with MPEG2 with a smaller filesize, I've never been able to get it to look better than with MPEG2. It is much more blocky.

Many of you have made many videos. Do you think your results look better than MPEG2?
I wanted to know if it is necessary to use the custom quant. matrix?

BiaTch 5.0

7th June 2003, 04:04

Try this (http://forum.doom9.org/showthread.php?s=&threadid=53136) guide & could you tell us what the file size difference between the mpeg2 & the XviD encodes are?

eng3

7th June 2003, 04:06

Originally posted by BiaTch 5.0
Try this (http://forum.doom9.org/showthread.php?s=&threadid=53136) guide & could you tell us what the file size difference between the mpeg2 & the XviD encodes are?

Yes I have read that guide already
The file size difference is zero.
I set the 2pass to encode the Xvid so it would end up being the same size as the MPEG2.

Xaanin

7th June 2003, 08:16

Well, what about this source? What codec was used, what resolution, what is your target resolution, how much noise is there in the source, have you tried using any noise reduction filters etc. ?

About the blocks you experience in the xvid, do they appear everywhere or only in high/low motion scenes, light/dark scenes?

eng3

7th June 2003, 08:31

I tried serveral different codecs capturing TV from a capture card.

Originally posted by Enrico Ng
Huffyuv, MJPEG... Or do you capture directly to XviD? Interlaced, progressive? Low
I've tried Huffyuv, MJPEG, Divx5 (1pass, 150000, fast, interlaced) and XviD (everything off, quant=3, interlaced), MPEG1, MPEG2
My results do not seem to be as good when using Huffyuv and MJPEG.

The resolution is 640x480 and I am leaving it at that res.
I can't say that its crystal clear like a DVD, but its pretty clear.
I'm not sure how to describe it. The source is interlaced.
I use the "smart" deinterlacing filter.

The blocks are pretty evenly distributed through out the scenes.
They are present in all scenes, but are far more visible in high motion scenes.

I'll run some more tests tonight (as I do everynight it seems)

Xaanin

7th June 2003, 08:39

kilg0r3

7th June 2003, 10:14

I'd try the hvs best or max-best matrix or even the 'professional quality' one which has been posted somewhere in this forum.

The problem with the standard matrices is that, since you cannot go below quant 2 in two-pass encode, they constrain bitrate where it might be needed. A matrix with lower values gives the codec the freedom to spend bitrate when it 'thinks' it is necessary. [edit] Furthermore it will be abel to adjust in a more fine graned manner to the needed bitrate, consequentially not killing more detail than necessary ( which it would have to reconstruct later on btw, because once gone it can no longer be used for prediction) this is why, for me, I multiplied all values in the mpeg matrix with 2/3. Only the values in the bottom right corner (4*4) are higher in order to compress stronger the very-high-detail levels. The hvs matrices use a similar approach afaik.

About b-frames, it might be useful to use very conservative settings like 2,125,100,50 or to disable them altogether.

VHQ: I'd use the 03052003-1 koepi build with vhq set at 1. But this is just me not trusting the younger algo.

Which bitrate are you using?
How much noise is there?
Are there many fades, much smoke, fog or similar things?
Is there much motion?
Is it good (predictable) motion or handy-cam-like jerks

The bottomline is that, if the clip is of a nature that does not allow to use advanced features of mpeg4, they will do no good.

BTW I'd be curious how rv9 would do in your case.

Blueseb

7th June 2003, 10:24

please please please be more specific and exhaustive while being synthetic: it took about 8 posts just to know resolution, and still we don't know the bitrate!!! this argument is very interesting to me, but by now it's impossible to give any opinion. to post Xvid settings in an effective way maybe you can use xvidset,do a search to find it.
furthermore,doin' XviD 2pass, did you retained 1st pass? does it shows same defects or not? and what is 2nd pass/1st pass percentage?

Nic

7th June 2003, 10:42

Im interested by this too, as to me, MPEG-2 is far more simple than MPEG-4, so in theory MPEG-4 should beat it always.

So I asked syskin about it...he stated:
"I think it has to do with quantization range - at high bitrate, mpeg4 has to work in range between 1 and 3, so it doesn't have a big choice"
(i.e. at high bitrates the difference between quant 2 and 3 is large and that causes loss of details)

Try with info from these threads:
Professional Quant Matrix
http://forum.doom9.org/showthread.php?s=&threadid=53924
Studio Quality
http://forum.doom9.org/showthread.php?s=&threadid=53632

Let me know how everyone gets on :)

-Nic

kilg0r3

7th June 2003, 11:04

Originally posted by Nic
[B]Im interested by this too, as to me, MPEG-2 is far more simple than MPEG-4, so in theory MPEG-4 should beat it always.

.. at high bitrate, mpeg4 has to work in range between 1 and 3, so it doesn't have a big choice"
(i.e. at high bitrates the difference between quant 2 and 3 is large and that causes loss of details)

Hi Nic,

Lowering the values in the qmatrices, if i understand it corectly, should result in an overall higher quantizer, in a 2-pass scenario of course. This, in turn would mean that the codec can adjust the quantization more finely(?:)) because e.g. the difference between qmatrix value*quant4 and qmatrix value*quant5 is less than between qmatrix value*quant2 and qmatrix value*quant3.

Do I make myself clear?

BiaTch 5.0

7th June 2003, 11:26

Originally posted by Nic
Try with info from these threads:
Professional Quant Matrix
http://forum.doom9.org/showthread.php?s=&threadid=53924
Studio Quality
http://forum.doom9.org/showthread.php?s=&threadid=53632

Let me know how everyone gets on :)

-Nic Not that I’m knocking Bishep but he changed his opinion about Quartelpel 3 times. I myself was wondering if I should use Qpel.

Nic

7th June 2003, 11:45

Ok, today the two things I asked syskin about was MPEG-2 vs MPEG-4 and QPel (what a coincidence hey? ;) )

Anyway, I couldnt understand why QPel wouldnt improve the image always. (apparently the more accurate ME helps but the sub-pixel filter changes...and that doesn't help).

I.e. QPel will generally lower the PSNR :( But on the plus side sharpens the picture :)

You'll have to make your own choice on whether to use it or not. And if you like its effect.

-Nic

@kilg0r3: Yup thats correct. Apparently with H264, the range is 1-64 so the quants would change at around 6-7 at high bitrate which is a lot less difference than 2-3. Sounds like H264 should perform well no matter the bitrate then :)

7th June 2003, 12:10

MPEG2 Quality is laughable. Remember that I encoded that matrix reloaded trailer at insane resolution (for people who don't know what I'm talking about: I encoded the Matrix Reloaded trailer at 1600x864 res with a target size of 100MB) ? I still have the original VBLE file (1.4GB), and I encoded it into maximum DVD resolution and bitrate (that is 720x576 anamorphic 16:9 at 8Mbit/s) with CCE. The file is 30MB bigger, and the quality (not to mention sharpness because of resolution loss) is much lower. Viva MPEG-2 :D.

PS: The quality was higher than that of ANY retail DVD you could get your hands on though :D. It also had no noise (only DCT noise from encoding).

Owen

7th June 2003, 14:14

I find this thread very interesting.
I have been using Xvid to capture digital satelite (mpeg2).
The best setting I have found is 1 Pass, fixed Quant 2 in both MPEG and H.263. I force a key frame every second (25frames)
I have tried many combinations of settings and custom matrix.
I capture at 720x576 PAL and have never been happy with the macroblocks and general compressed look.
Even at bitrates higher than DVD it is still blocky.
I want DVD quality or better at DVD bitrates . I just can’t get there.
I would use an Mpeg2 codec if I could find one that could be used for capture.
I have been looking for a solution for 3 years now.
My idea of quality must be higher than most.
I just dont think Xvid can do it.

Regards,

Owen

snowbeach

7th June 2003, 14:22

my opinion on this, because I tested CCE and XviD for a 2 CD rip of a DVD (movie length: 2h18min) last days. result: XviD won. the picture was clearer and less noisy than the one of CCE. But there was no big difference. Movie encoded: Black Hawk Down

this discussion is quite funny, because I think you cannot compare MPEG-2 and MPEG-4! it depends on that what you want at least.

in my case I backup very important movies to me to 1 DVD+R using CCE, because then I have a perfect rip - no visible difference between DVD and my rip, which is playable on most DVD players. So I use CCE with bitrates that are higher than 3.800 kBit/s.

If a movie is not important to me, but I still want a good rip, I use MPEG-4 encoder. Then I use XviD. ;) Movies that lasts longer than 2h I rip to 2 CDs and movies that last below 2h I rip to 1 CD and I get pretty good results. So at least I am pending at a bitrate from 700 kBit/s to 1.300 kBit/s.

Because of 2 different technologies, MPEG-4 compresses stuff very good, you cannot compare MPEG-2 and MPEG-4 at low bitrates. Ok, I did it, but I always want to see it.... to prove, I am right... :D

7th June 2003, 14:24

Originally posted by Owen
I want DVD quality or better at DVD bitrates . I just can’t get there.
I would use an Mpeg2 codec if I could find one that could be used for capture.

Ever thought of doing a direct stream dump of the sattelite stream?

Owen

7th June 2003, 14:32

Yes, but the equipment to do this is not yet available in Australia.
One sat card has just been released to our market. But it is untested at this time.
I sill cannot understand why Xvid cant do a better job with a data rate of about 6Mbits.

Regards,

Owen

kilg0r3

7th June 2003, 15:06

@Owen

get yourself one of those matrices mentioned above. If it is not enough for you, you can lower the values even more. Below 16 ffdshow does not decode correctly, though. You will have to use xvid.dll or Nic's decoder then.

eng3

7th June 2003, 17:53

Xaanin:
I thought about showing some images, but I thought that it would not be very accurate since XviD may encode every frame differently. So far I have been using the standard quant matricies. As for settings I have tried almost all combinations of (Motion Search=5,6, VHQ-0,1,4, Qpel, Chrom motion, (MPEG, H.263), B-frames or not). In all those cases I have not beenable to get any to look as good as my MPEG2

kilg0r3: I am trying custom quant. matricies for the first time now, I'll tell you how it goes. Thanx for the explaination. I am setting the filesize to be eqaul to my MPEG2. It is a TV capture so it is more noisy than a DVD, but it is still pretty clear. It is Interlaced so I run it through the "smart deinterlacer" first. There are fades and some fog in the clip and some motion, but even it almost no motion scenes, it is still more blocky than my MPEG2. rv9? never heard of it, I'll look around. Oh and I read the prof. quality thread already too, I'm trying out the matrix now.

Blueseb: I was not that specific about my settings because I did not think it matter in this case because I thought that mpeg-4 should always be better than an MPEG-2 of the same size. I have tried many settings which I have posted above a few times but none result in a video better than the MPEG2. I do not know what the bitrate is, I just set the filesize to be the same as the MPEG-2. I did not originally post the resolution because I also thought it did not matter because it is the same in my MPEG2 clip. Retain the 1st pass? do you mean the video.stats file? As I have said before I have tried many settings so I did not retain any. I tried so many different one's I would not know which one to show you.

Nic:
MPEG-2 is far more simple than MPEG-4, so in theory MPEG-4 should beat it always
Yes this is what I thought too. So do you mean that mpeg-4 is better for low bitrates but not high bitrates? Eitherway it still does not look as good as MPEG2 and lowering the bitrate will just make it worse.

snowbeach: I am a little confused by your post. Do you mean that between 700 and 1300 kbit/s your video looks better than mpeg2?

Well everyone, thanks for your comments so far, I will try some custom matricies and see how much that helps.

Blueseb

7th June 2003, 21:39

Originally posted by Enrico Ng
I do not know what the bitrate is, I just set the filesize to be the same as the MPEG-2.
well, when you're dealing with compression, you should know at least how much do you compress, don't you? ;)
Retain the 1st pass? do you mean the video.stats file? As I have said before I have tried many settings so I did not retain any.

no, i mean the avi clip generated during the first pass (usually very huge): to do this just uncheck "discard first pass" in th two pass tab. the first pass output provides the maximum quality reachable with your settings; if it looks bad the problem is settings-dependant, if it looks good problem [u][b]may be[b][u] final filesize too small (AKA low bitrate)

snowbeach

7th June 2003, 23:23

@Enrico Ng

Right!

Tommy Carrot

8th June 2003, 01:17

Don't forget that Mpeg4 was mainly improved for the low bitrates (the original aim was cell-phones). Mpeg2 wasn't usable for low bitrates, because the too frequent keyframes. Mpeg2 has to use keyframe for every 15th frame, while mpeg4 hasn't.

But this has much less influence in the high bitrates. But in theory, mpeg4 should still be a bit better, but i guess the motion estimation of xvid plays role in this issue.

BTW, you should _never_ use qpel for high quality encodes. The additional noise just doesn't worth it. :(

symonjfox

8th June 2003, 15:27

My tests show (and confirm) that MPEG4 is better than MPEG2 at low bitrates.

Now, on high bitrates, most of you consider as source a DVD, a DVB capture or an analogic capture. None of these are UNCOMPRESSED, so they might have some noise, artifact or anything else.

I think that is like REcompressing an MP3. If I take a WAV and encode it to MP3 @ 192 kbs, it sounds great. Now, if I take this MP3 and reencode to OGG Vorbis @ 160 kbs, this file will sound WORSE that a direct WAV -> OGG compression. Else if I reencode the MP3 to OGG Vorbis @ 80kbs this file will sound a very little worse that a WAV -> OGG conversion. I hope to be clear. If you don't believe me, try yourself.

So if I take as input a DVD (High Bitrate MPEG2 compressed), and I reencode it to 3 CD MPEG4 (High bitrate MPEG4), it would be like MP3 192 -> OGG 160. So this file will NEVER reach the original MPEG2 quality, you can try any matrix or filter or whatever you want.
Else if I convert this DVD to Low bitrate MPEG4 (as most people do), the difference won't be noticable from an UNCOMPRESSED -> LowBitrate MPEG4.

For REAL testing, we should have some DIGITAL uncompressed material to try. I think that a DV camera should give enough quality for testing (but I don't know since I haven't one).

Bishep

8th June 2003, 18:09

In due time I patiently searched for the best and universal audioformat which would allow good quality as on low bitrates – for the Internet, and super-quality on high – for audiophiles. :confused: After visiting some forum topics on HydrogenAudio.org I should agree, that the universal audioformat which equally meets the requirements of economy and ideal quality simply doesn’t exist for today (even if to forget about support of 5.1 and 7.1, 24 bits, 48 and 96 kHz :scared: ). On low bitrates Vorbis and RealAudio (…) show great results, on high – MPC (…). ;) The Reason for this concerns to the different coding systems. Vorbis uses complex transform-coding that gives excellent quality at small size, but cannot give transparent sound (if to compare to the original) even at rather big size of a file. Best or transparent quality MusePack (MPC) gives because it uses simpler subband-coding, that deforms an original sound less. :rolleyes:

What for I speak about it… :p

Actually I want to carry out an analogy between videoformats MPEG2 and MPEG4. MPEG4 uses complex system of coding, therefore it is as much as possible adapted for low and average bitrates – it considerably deforms the original image by its nature (the more transformations, the distortion of the original is more significant), but gives visually pleasant quality and a small file size. MPEG2 uses considerably more simple system of coding, therefore deforms the original image less, however gives rather big file size.

MPEG4 is definitely created for low and average bitrates, it needs some additional work to use it effectively for high bitrates. Such work is named as tuning of videocodec. I should tell, that XviD Team has achieved the most significant results in this area (For example, custom quantization matrices & complex b-frames support). :D For audioformat Vorbis such work was made by Garf – Vorbis GT3b1. :D

BiaTch 5.0

9th June 2003, 01:27

Keep in mind DVD Video audio is AC3 & must be transcoded to another format, so I don't think one could say that mpc would be the best performer without testing on a AC3 source.

As far as audio to me it's very simple:
Usually (when source is 5.1) - 128kbps=mp3, good quality
Trying to keep 1CD rip - 96kbps=ogg/aac, low quality
Lucky, got 2chl - 192kbps=Original AC3, :)

I think I have asked this before but here goes again, is there a way to keep only 2chl's from 5.1 ac3 source? this would be extremely useful for 1CD rips.

Morbo

9th June 2003, 01:39

I use a DAzzle DVC II,and I think MPEG2(hardware) is alot better than a software/vivo capture.

I can capture @ 1400 kbits SVCD and its compatible with most DVD players by nature....no transcoding at all.

It all depends on what you do and want.

eng3

9th June 2003, 19:45

So basically from what I am reading here, MPEG2 is better than MPEG4 except for lower bitrates.

What kind of bitrates are low.
The MPEG2 I had was about 2000 kbps
so I guess it would not be possible to create a mpeg4 that looks better.

BiaTch 5.0

9th June 2003, 21:15

Originally posted by Enrico Ng
So basically from what I am reading here, MPEG2 is better than MPEG4 except for lower bitrates.

What kind of bitrates are low.
The MPEG2 I had was about 2000 kbps
so I guess it would not be possible to create a mpeg4 that looks better.
IMO below 1500kbps I too found that SVCD encodes have looked better than my DivX/Xvid. I thought that if you restrict quant 2 & use no B-frames, Qpel, GMC you got 100% DVD quality?

symonjfox

9th June 2003, 22:00

For me we should consider the Bitrate/Area.

For example a 2000 kbs SVCD (480*576) will look worse that the same Xvid (720*576). Believe me, I made a lot of SVCDs but at the end I switch to CVDs (352*576). These discs have better quality than SVCD because they've less area to encode. Yes, SVCD is a little bit sharper, but since SVCD and CVD has limitation in bitrate, CVD is the best IMO.

Nibor

9th June 2003, 22:17

Originally posted by BiaTch 5.0
I think I have asked this before but here goes again, is there a way to keep only 2chl's from 5.1 ac3 source? this would be extremely useful for 1CD rips.
I'm a big big newbie in audio encoding (for me ogg vorbis with -q 2.00 sounds perfect *g*)...
But with BeSweet you can get 2ch from a 6ch ac3 file...
I did a couple of 1CD rips with 2ch ogg sound from a source with 5.1 ac3 sound...
Maybe I'm blaming me now with this answer and your question was about "something totally different" :D
But at least I'm trying to help.... So.... Forgive me! :)

Phanton_13

10th June 2003, 00:41

Sorry for my english is bad.

In the video or audio lossy recompresion exist a problem when it is used a diferent format for recompresion and isn't procesed or filtered the video.

If we part fron a real digital losless source ( ej. capture direct from a DV camera ) and if we compress directly it to MPEG2 and MPEG4 we obtain beter quality at the same bitrate with MPEG4, but if we recompres the MPEG2 to MPEG4 and MPEG2 we obtain better quality with MPG2, in the other hand if we resize the video to other resolution we obtain beter quatily with MPEG4. The reason for this is very simply, when we compress with lossy altgoritm it is introduced a frame simplification in mode of a diferenzes from the real frame, this simplification can be considered a type of noise or a distortion , the distortion for MPEG2 and MPEG4 are diferent for this when we transcode MPEG2 to MPEG2 with not video process we obtain better quality at high bitrates than we transcode to MPEG4 and it is for when we trancode to MPEG2 whe have the same distortions at the same points and it is virtualy masked by the old distortion but when it is to MPEG4 the distortions aren´t the same for the same points an they is more visible and in some case cam be exargerate this distortion for the ol distortion, but alwais exist a inflexion point and in this case on whe obtain better quality with the MPEG4 the reason it's simple: MPE4 is more eficient than MPEG2. It's similar with we trancode in the same format with diferents quantification matrices at the original bitrate.

BiaTch 5.0

10th June 2003, 08:03

Originally posted by Nibor
I'm a big big newbie in audio encoding (for me ogg vorbis with -q 2.00 sounds perfect *g*)...
But with BeSweet you can get 2ch from a 6ch ac3 file...
I did a couple of 1CD rips with 2ch ogg sound from a source with 5.1 ac3 sound...
Maybe I'm blaming me now with this answer and your question was about "something totally different" :D
But at least I'm trying to help.... So.... Forgive me! :)
Yeah I mean keep only the 2 chl's of AC3 without decoding, this would mean no quality loss what so ever..

Acaila

10th June 2003, 08:32

Yeah I mean keep only the 2 chl's of AC3 without decoding, this would mean no quality loss what so ever..I do hope you mean downmixing those 6 channels into 2 channels (which will give you a slight quality loss, if I'm not mistaken, because it's an approximation). You're correct that by keeping just those two channels won't degrade quality for those two channels, however you will lose audio information. The center channel is what 99% of the time caries the voices from actors, the 2 rear channels give you audio from the environment, so you see that by just keeping the left and right front channels you'll miss a lot!

As for dowmixing from 5.1ch->2ch, both BeSweet and HeadAC3he can do that just fine.

ChristianHJW

10th June 2003, 09:25

Originally posted by Enrico Ng
I tried serveral different codecs capturing TV from a capture card.
The resolution is 640x480 and I am leaving it at that res.
I can't say that its crystal clear like a DVD, but its pretty clear.
I'm not sure how to describe it. The source is interlaced.
I use the "smart" deinterlacing filter.
The blocks are pretty evenly distributed through out the scenes.
They are present in all scenes, but are far more visible in high motion scenes.
I'll run some more tests tonight (as I do everynight it seems)

When reading through this, and comparing with the results i could get in very best case with MPEG2 ( trying to put a 2+ hrs movie on 2 SVCDs, with 5 - 6 passes in CCE ), there is only one conclusion :

You have problems with your deinterlacing filter !. The blocks you are talking about are maybe more like horizontal stripes ?

Actually, could you please describe precisely how you are encoding your movie ? What container format is your source in, is it an .avi file coming from your capture card ?

What programs are you using to encode to XviD, and how do you add the deinterlacing filter ?

Same for MPEG2, what program and how do you make sure it really not interlaced ?

I assume that you are using CCE for MPEG2 and Virtualdub for XviD into .avi, and if i am not completely mistaken then CCE has the capability to detect interlaced sources automatically and to encode it interlaced then ( you wouldnt even notice ), while Virtualdun is highly depending on the deinterlacing filter, else you will get a lot of vertical lines in the picture, showing up especially in motion scenes and eating bitrate like crazy ....

Could you please clarify Enrico Ng ?

10th June 2003, 12:03

Originally posted by Phanton_13
If we part fron a real digital losless source ( ej. capture direct from a DV camera )
That is wrong. DV is lossy - it uses about the same principle as MJPEG.

eng3

10th June 2003, 14:59

ok I will try to clarify
I use VirtualVCR to capture
I tried using Huffyuv, MJPEG, Divx5, Xvid, mpeg1
For divx5 I set 12000 bitrate, slowest, interlaced
For Xvid, everything off, quant=3, interlaced
using YUY2 for all
This is all from a TV capture of a live broadcast.

The MPEG2 stays interlaced I think. I may have used the de-interlacing filter that comes with TMPEGENC but I don't remember if it has one. I also tried with my capture card's hardware mpeg2, I think this stays interlaced.

Then I use virtualdub and compress as I described.
I use the "Smart Deinterlace" filter. This is the best one I have found. I have tried several others and also have tried just leaving the "enable interlaced" checked, but when playing the video, you can see the interlacing.

Please tell me if you need anymore details.

Phanton_13

10th June 2003, 16:52

Originally posted by mf
That is wrong. DV is lossy - it uses about the same principle as MJPEG.
:( Sorry for my litle confusion, this is true but is the most quality source than I can be optained and it's the same for anione normal user; and in the reality any digital format is a lossy format :D but this is other History.

10th June 2003, 16:58

Originally posted by Phanton_13
in the reality any digital format is a lossy format :D but this is other History.
Less lossy than analogue film. You sacrifice resolution (which is almost infinite, up to atomic level, in analogue film) for the ability of creating 1:1 copies and lossless editing. You lose in the initial conversion to digital format, but from then on, it's lossless, unless you compress (lossy).

((( atom )))

10th June 2003, 19:09

personally i also tend to believe that one should be able to achieve higher overall quality with mpeg4 than with mpeg2, especially at the same filesize, simply because mpeg2 is limited in its resolution. if it wasn't and if we wouldn't care about filesize, i am sure, mpeg2 would win.

if i raise my resolution, of course my artifacts like blocks become smaller compared to the whole picture and will be less noticeable.

i could get my hands on either mpeg2 i-frame only material wich is made especially to transcode it with minimal artifacts. these files are HUGE (up to 270mbit/s!) but they kick ass of course. i will see what i can find and maybe i can put some snipplets on our ftp-server @ work (2mbit) and you could just download and compare these. that would lead to definite results.

should i?

Im interested by this too, as to me, MPEG-2 is far more simple than MPEG-4, so in theory MPEG-4 should beat it always. there is nothing more "simple" than a chian of raw bitmap images would be and these would give the best result of course.

also you could try and set min and max i-frame interval to 1 and get 1-frame only mpeg4 (idea stolen from mpeg2 broadcast habits). i suggested this in the capturing forum with little responses only. i believe this to be a great choice for keeping max quality for further editing or whatsoever. also if every frame is a keyframe you can cut on every frame you like. filesizes go up to the ususal dvd mpeg2 range. some short tests didn't let me spot out the used dvd-source anymore, wich was amazingly good in the first place (looked a lot like progressive hdtv camera whilst recording in a studio with brilliant light setup -> no noise at all and extreme detail).

OUTPinged_

12th June 2003, 13:55

You sacrifice resolution (which is almost infinite, up to atomic level, in analogue film)
Not true. Even premium quality Hollywood 35mm films cant show much more than 3500 of b/w columns without excessive blurring. It is most like VHS analogue width - you can't represent it exactly by pixels, but that doesnt mean that VHS is better than 720 pixel DVDs.

Nic

12th June 2003, 15:28

"there is nothing more "simple" than a chian of raw bitmap images would be and these would give the best result of course."

Whats that got to do with anything??? MPEG-4 should always win be cause it can compress better, but it doesnt because of the quantizer range at that kind of bitrate. It could be tuned to be better than MPEG-2 though in high bitrate circumstances. MPEG-4 part 10 will beat it in all cases ;)

-Nic

12th June 2003, 15:38

Originally posted by OUTPinged_
You sacrifice resolution (which is almost infinite, up to atomic level, in analogue film)
Not true. Even premium quality Hollywood 35mm films cant show much more than 3500 of b/w columns without excessive blurring. It is most like VHS analogue width - you can't represent it exactly by pixels, but that doesnt mean that VHS is better than 720 pixel DVDs.
The "resolution" of film is the grain size. However, you cannot represent that as resolution in the way you can digitally, because grains are on random spots - they can be displaced for VERY small amounts, up to atomic level. So even if you have film of "3500x1890 grains" - however blurry it might be, digitizing it with 3500x1890 pixels will not give you the exact same resolution as your analog material. That was what I meant.