VC-1 and POSTERIZATION

[TheLion]

This thread is in no way intended to "bash VC-1" as a codec per se but to make the community aware of one of the single
most annoying and severe picture artifacts we still have to bear with our next generation formats - POSTERIZATION.

This is not a format war topic either as both sides suffer from it equally.
And this artifact is in now way limited to VC-1 as all three available codecs are prone to it.

Why did I choose to state explicitly VC-1 in the topic title then?
Well for once VC-1 is quite well represented here @ avsforum - I sure would love to hear some ("marketing-free") insights from Amir and Ben about this.

Other than that posterization just happens to be a really big issue especially with some of the latest VC-1 releases.
So even with the "very latest", applied generation of VC-1/PEP and considering that we are talking about (most probably highly optimized) high-profile
releases there is strong indication that there is in fact an inherent problem with the VC-1/PEP workflow regarding difficult to avoid posterization.

To put it very simple these are potential causes for the posterization we see all too often (in no particular order):

- it is already present in the source/master

- "flawed" 4:4:4 10/12bit -> 4:2:0 8bit conversion

- unsufficient hand-tuning by the compressionist for problematic sequences. eg. proper dithering can help to hide any obvious banding/posterization artifacts

- unsufficient bitrate/bandwidth for particular sequences. Problem is that the
encoder likely is automatically applying very low bitrates due to the "simplistic nature" of some of these scenes - this "should" be manually corrected in the second pass.
Best case scenario is that the encoder is able to indentify such sequences automatically.

I invite Amir and Ben to complement and/or correct this simplistic list.

To demonstrate the problem I will post the following screenshots of recent VC-1 releases.
My thanks goes to members Xylon and House for all their efforts in providing these.
These screenshots are for demonstration purposes only - results may vary slightly depending on the playback configuration
BUT I personally verified these caps with my XA2 and the PS3 - both times the results were just as bad as shown here :

Happy Feet , VC-1, HD-DVD/Blu-Ray, unaltered, provided by member Xylon


Direct digital transfer so no master to blaim here


Planet Earth, VC-1, HD-DVD/Blu-Ray, provided by member House




VERY interesting to say the least is the comparison with the BBC h.264/AVC broadcast:



So again this leads to the conclusion that the source/master is not the issue here (how could it ever be with such horrible posterization as shown here)
but the VC-1 encoding. Everybody is free to draw any conclusions from the comparison with the BBC capture.

These are just single samples of the problem at hand. eg. the recent Planet Earth release shows severe posterization (among other artifacts) not just occasionally
but a several times per 45min episode - some of these shots last for many seconds (like the one shown above) - starting with the initial BBC logo...

As another example I haven't really seen any high-def release with underwater sequences without a hint of posterization
(-> eg. VC-1 releases U571, Superman Returns, Happy Feet,... AVC/h.264 -> Ice Age 2...).

The important question is: What can be done to avoid it in the future?

I sincerely hope that Amir, Ben and other insiders admit this issue and provide insights of their strategies to avoid it.

[gooki]

- unsufficient bitrate/bandwidth for particular sequences. Problem is that the
encoder likely is automatically applying very low bitrates due to the "simplistic nature" of some of these scenes - this "should" be manually corrected in the second pass.
Best case scenario is that the encoder is able to indentify such sequences automatically.

Is exactly what's going on - any of the three codecs (MPEG2, AVC, VC1) will disply the similar levels of posturization when the above scenarion happens.

Basically it comes down to quality control and following best practices at the point of encoding.

[1031982]

Yes, but this can be manually corrected. I have done some video encoding, nothing major, but enough to know some things. If I use a variable bitrate, similar things happen, but if I use a 2 pass encoder, forcing the encoder to use a higher average bitrate, I end up with a only slightly larger file and a much better encode. By slightly, I mean I was getting about 300MB on the single pas, and around 330 with the two pass. Now, I realise that it gets a lot bigger with HD video, but it could get more advanced then I do by forcing the encoder to use higher bit-rates at certain points of the video. I personally have not done this, as I don't need to or know how, but I am pretty sure it can be done.

[ryoohki]

The things that happen here is thnat the encode see the scene as a BLACK scene 1 color, it doesn't see the whole color. In case of HF, i'am pretty sure the encoder saw a all black scene and putted 5mbits (witch is the bitrate for that scene) just like the Credit at the end and the computer didn't see the subtle gradiant of color.

Same with the sun, the color gradiant is so subtle and that computer must have seen only 1 color and dropped to 4-5 mbits (i don't know the bitrate scene i don't have planet earth) and voila Banding!

Those have to be hand corrected , i did a multiple of encode using x264 and i always needed to tweak those scene, especially if there's no grain. When there's grain, it not a problem because the encoder see it and boost the Bitrate and that compensate for it (altought in might need adjustements), you rarely see Posterization is a FILM filmed with Filmed, mostly only happen on Digital filmed stuff (with no grain) and CGI animation.

NIN show excel in that matter, it got TONS of gradiant and to my knowledge it never band..

[Timothy Ramzyk]

Originally Posted by ryoohki

Those have to be hand corrected , i did a multiple of encode using x264 and i always needed to tweak those scene, especially if there's no grain. When there's grain, it not a problem because the encoder see it and boost the Bitrate and that compensate for it (altought in might need adjustements), you rarely see Posterization is a FILM filmed with Filmed, mostly only happen on Digital filmed stuff (with no grain) and CGI animation.

NIN show excel in that matter, it got TONS of gradiant and to my knowledge it never band..

Now that's fascinating to me. It's another reason why people shouldn't be so eager to eradicate a films original grain from the process.

I found a lot of this on the SD of THE HUNGER, and remember reading somewhere that the materials on this film (though only 20 years old) basically had to be colorized due to fading. I wonder if the heavy digital manipulation left it more susceptible to this banding?

I know the BD of AMERICAN PSYCHO has been criticized for banding, and I think that's MPEG-2?

[mfuhlendorf]

This is very bothersome for me too, I notice it all the time on my Toshiba CRT rear projection with HDDVDs.

I'd go as far as saying that this sort of posterization is the worst flaw of HD DVD and Blu-ray, as they become obvious when the rest of the image properties are so perfect.

[Deja Vu]

If you have a Lumagen scaler with the 11 point gamma correction you can increase or decrease both banding and posterization - to increase the visibility of either, one would increase Luma at low IRE levels and do the opposite at to decrease both artifacts. I completely agree that this is a problem, especially for some displays, and could be rectified by employing TLC to these scenes. It's not a problem that is exclusive to VC-1 as has been pointed out. My opinion is that the problem lies with those applying VC-1 not being aware - in other words someone is asleep at the wheel or possibly the display being used isn't sceptible to these problems due the type of monitor and calibration.

Here are some examples I can think of - underwater scene in second Borne movie where the jeep goes over the bridge; Superman Returns has banding in the space pan down to upper atmosphere and posterization in the underwater scenes and of course Happy Feet underwater scenes (banding if I remember correctly). I suspect that with some displays this will hardly be noticable or may not be noticable at all and that's why some here will just shake their heads and wonder what we're talking about. If a friend says he doesn't see the problem you can simply watch the scenes pointed out here on his display and chances are you won't see the problem either. Let him watch on your display and see his reaction when those scenes are viewed.

Cheers,

Grant

[Li On]

Originally Posted by ryoohki

you rarely see Posterization is a FILM filmed with Filmed, mostly only happen on Digital filmed stuff (with no grain) and CGI animation.

Hmm, how about "Open Season"? It's pure CG and looks PERFECT!

I hope it's not a "flaw" of VC1. I can buy "sleeping" at the wheel. But Microsoft keep telling us they have THE BEST tweak for VC1. If this is the BEST they can get, I wonder what if someone is in a coma at the wheel!

Btw, according to Microsoft, more bitrate won't help because that's already as good as it gets. Their aim is to REDUCE bitrate with more tweak as time goes on!

regards,

Li On

[xradman]

Originally Posted by Li On

Hmm, how about "Open Season"? It's pure CG and looks PERFECT!

I hope it's not a "flaw" of VC1. I can buy "sleeping" at the wheel. But Microsoft keep telling us they have THE BEST tweak for VC1. If this is the BEST they can get, I wonder what if someone is in a coma at the wheel!
Li On

According to this post on Blu-ray thread, there is posterization/banding on Open Season.

Originally Posted by WilliamG

For those that say there's no banding in Open Season, I have proof now. I was at Best Buy, and literally watched the BD version on the 50" 1080p Panasonic plasma for just 2 minutes before seeing it glaring at me, mocking my very existence.

So, again, this is banding in OPEN SEASON:

18:01, and 18:43, in the sky. Banding is obvious, but...

The worst offender: 18 minutes, 43 seconds. Awful banding in the lower right of the screen, and pretty bad everywhere else on the screen.

This really just goes to show that some people just do NOT see banding, even when it's staring right at them.

Post on Banding on Open Season

[Rob Tomlin]

Interesting. Many people on AVS have been saying that the Blu-ray and HD-DVD versions of this are "much better" than the original BBC and Discovery HD broadcasts.

What bitrate was used in the BBC broadcast?

[House]

Originally Posted by Rob Tomlin

Interesting. Many people on AVS have been saying that the Blu-ray and HD-DVD versions of this are "much better" than the original BBC and Discovery HD broadcasts.

What bitrate was used in the BBC broadcast?

CBR 19.44Mbit/s h264.

[Li On]

Originally Posted by xradman

According to this post on Blu-ray thread, there is posterization/banding on Open Season.

Just checked. No banding whatsoever on "Open Season". The poster had no idea of the topic IMO.

regards,

Li On

[Li On]

I'm watching "Sky Captain and the World of Tomorrow". So far there were 2 very bad banding scenes:
- chapter 7, 36min48-51sec, the plane just dove into water, MPEG2 at around 20mbps
- chapter 7, 37min12sec, underwater scene with the plane, MPEG2 at 17.4mbps

Besides that the blu-ray looks great!

regards,

Li On

[Rob Tomlin]

Originally Posted by House

CBR 19.44Mbit/s h264.

Thanks.

Based on this, it would seem that if the BBC version didn't have posterization, there is no reason the HD-DVD or Blu-ray versions should either. Obviously.

[xradman]

Originally Posted by Rob Tomlin

Thanks.

Based on this, it would seem that if the BBC version didn't have posterization, there is no reason the HD-DVD or Blu-ray versions should either. Obviously.

But there is posterization in the BBC version, just not to the same degree. This is a very simple picture (content wise) and the fact that 19Mbps AVC encode still has posterization tells me that there is something more than just compression causing this problem.

[Kram Sacul]

Sounds like the encoding box that did the BBC version deserves a promotion.

[tvted]

Originally Posted by Rob Tomlin

Thanks.

Based on this, it would seem that if the BBC version didn't have posterization, there is no reason the HD-DVD or Blu-ray versions should either. Obviously.

This posting from this thread suggest the bitrate for the shots The Lion posted are considerably lower. You would have to question the accuracy of the bit rate meter on the PS3 but I would doubt the error would be this great.

I see banding in both images The Lion provided for what its worth.

ted

[abr27440]

Just goes to show what happens when proper dithering is not done. Sad really, but its not the codec's fault, thats just what was fed to it.

In my opinion the BBC version has a bit too much banding also.

[benwaggoner]

As suggested, that kind of banding is normally due to source processing issues. We have a great technology as part of PEP/CineVision PSE that prevents banding from being introduced as part of the conversion to 4:2:0 8-bit, but if there's banding in the source, than there's going to be banding in the output.

Also, comparing a broadcast to a disc title isn't always that informative, since they could have been different telecines from all we know.

Encoding at too low a bitrate can also cause banding, but as was suggested above, that's exactly what segment reencoding is for. Without access to the source, I can't say what the issue is in particular, but there's nothing intrinsic to VC-1 that could cause that.

[bferr1]

Originally Posted by benwaggoner

Also, comparing a broadcast to a disc title isn't always that informative, since they could have been different telecines from all we know.

Would a telecine even be involved, since the broadcast/disc comparison was of Planet Earth, which was shot in HD?

[TheLion]

Originally Posted by benwaggoner

As suggested, that kind of banding is normally due to source processing issues. We have a great technology as part of PEP/CineVision PSE that prevents banding from being introduced as part of the conversion to 4:2:0 8-bit, but if there's banding in the source, than there's going to be banding in the output.

Also, comparing a broadcast to a disc title isn't always that informative, since they could have been different telecines from all we know.

Encoding at too low a bitrate can also cause banding, but as was suggested above, that's exactly what segment reencoding is for. Without access to the source, I can't say what the issue is in particular, but there's nothing intrinsic to VC-1 that could cause that.

Ben,

thank you for your comments.

I have a pretty hard time to believe that the posterization galore that is the Planet Earth VC-1 release is anywhere near to being representative let alone transparent to any given master/source Warner/BBC could have come up with. Do you really suggest that posterization this severe and common throughout the runtime of Planet Earth is due to a "flawed source"? What about all the other compression artifacts - eg. as Fettastic so eloquently put it the flock of birds (beginning of the first episode) looks like it is being attacked by a swarm of killer bees... Lots of jelly-fish to be seen as well btw

Talking about Happy Feet - there is certainly no problem with the source as this is a direct digital transfer. You are stating that you provide "great technology as part of PEP/CineVision PSE that prevents banding from being introduced as part of the conversion to 4:2:0 8-bit" - great, so this is no conversion problem as well.
That leaves us with an encoding problem -> theory: encoding at too low a bitrate. This certainly is no issue with VC-1 per se as I clearly stated in my initial post:
" there is strong indication that there is in fact an inherent problem with the VC-1/PEP workflow regarding difficult to avoid posterization."

Therein lies the problem. Given the (rather questionable) trend of ever decreasing applied average bandwidth/bitrates for common VC-1 encodings (an approach and "accomplishment" that your team in general and Amir in particular seem to be quite fond and proud of) don't you agree that MS's PEP application "should" be able to indentify sequences that are prone to severe posterization and therefor AUTOMATICALLY apply sufficient bitrates during the first pass. This would make the quality of the encoding regarding posterization somewhat independent from the attentiveness of the compressionist.

Sure, you can always argue that there is no "problem" with the VC-1/PEP workflow that causes posterization - it is just the responsibility of the compressionst to hand-tune these sequences. BUT in the end we as customers have to agonize over many releases with severe posterization artifacts.

I think high-profile releases like Planet Earth, Happy Feet and Superman Returns clearly demonstrate that relying on the proper actions of any given compressionist in order to avoid posterization is certainly not an effective approach - at least not with our 12-15MBit/s avg <5Mbit/s minimum encodings.

[teknoguy]

Hi folks,

Allow me to show my ignorance but could someone tell me the difference between "posterization" and "banding"? They seem, to me, to be similar things.

Just trying to educate myself...
-t

[Neo1965]

I'll open up my P.E. over the weekend and take a look, but my best guess is that for those segments, quantization is set too high which results in lower bitrates but shouldn't really be used as the P.E. picture was so simple, it would have encoded well even with with lower quantization (lower is better).

It is possible that this and the MI3 example was merely operator (compressionist) error in dialing in too low a value for these frames. I think the only way we can prevent things like this is to try to keep a cap on how high quantization can go, even if the bitrate cap is 28Mbps, that should be high enough for most materials to work well --- and the P.E. frame was so simple there is no reason for that kind of posterization. (Note that the AVC version also had slight banding, but it became much more pronounced on the VC-1 version) Someone at MSFT should just take a VC-1 analyzer and look at that P.E. segment frame by frame and see if the theory of quantization set too high is correct.

[Mr. Hanky]

Just so we have the information on record, has anyone profiled the data specs on the BR/HD discs for this title? It's 4 discs for either format, correct? Are these sl or dl, though? What is the file size per disc? Is it about 2.5-3 hr of footage per disc with a single DD track?

[Graham Johnson]

90% of the banding issues I have seen are display related. What's apparent on my Plasma as banding, is totally missing on the 9 inch CRT projector.

I have yet to see a Plasma or LCD that doesnt have banding when the contrast ratio is limited by the scene in question.

Until a poster can assure us somehow it isnt his display, this is a non issue.

This seems to have become worse since HDMI and DVI became the standard transmission method. When component or RGBHV are used. Evidence of banding is largely eliminated. To me this points to contrast being limited by available 8 bit DVI or 10 bit HDMI transmission methods.

[House]

Originally Posted by Mr. Hanky

Just so we have the information on record, has anyone profiled the data specs on the BR/HD discs for this title? It's 4 discs for either format, correct? Are these sl or dl, though? What is the file size per disc? Is it about 2.5-3 hr of footage per disc with a single DD track?

HD-DVD;

Disc 1: Size - 23,792,599,040, Length - 2:32:17
Disc 2: Size - 23,658,328,064, Length - 2:32:44
Disc 3: Size - 24,005,175,296, Length - 2:33:11
Disc 4: Size - 15,973,357,56, Length - 1:42:13

Just the single audio track, DD5.1 448Kbit/s.

[WirelessGuru]

Originally Posted by TheLion

This thread is in no way intended to "bash VC-1" as a codec

Ummm.... sure it isn't.

[gooki]

90% of the banding issues I have seen are display related. What's apparent on my Plasma as banding, is totally missing on the 9 inch CRT projector.

I have yet to see a Plasma or LCD that doesnt have banding when the contrast ratio is limited by the scene in question.

Until a poster can assure us somehow it isnt his display, this is a non issue.

Agreed. If someone talks about noticeable banding, and blames the source, i'd definetly expect direct screenshots from the source before i accepted that argument over the display being at fault.

[Wesley5]

Originally Posted by House

HD-DVD;

Disc 1: Size - 23,792,599,040, Length - 2:32:17
Disc 2: Size - 23,658,328,064, Length - 2:32:44
Disc 3: Size - 24,005,175,296, Length - 2:33:11
Disc 4: Size - 15,973,357,56, Length - 1:42:13

Just the single audio track, DD5.1 448Kbit/s.

Too bad, with so much space left on HD DVD 30, they wouldn't include some extra or better audio I guess they are limited to 25 GB SL BD.

[Mr. Hanky]

They limited themselves, since there is no requirement to use one encode for both formats, nor is there a requirement to only use sl bluray discs.