AVS Forum banner

Status
Not open for further replies.
1 - 20 of 38 Posts

·
Registered
Joined
·
833 Posts
Discussion Starter · #1 ·
[In the 'Van Ling's take on EE' thread, djs brought up the age old question "Could EE be an MPEG artifact?". I decided to post this in a new thread, because this discussion comes up every so often and isn't only related to the thread above.]




A 'few' thoughts... **cough**




Yes, compression can and does lead indeed to ringing.


The important part here, is NOT to search and assume a SINGLE reason for the ringing that we see on DVD. So the question 'Could EE be a compression artefact?' (which has been brought up in every 2nd post on this issue over the last 2 years, btw :)) should be modified to 'Is compression part of the reason for the ringing on DVDs?'. And the answer is 'yes'. Is it the sole reason? 'no'. Is it a major or minor reason? 'depends'.


I can think of about 10 reasons in the film->DVD process that lead to ringing. Its always a combination of all of these. Compression is always a part of it. Some may want you to believe that compression is THE reason for EE, but it just isn't that simple. Its the 'easy' excuse that you might hear from some compressionists. Because it puts the ball out of their yard. If its mostly a compression artefact, then its inevitable, so they aren't guilty of anything.


The important term is 'inevitable'. Is it? The simply fact that quite a few DVDs ARE smooth as silk without any ringing is enough evidence that it isn't. So, maybe these EE free discs are all encoded with super high bitrate and were easy to compress (short running time; not grainy; etc)? No.


Braveheart comes to mind. Basically no EE, smooth film-look, yet impressive detail (not false detail through high frequency boosting, but real detail) and that with such a long running time, and such dark and grainy scenes.


Or Titanic. Long running time, ultra smooth picture, incredibly high detail (for a non-anamorphic title), many dark scenes. Yet no EE. This title is interesting, since it DOES have a considerable amount of mosquito noise, blocking and other compression artefacts (more than most of the EE offenders!), yet there is basically NO ringing at all.


Yet, on other titles with only half the running time, same average bitrate, almost grain free, not half as many typical compression artefacts as the 2 mentioned above, the ringing/halos are supposed to be an inevitable compression artefact? Yeah, sure.


Thesis: 'EE is mostly a compression artefact'


Without being technical, simply through some common sense, I could probably list a dozen of hints, why you could/should doubt that compression is the major cause for ringing:

1) Statistical pattern.


Take a few dozend of DVDs. Some long, some short, some grainy, some clean, some with high bitrate, some with low. Find a pattern in regard to EE? No! I have just as many long, complicate-to-compress titles that are EE free, as short clean transfers which are ringing desasters.

2) NTSC vs PAL


Take a dozen of NTSC and the corresponding PAL title (mastered from the same studio, not someone else). PAL has 20% more vertical resolution. Given the same bitrate, the PAL title should have MORE compression induced ringing, because of that. Is that the case? No! Most of the PAL titles that i compared to their NTSC counterpart have LESS vertical ringing (often considerably), yet MORE typical compression artefacts (blocking, mosquito, etc). This completely contradicts the thesis.

3) 2.35:1 vs 1.85:1


Take some movies with comparable average running time and comparable bitrate, half of which are 2.35:1, the other half being 1.85:1. For example titles from Columbia Tristar (CTHV), which always (few exceptions) exhibit ringing. Given the same running time and bitrate, the 1.85:1 titles should have more compression problems, since they have 27% more vertical picture information to encode. So they should have much more ringing than 2.35:1 movies. Is this the case? No! Exactly the opposite is true again. Most 2.35:1 CTHV movies have lots of ringing, while a large part of their 1.85:1 movies are quite smooth and ringing free. How could this be possibly true?


Now some more technical thoughts:

4) Mosquito noise and Compression induced ringing


Ringing at 'edges' as a compression induced artefact only occurs together with mosquito noise, because they are the same thing. Not having enough high frequency capability to perfectly 'model' an hard edge, will not only lead to that 'edge' showing ringing, it will lead to the whole block getting 'ripply' (because you throw away the high freq's of the whole block). Thus every block that hard edge will lead through, will have ripples (either in only one or in both dimensions, depending on the edge(s) in that block).


These ripples are called mosquito noise. "False, high-frequency ripples around hard edges, that follow a block pattern." You cannot have strong halos, which contour your every edge precisely and NOT have strong mosquito noise (with the same frequency and amplitude) in those blocks as well. Yet, many, if not most DVDs with horrible EE have just that. Horrible halos, yet no mosquito noise or other noticable compression artefacts.

5) EE assymetry


Most of the DVDs which are the main EE offenders have a strong assymetry in their ringing characteristic. TPM for example has basically zero horizontal EE (visible at vertical edges!), yet a tremendous amount of vertical EE (visible at horizontal edges).


For this constallation to be even remotely compression related, the transfer would need to have 10 times the horizontal frequency response than vertical. But this is not the case. The detail characteristic is 'mostly' symmetrical. Also, the frequency boost of the EE in TPM is much lower (lower freq EE -> thicker halos) than the finest vertical reproduced detail. This is impossible for compression related ringing.


I haven't seen an assymetrical chosen DCT coef matrix yet (Ron?), which would be needed for assymetrical ringing characteristics in the first place.


And again, a lot of the EE heavy hitters HAVE strong assymetrical ringing. For example: Forrest Gump and TPM strong vertical EE, almost no horizontal. Tombstone Vista and Die Hard 3 S.E. strong horizontal EE, only little vertical.


A favor towards vertical EE is very common on NTSC DVDs. While PAL DVDs, which have a slightly higher vertical resolution are often completely free of any vertical ringing. Odd, no? That does sound an aweful lot like someone thought "Oh, NTSC doesn't have as much vertical resolution, let's boost it up some!".

6) Ringing alteration and 'contouring'


Compression related ringing (along with mosquito noise) should vary strongly from shot to shot, since it depends on bitrate in relation to detail and motion. If even in I-Frames with a reasonable bit budget, yet only little detail and no motion in the shot, objects exhibit strong halos that 'precicely' contours edges without any or only little mosquito noise, its very unlikely, to be compression related. Yet this is often the case.



I could go on and on, but this should do for now.



All of that doesn't mean that certain MPEG encoders (like the sadly popular Sony) aren't responsible for a certain amount of ringing in many DVDs.

Done with the Sony, they often do indeed show:


a) Ringing at edges, PLUS mosquito moise of same amplitude/freq

b) No carried response above this ringing freq

c) symmetrical detail and ringing


On those discs, a large part of the ringing is indeed an 'compression artefact'. The choice of brick wall filtering in the Sony, coupled with the used DCT coef matrix and the overall efficiency and intelligence of the encoder design may lead to this.

But the very important question in this regard is: "Does the fact that one brand/model MPEG encoder produces ringing edges all over the place by default, without any deliberate action of the compressionist, automatically mean that this is a 'compression' artefact, per se?"


My answer is a clear NO. Because there are other encoders which seem to have no problem producing images with much less or no edge artefacts. The term 'compression artefact' in my mind is coupled with it being an INEVITABLE fault of the general MPEG encoding/decoding scheme.



And its not like all encoders which do produce less to no edge artefacts have unacceptable low frequency response characteristics (per given bitrate). Au contraire. The 'measured' frequency response might be lower, since no frequency equalization (EE) is used to compensate high-freq rolloff, but the actually encoded frequency response of the original SD frames is just as high if not higher on those smooth EE free transfers. At least the actual original freq response is more 'apparent', since its less overwhelmed and intruded by false contouring.



In the domain of the compressionist, its important to differentiate the primary responsibility:


a) either the compressionist deliberately uses certain encoder settings or EE pre-filters to achieve some 'sharpening'

b) or the 'engineer' of the chosen MPEG encoder designed a certain equalization behavior (EE) on different input material (NTSC, PAL, 2.35:1, 1.85:1 etc) under certain circumstances (high/low bitrate, etc)


The point is, even in case 2, the 'actual' responsibility is with the compressionist. "I didn't do no nothing" simply isn't an excuse, IMO. The result counts. If the ringing was not due to the compressionist's intend, but due to the encoder and its default behavior, its still up to him to indentify the ringing, and if connected to the encoder, try another.


A funny sidenote in this regard, is that several compressionists told me flat out that some (or even most) of their clients 'prefer' the snappier look of the Sony encoder. THIS preference for the resulting picture with more ringing and a higher (yet false) frequency response, is NOTHING else but 'edge enhancement', whether the ringing was 100% compression related or actual equalization (EE). The 'choice' in itself, for the 'edgy' picture, is EE.



Other sources for ringing, not related to the compressionists domain:

A) Aperture correction at the telecine (Film -> HD master). AC is also called 'enhancement'.


Aperture correction is used to compensate the inevitable 'aperture effect' of the scanning sensors, a high freq rolloff at the bandlimit of the sensor's native resolution (e.g. 2k), due to the high sensor fillrate, which is necessary in terms of light efficiency. The only and ultimately inevitable solution to actually defeat this rollof, is oversampling. E.g. scanning at 4k and downconverting to HD.


If this isn't possible, you can either live with the AE rolloff, or you try to equalize it with EE.


Aperture correction = High-frequency Equalization = Enhancement = Edge Enhancement


Most of the HD that is broadcasted today (or sold on D-Theater) is edge enhanced. Thankfully, the frequency bands that are boosted here are much higher, thus, the halos are mostly much thinner than on DVD. Sadly, often, EE is used to the extend that its still 'benefitial' (cough) at SD resolutions. Ugh. A nasty habit from the earlier days, where the primary purpose of HD master, was to produce snappy SD masters afterwards.

B) Downconvertion.


You cannot resample 1920x1080 HD pixels to 720x480 SD pixels without at least a slight bit of ringing at the SD bandlimit. If done properly (Sinx/x), or at least b-spline based, or bicubic, this isn't all that relevant in comparison to actual deliberate EE or badly designed MPEG encoders.


I would be interested in knowing what algorithms the commonly used anti-alias boards work on, for example. Why are they used in the first place. A properly downsampled HD->SD source, doesn't need to be anti-aliased.


Will have to investigate this area some further.




That has to be enough for today. Most will probably be asleep already.


Regards

Bjoern
 

·
Registered
Joined
·
1,390 Posts
Absolutely.


Here's a quick photoshop file I created. JPEG compression IS highly related to MPEG compression, and here's a highly compressed JPEG file:
http://www.aaronoz.com/images/test.jpg


Now here's a zoom in on that file, showing what appears to be EE, and I PROMISE I didn't do any unsharp mask or anything... it is simply a zoom of the above highly JPEG compressed file:
http://www.aaronoz.com/images/test.gif


BTW - For the zoomed file, I used a GIF format for the save so as not to compound the JPEG compression with more JPEG compression.


To me, if nobody will admit EE on TPM, I really feel that it simply looks like a highly compressed image.
 

·
Registered
Joined
·
833 Posts
Discussion Starter · #3 ·
Aaron,


i don't know if that was your intention, but your example serves the exact opposite purpose of what you write.


- The high compression level induces a high level of mosquito noise, especially at the corners.

- There is NO EE-typical 'contouring' halo around the edges like with EE, but a block based ripple effect, where the effect on the edge is determined through its relative position within the MPEG 8x8 blocks.


This is exactly what solely compression based artefacts look like and in no way resembles the EE that we see on DVD. Somewhat similar, but not the same. I hope that is readily apparent.


Only few DVDs have that strong compression artefacts, btw. This shows that even with a compression level that is much higher than commonly used on DVD, you cannot force typical EE artefacts.


On typical EE laden DVDs, with normal to strong halos, only about 10-20% of those are due to a strongly subdued form of the compression artefacts that can be seen in your example. The rest is actual EE.

Quote:
To me, if nobody will admit EE on TPM, I really feel that it simply looks like a highly compressed image.
Did you actually read my post?


Bjoern
 

·
Registered
Joined
·
1,750 Posts
I've tested this many times myself. My results are consistent with what Bjoern is reporting. There are edge artifacts with MPEG2 encoding, especially at low bitrates, but they are typically more random, ragged looking lines or bands. Not these thin well defined lines circumscribing objects people usually show in their edge enhancement screen shots. No way to tell for sure without having the original source material though, I guess.


Joe
 

·
Registered
Joined
·
746 Posts
Especially with the Episode 1 example, there's multiple examples of it being encoded with multiple levels of edge enhancement. Take the R2 PAL version, or the trailer of the film, or even the deleted scenes. At some point, the R1 version especially, the master was "tampered" with and lots of enhancement was added.


I've seen plenty of DVDs that look absolutely gorgeous and are missing this "feature" the Phantom Menace has. I am willing to believe that SOME encoders out there may introduce some kind of edge enhancement without user input (and in my opinion, are doing a very poor job at what they are supposed to do), but there are plenty of counter examples showing that encoding DOES NOT lead to a certain amount of edge enhancement. The Phantom Menace is a great example especially since it is the same source material and (most likely) encoded by the same people.


Keep up the good fight, Bjoern. ;)
 

·
Registered
Joined
·
833 Posts
Discussion Starter · #6 ·
Quote:
I've tested this many times myself. My results are consistent with what Bjoern is reporting. There are edge artifacts with MPEG2 encoding, especially at low bitrates, but they are typically more random, ragged looking lines or bands. Not these thin well defined lines circumscribing objects people usually show in their edge enhancement screen shots.
Thanks for the backup, Joe.



Mike, the different versions of TPM material (R1, R2, trailer, deleted scenes) that i show in my review, are indeed IMO an excellent example. The trailer, which has the most classical compression artefacts, has basically no EE. The R1 feature presentation, which has the least amount of compression artefacts has by far the biggest amount of EE.



I would like to keep this thread for the general discussion of MPEG vs EE artefacts and not focus on TPM. The strong vertical EE in TPM is not compression related. Period.
 

·
Registered
Joined
·
833 Posts
Discussion Starter · #7 ·
I put together a few demonstration pics.


Here is the first one:

http://www.videophile.info/Misc/EE/EE-MPEG_01.jpg http://www.videophile.info/Misc/EE/EE-MPEG_01_mini.jpg
Click on picture to enlarge


At the top left, you see the original picture. The color combination is rather typical for a problematic shot on DVD. A dark silhouette against a brighter background (think pole or person against sky here). The dark not too dark, the bright not too bright, so that actual over- and undershoot ripples can be seen better.


Now, in the picture below the original, i applied quite a bit of EE. With considerable amplitude and rather typical frequency.


In the top right, a moderate MPEG compression ratio is applied, that is common for a mid-class (in terms of compression) DVD transfer. The false pixel swarms around the edges are called 'mosquito noise'. Note that they don't follow the whole edge, but are tied to blocks, see below. What you can see, is that there is no real contour tracing with halos.


The picture below that, shows very heavy MPEG compression. Much stronger than what DVDs are encoded at. Think below 1.0 MBit here.


The MPEG block structure is very apparent here. In some of the blocks, you can see black and in some also white halos around the edge, 'similiar' to what EE looks like. The difference is, that it doesn't really 'trace' the edge with these halos, like with EE, but the visiblility and severness strongly depends on how the edge passes through a given block. Also note, that in some of the blocks, there are multiple ripples or they aren't parallel to the edge, but are simply ripple patterns which try to approach the original content (and fail miserable at that bitrate ;)).


If you have difficulties to see that 'block' pattern, take a look at the lower right, were i overlayed a slight red/white pattern that shows the 8x8 pixel blocks.


You can see that with compression alone, you don't really get the 'look' that the typical EE on most titile (especially the heavy hitters) exhibits. They do have a few mosquito noise and blocking artefacts, but that is rather light and on TOP of actual EE.


The picture in the lower left is just that. A combination of actual EE and a reasonable compression ratio (yielding some mosquito noise).



Here is another example:

http://www.videophile.info/Misc/EE/T2_Protect.jpg http://www.videophile.info/Misc/EE/T2_Protect_mini.jpg
Click on picture to enlarge


I did this example some time ago. It fits here nicely. The picture in the upper left is an actual crop from a T2:UE DVD frame.


In the lower left, i created a little text-on-background pic, that mimics the original crop. Same source resolution of 720x480, btw, look just how perfectly clean a flat 480p response would look like!


The 2 pictures in the middle show light and heavy MPEG compression on that simulated pic. You can see lots of mosquito noise around the edges, but nother that really resembles contour tracing EE. Not even at the 'heavy' compression setting, mind you, which is much stronger compressed than the original T2 frame (look at the amount of mosquito noise).


In the top right, i used EE with a frequency and amplitude that simulates the halos in the original frame. See how the EE really actually mimics those halos and how it perfectly traces the outline. The pic in the top right is a bit unrealistic, since its not compressed, so i applied a reasonable amount of compression, to simulate the amount of compression that can be seen in the original.


As you can see in this practical example, at least 95% of the ringing is due to actual EE and NOT compression. And this is the way it is on most if not all EE laden titles. The ratio might decrease to 2/3rd EE and 1/3rd MPEG ringing on a FEW really bad encoded titles (read: compression mess), but the dominating artefact is alway actual EE.


If an MPEG encoder like the Sony yields this amount of ringing without any EE controls, then its simply the way its (mis-)designed (see first post).


Regards

Bjoern
 

·
Registered
Joined
·
1,390 Posts
I'm getting mixed signals, and quite confused I guess. Since there are arguments from the DVD mastering techs stating that they are not adding EE, and the EE doesn't exist on the masters, then the EE must added during compression. What can happen in those mastering studios is nearly limitless. Software types hardware problems, incorrect multiple compression passes... what I see from JPEG compression and from EE'd movies looks too similar to discredit.
 

·
Registered
Joined
·
1,746 Posts
Bjoern,

Thanks for the excellent dicussion. Guess I should have searched for this, but I thought I was being clever!


I was getting fed up with 'the studio insists they didn't add EE, but it's there so they must be lying' threads.


Dave
 

·
Registered
Joined
·
4,580 Posts
Having deleted one comment, let me say that this thread is for people who want to seriously discuss the issue of edge enhancement. It's not the place for frivolous remarks or comments meant to demean your fellow member(s). Thank you.
 

·
Registered
Joined
·
4,940 Posts
Quote:
Originally posted by Bjoern Roy

That has to be enough for today. Most will probably be asleep already.

[/b]
Not at all. Excellent posting!

MH
 

·
Premium Member
Joined
·
16,963 Posts
I wonder if display technology also plays a roll enhancing EE/ringing.


I say this because I have noticed EE is not much of an issue when using a digital projector. When I view the same disc on a properly setup crt with a good processing chain EE becomes more noticeable.

This is something I have noticed while playing with numerous projectors over the last few years.

Anyone else notice this.


By the way I am finishing a theater room in my home and Im installing a 9"crt "no digital" so this is not a bash on crt just an observation.



Any thoughts?
 

·
Registered
Joined
·
1,390 Posts
I've heard from several sources that an ananlog link (especially over 20ft)increases ringing. Is there a possiblity that the digital projectors you're referring to have a digital DVI link between them?
 

·
Premium Member
Joined
·
16,963 Posts
Aaron


No digital in just analog.


With crt I wonder with such a long video path within the projector via numorous cards ect if that adds to the cause.
 

·
Registered
Joined
·
24,883 Posts
Alan,

I don't know about all digitals but bad EE still shows up as bad EE on many. I still keep thinking that if the cables,cable length,processor, or PJ technology was at the heart then zero EE titles would not exist in that chain. I guess it is possible that it exacerbates or displays it in all it's glory but I don't think that it introduces it.


Art
 

·
Registered
Joined
·
5,318 Posts
Art, I believe you are correct, however I think that the device feeding the projector also has some effect.


For example:


I just recently installed a Radeon 9700 Pro, DirectX9 RC2, and built some new filtergraphs to use the VMR9 with ffdshow in the path as well. Using Sonic CP 1.5 current release.


(I also revamped my subwoofer/transducer setup, but that's another story.


Anyhow, as a test of the new subwoofer setup, I played TPM DVD, and I believe that the severity of the visible edge enhancement has been reduced from what I saw with the old DX 8 - overlay mixer setup. It's still bad, but not as offensive.


I viewed some of TPM on my other partition that is still using DX8 (same 9700 Pro video card) and DX9/VMR9 seems to be the significant factor here.


I do find the image with the VMR9 to be superior in resolution and smoothness to the older DX-8 - 8500 setup on all the DVD's I have watched.


Just my $.02


Vern Dias
 

·
Registered
Joined
·
4,649 Posts
Originally posted by Bjoern Roy:


"Most of the HD that is broadcasted today (or sold on D-Theater) is edge enhanced. Thankfully, the frequency bands that are boosted here are much higher, thus, the halos are mostly much thinner than on DVD. Sadly, often, EE is used to the extend that its still 'beneficial' (cough) at SD resolutions. Ugh. A nasty habit from the earlier days, where the primary purpose of HD master, was to produce snappy SD masters afterwards."


So even HD is being EE'd to death... :mad:


This practice sure harks back to the early days of CD mastering when audio engineers continued boosting high frequencies in the same manner and rates they did in an attempt to compensate for the natural HF losses that occurred with the vinyl archival platform, which predictably resulted in screechy, shrill, over-bright high frequency ranges. It wasn't until the practice was omitted by the most savvy engineers that things began to greatly improve sonically.


I don't hold the same hope for digital video software as we're dealing with an essentially inherent lossy, highly compromised format. Your thesis only helps enhancing my thoughts and feeling about the situation even more.

To tell the truth, I've lost a great deal of enthusiasm for the DVD, although I continue collecting movies in this format.

I am holding on to the HD format only because it indeed provides a more cinematic experience spite of everything else, but highly decry the practice and wish it stops altogether.


As far as I am concerned, "EE" is without a doubt one the greatest culprits that deny us sampling truly immersive cinematic experiences via our home theater systems. It should be far more of a complete thrill ride... :(


-THTS
 

·
Premium Member
Joined
·
16,963 Posts
Frank


I was told by Mark Rejhon that ffdshow has a feature for eliminating

EE. You can dial it in ( I assume frequency) and eliminate it.

Something similar to whats built into the Teranex. He told me ffdshow was not for the novice so I have stayed away from it.
 

·
Registered
Joined
·
976 Posts
Quote:
Originally posted by Alan Gouger
I wonder if display technology also plays a roll enhancing EE/ringing.


I say this because I have noticed EE is not much of an issue when using a digital projector. When I view the same disc on a properly setup crt with a good processing chain EE becomes more noticeable.


Any thoughts?
EE is much easier to see, even on a 58" to 65" RPTV driven by 9" crts, than on an 8 ft or larger picture projected by a dlp. I always wondered why this observation rings true, and can only conjecture why.


Crts typically have much higher luminance bandwidth than a dlp. If you want to see a finely detailed picture, watch it on a good crt. Dlp systems (signal processing, dmd, bulb, lens, and end result on screen) are also non-linear in luminance space, especially for variations at low IRE values; two anolgies to audio might be distortion artifacts that vary with amplitude, and listening music on a bandlimited system. Sounds great when the S/N is high or the signal is within the sweet part of the frequency domain, but smorges the signal elsewhere. Now, you can still see EE on dlps, but something in the technology is squishing it a bit.
 

·
Registered
Joined
·
288 Posts
Quote:
Dlp's are also non-linear in luminance space
Is that based on measured data? I would think that the PWM technique used to produce grey-scale from a DMD chip would make it a solidly linear device. In fact one doesn't have to bugger around with gamma curves to account for nonlinearity in light output like one has to do when using CRT.
 
1 - 20 of 38 Posts
Status
Not open for further replies.
Top