Robert Harris on The Bits - 8/3/04 column - OFFICIAL THREAD (1 Viewer)

[ToddF]

I rented 'Cold Mountain' today and I agree with the many posts citing poor video quality. Yes, the image reminded me of watching VHS. I think whatever process that was used to make this DVD presentation needs to be seriously re-examined. I own a Toshiba 57HDX82 widesceen TV.....

 

[David Grove]

Eric,



I have to admit I'm not really up on the Lanczos sigma factor.



Regarding the relation of the DCT to Fourier-- you can think of the DCT as a sort of special case of the Fourier Transform. All the Fourier stuff applies.







All,



I have to admit to getting a little carried away on the Gibbs stuff. I was pretty much extemporizing, and, on review, I don't think I should have "blamed Gibbs" for all the potential bad effects of brick wall filters.



All that stuff about sharp transitions (square waves) producing bad stuff, like ringing, because of Gibbs is true. But, there is a ton more to the story, and undesirable effects from brick wall filters are more due to the way frequency domain filtering works.



Specifically, multiplication in the frequency domain (in the "brick wall" case, this is multiplication by a "step function") manifests itself as a much more complicated operation in the original domain. It's almost magic. Indeed, this is one of the main motivating factors for using this stuff in the real world-- A desired operation that is complicated (expensive) to perform in the original domain is very easy to perform in the transform domain. The recipe is easy as 1, 2, 3. 1) convert the data to the transform domain; 2) perform an operation (multiplication)that is simple (cheap) in the transform domain; 3) convert the data back to the original domain. Voila! you have accomplished the very complicated operation (in our case that would be filtering an image), but by using the transform it was done faster and cheaper. Anyway, the equivalent "complicated operation" in the original domain can also cause "echoes".



It has been quite a while since I was active in the digital signal processing field, and without some study and review, I'm not sure I trust my memory for more than just the bare fundamentals. So, I apologize for making Gibbs appear to be the only bogey man.



In any case, you can be sure that, to the folks who really developed this stuff (MPEG, etc,), this is nothing new. They were fully versed in DSP topics, including Gibbs, filtering, compression, etc. You can be sure that the theory was carefully considered.



However, it may be possible that some folks who apply the technology may not always have a deep understanding of the tools they are wielding.



Just as in many professional or technical fields these days, the easy commercial availability of powerful digital signal processing software is a double-edged sword-- It is easy for folks or firms to pick up a little knowledge and then acquire hardware and software tools that are amazingly powerful, easily operated. It might require additional, substantial effort (expense) to wield those tools to best advantage.



Truly a little knowledge can be a dangerous thing.





DG

 

[george kaplan]

You think things are bad now, just wait til they jump on the latest corporate trend and start outsourcing it to India or elsewhere.

 

[Eric Stewart]

Cassy,

quote:What does this tell me? These films are transferred into High definition and look great. They are then downconverted for DVD and StarzHD and that is where the EE is added. From there the downcoverted master is compressed for DVD and released as the abominations we've seen in COLD MOUNTAIN, etc.

EE is being added:



(1) following the HD transfer

(2) prior to the video-format downconversion ...

(3) which precedes the MPEG compression.



That seems to be the asumption here. I don't dispute it. Not for a minute.



Not for a nanosecond.



But then it may still be the case that the Gibbs effect which David Grove speaks of as the outcome of "brick-wall filtering" that is done in conjunction with the MPEG compression can make it worse.



There may be two culprits here, not one.



On top of that, I've just stumbled across a web page ...



[url=http://www.asahi-net.or.jp/~rt6k-okn/ddp/digital.htm]http://www.asahi-net.or.jp/~rt6k-okn/ddp/digital.htm[/url]



... that claims, "Chemical processing of film also produces an edge enhancement effect." Scroll about 1/5 of the way down the page to a photo of the edge of the moon set against the darkness of space. There seems to be edge enhancement going on. Is it the same sort about which we've been complaining?



I recently saw an anti-corporate documentary, The Corporation, projected in a movie house. Some of its scenes showed the likes of electrical power lines and their support towers set against a diffuse white sky ... and there appeared to be edge enhancement in the image! I wondered if these shots might not have been passed through the digital-video domain en route to final celluloid.



But maybe it was an EE effect produced by chemical processing of film.



Mr. Harris's expert input on this particular topic would be very much appreciated.

 

[Eric Stewart]

RAH,



I think it would be wonderful if a certain well-known film restorer/preservationist with ...



(a) eagle eyes

(b) technical smarts, and

(c) impeccable credentials for having resurrected films like Vertigo, Lawrence of Arabia, and My Fair Lady



... were to be invited by an important film and DVD distributor such as Disney/Miramax to sit in on the process of trasnsferring and compressing an important motion picture for home video. Then he could learn first-hand at what point the gremlins creep in ... and maybe help keep them from creeping in in the first place.



Coverage of the event would make an interesting feature story for one of the home-theater enthusiast magazines. So perhaps if the well-known film restorer/preservationist had any contacts at such a publication, he could persuade them to broker such a deal.



Is there any chance of that kind of thing happening? It would definitely eliminate excuses, discussions and hyperbole, at least for that one DVD. And by raising consciousnesses, it would move the whole state of the art ahead by leaps and bounds.

 

[Eric Stewart]

David Grove,



Again, thanks for all your discussions re: digital signal processing. OK, Gibbs is not the only bogey man ... that's established now, and it's important. But one thing I'm not clear about is whether the digital processing of which you speak ...

quote:... multiplication in the frequency domain (in the "brick wall" case, this is multiplication by a "step function") ...

... is an intrinsic part of the MPEG compression step, or is it a separate step entirely? If the latter, at which stage along the way from the original film-to-video transfer to the finished DVD does it most likely occur?



*** Edit: Let me edit my original post to clarify the question: Is the brick-wall filtering or whatever done in a separate, explicit filtering step, or is it implicit in the Discrete Cosine Transform, quantization, etc. that are part of any MPEG compression? ***



Cheers,

 

[Robert Harris]

To Eric Stewart...



While it might make for nice press, there would be little sense in said archivist sitting in on said work, as there are many more individuals with skills in said arena far above those of said archivist.



While I'm not suggesting that the fox be given the keys to the henhouse, there are certainly enough highly motivated people involved that the problem can be solved.



RAH

 

[Max Leung]

Eric, just to comment on The Corporation, but I think that film may have been shot with digital cameras. Damn, I should have asked the director what equipment he used, but I didn't want to interrupt his impromptu "Burning Man" festival slideshow while we were at the bar. I wonder if he would have been sober enough to answer the question anyways.

Most low-budget documentaries nowadays are shot with digicams; I seriously doubt we can use them as benchmarks for quality anymore!

 

[Patrick McCart]

Here's a comparison between a split-screen shot in Kill Bill Vol. 1 and Looney Tunes: Back in Action.

 

[ChristopherDAC]

It is possible to generate edge halos in on film; working with silver emulsions, however, is so well understood after more than 150 years that only total incompetency or deliberate intent would be likely to have such a result. For instance, I suspect overexposing a scene of sharp bright-dark transitions, onto low-contrast film, and then overdeveloping, would be an example of a photochemical process which could lead to pigment depletion and leaching, generating haloes. BUT: what we are seeing on DVD does not look much like the effect of such a processing step; the effect on these DVDs is so egregious that a film which looked like that would scarcely have gotten past the "dailies" stage, and would have been remarked upon in the theater; the fact that different DVDs from the same source and even the same master exhibit different levels of the phenomenon indicates that it does not arise primarily in the film part of the signal chain; and the fact that the effect is typically most marked in the horizontal direction suggests that some of it was introduced during the video part of the chain, specifically at a step when the video signal is a 1-dimensional stream [either analog electrical signal or digitised bitstream, but not the 2-D MPEG matrix]. Unfortunately there is little guarantee that a similar class of processing equipment and operators, with just as many problems, will be employed in the forthcoming High Definition signal chains -- considering that nost HD work up to now has been done by a very limited number of operators and production houses, when it finally becomes a mass-market commodity we may see the same effects as now with DVD, particularly if [as seems likely] the HD recording format is rushed to market in the same way DVD was.

 

[David Grove]

quote:question: Is the brick-wall filtering or whatever done in a separate, explicit filtering step, or is it implicit in the Discrete Cosine Transform, quantization, etc. that are part of any MPEG compression?

To the extent that MPEG zeroes out all coefficients above a certain threshold, while retaining all coefficients below it, it is a brick wall filter.





DG

 

[Dave Mack]

To me "Gangs..." is still the worst. The opening scene in the snow is atrocious. Unwatchable. NO detail and forcefields. Than I put on State of Grace with NO EE, tons of detail.. Interesting how a little catalog title looks so much better.....

 

[PeterTHX]

Forgive me for skimming the thread (after 3 pages...) I can't believe that NO ONE has brought up the fact that someone's DVD PLAYER can have a huge effect on presentation quality, especially on large screen TVs.



Cassy was one of the more vocal opponents of the "Cold Mountain" DVD. Her player is the Panasonic RP-91. From HomeTheater HiFi's progressive shootout: "All in all, the RP-91 is a good machine, and worth considering for its many great features, but if deinterlacing performance is very important to you, you may want to look elsewhere."



I have had opportunities to "shoot-out" DVD players, with some of the finest TVs available. The quality of the DVD player's Video D/A and Progressive chip can sometimes have dramatic effects on a presentation. I'm not at all fond of Panasonic DVD players, and some borderline DVDs were indeed "unwatchable". Put in a Sony with the DSV 12bit-108MHz chip and the picture was often smoother, more detailed, and VERY watchable.



So, this may very well explain why some people find some DVDs more "objectionable" than others. I haven't seen "Cold Mountain" and it may very well be a poor presentation. At the same time, the reason some reviewers found it OK, and people like Cassy who hated it with a passion is probably the playback hardware.

 

[Cassy_w]

Peter, with all due respect, the RP91 is an outstanding player. Certainly the best we have owned (all around). The Miramax and Disney titles clearly look inferior to pretty much everything else out there. I can pop in MATRIX REVOLUTIONS or PITCH BLACK or STAR TREK II and be completely satisfied. But if I try and watch COLD MOUNTAIN or GANGS OF NEW YORK or HIDALGO, the picture is ugly, low in detail and high in halos.

The Miramax problem cannot be explained away by using a player excuse.

By the way, we also have an RP62, which is even better than the RP91 in regards to deinterlacing. We moved it to the bedroom, since it doesn't have scaling.

 

[Reagan]

Deinterlacing would have nothing to do with D/A conversion or loss of high frequency detail or EE.



Unless everything I've read on the shootouts is wrong...



-Reagan

 

[Eric Stewart]

To one and all,



We have proposed in this thread several possible explanations for how ringing/edge enhancement intrudes on DVD images like that on Cold Mountain, and also for how there may be a loss of the high-frequency information which encodes fine detail.



The all have one thing in common … filtering.



It comes as no surprise to most of us that a video signal can be and will be subjected to explicit filtering, very possibly (as ChristopherDAC says) “when the video signal is a 1-dimensional stream [either analog electrical signal or digitised bitstream, but not the 2-D MPEG matrix].”



But it may come as a surprise to some of us that MPEG compression, though not usually thought of in this way, is itself a kind of implicit filtering … David Grove’s main point. In this particular case, it is “the 2-D MPEG matrix” that is in fact the locus of the filtering.



Either way, it’s still filtering.



Sounds like the mere presence of filtering might be bad, however it might be accomplished, right? Wrong! For one thing, frequencies at and above the Nyquist frequency (one-half the sampling frequency) absolutely must be discarded, or pesky aliasing will occur … as in a false moiré pattern. In sampling theory, an alias is a waveform at the wrong frequency, constructed in the process of trying to reconstruct the original waveform from a set of digital samples. Sampling at below the Nyquist frequency avoids this type of alias.



So some kind of low-pass filtering is a must … but it can also introduce artifacts.



There are several possible types of real-world filters (I’m summarizing Poynton’s book, chapters 16-18). They may work in the spatiotemporal domain — i.e., in two-dimensional image-frame space and/or in one-dimensional time — but they can be expressed also in terms of their characteristics in the frequency domain … as in how many cycles per unit time a sampled waveform has.



Versions of the Fourier Transform can be used to convert the mathematics of the filter from the spatiotemporal to the frequency domain (see p. 146). Looking at a particular filter in the frequency domain can reveal what artifacts it introduces. When a “brick wall” filter is looked at in the frequency domain — by means of deriving a Fourier Transform of its behavior in the spatiotemporal domain — the filter’s so-called “magnitude frequency response” (MFR) or “frequency spectrum” (FS) typically reveals that this “brick wall” filter necessarily introduces a “ringing” at certain frequencies.



Other, more sophisticated filter designs — or digital signal processing stages such as MPEG encoding which are not explicitly “billed” as filtering — can do much the same thing, if not always to as obvious an extent. If an MFR/FS of a particular MPEG compressor, using particular quantizing coefficients, could be derived, it too might reveal “ringing.”



“Ringing,” as in edge enhancement.



“Ringing,” which is the exaggeration of certain high frequencies representing fine detail, is not the only possible filtering artifact. A filter may, for example, attenuate intermediate frequencies … not because it was used expressly for this purpose, but perhaps as an unintended side effect.



That’s key, so I’ll repeat it: filters have unintended side effects.



When the filter is not explicitly called a “filter,” but is instead called something else like “an MPEG encoder,” the side effects can be truly insidious.



So … going from film to DVD, an image goes through a lot of explicit and implicit filtering stages. Several of them may contribute to the “ringing” we’re calling edge enhancement.



Some of edge enhancement may be done on purpose, as if for a transfer that is bound for VHS video.



Some of the edge enhancement may be a wholly unintended side effect.



And any and all of the types of filtering which produce edge enhancement, intentional or not, may occur during the life history of the video bound for a particular DVD.



Which brings us back to something Robert Harris has emphasized over and over, all along: solving this problem will require a high-level corporate commitment at Miramax (not to mention other studios). Only the high muckety-mucks can shepherd a DVD-in-development through all its “explicit and implicit filtering stages” — which very likely are done at two different post facilities, one for the transfer and one for the compression — and insist that the “filtering” always be done in the best possible way with the best available equipment.



Consciousness must be raised at all levels … but especially at a corporate level sufficiently high to avoid the blind-men-and-the-elephant problem. One blind man holds the trunk, one a leg, and one the tail – and all describe the elephant differently. Only the high muckety-muck perched atop the elephant knows the whole elephant.



But does the high muckety-muck possess enough “ridership sensitivity” to notice when the elephant is lame?



Keep up the good fight, RAH!

 

[Robert Harris]

Eric,



I couldn't have said it better myself.



To Ron Epstein:



Would you kindly pin this thread to the top after the new column comes out.



With studio and post facility readership of HTF, and with a general sense within the studios (more than this readership might realize within the studio technical area) that quality is of importance, I believe that we have a real possibility here of advancing the quality of the home theatre experience.



In this way we can keep the African elephant distinguished from its Indian counterpart, with all details intact, and not end up with "elephantus genericus."



It must also be noted that the individual sitting atop the elephant must also control the purse strings, and dependant upon the importance of the elephant, must decide whether or not to give it the treatment it needs to permit to walk properly.


Elephants aside, it would also be a nice thing if detail was left in the image portion of home video softeware.



RAH

 

[TedD]

quote:I can't believe that NO ONE has brought up the fact that someone's DVD PLAYER can have a huge effect on presentation quality, especially on large screen TVs.

That's probably because most, if not all of the captures and comparisons in this thread are being made from digital captures directly off a PC DVD drive using FWMM4 Directshow filters which are the current state of the art decoders and FFDShow. In fact, if you will notice, most of the captures are in PNG files to avoid any possibility of corrupting the capture.



These captures are truly the closest one can get to what is encoded on the DVD without picking up the original digital 1's and 0's from the DVD surface and tasting them.

Ted

 

[Ed St. Clair]

quote:You think things are bad now, just wait til they jump on the latest corporate trend and start outsourcing it to India or elsewhere.

Seen a few disc with "Made in Mexico", recently.

Scared me!

Any comments?

 

[ChristopherDAC]

Field guide to Elephants:

Loxodonta Africana: Flat head, Large Ears.

Elephas Indicus: Bulbous head, Small Ears.

Any questions?

The truth is that in order to cram 133 minutes of video per side onto a 12 cm disc read with a 650 nm laser, a compression ratio of greater than 60:1 is necessary. This simply cannot be accomplished while preserving all of the original picture information. It is therefore necessary to "lose" some of this information, either spatial detail or temporal detail or both. This is of course a filtering step.

What is desirable is to filter out information which is not readily percieved, without affecting what remains in any perceptible fashion. MPEG-2 encoding is a tool for achieving this purpose, which [as I am sure most on this forum can attest] can be made to work quite well providing it is done carefully. There may be other tools which can work better, but MPEG-2 is the one we are stuck with. The question which we must answer is how we can consistently make the most effective use of it.

The answer seems to be keeping a close watch on the transfer-encode chain, with careful inspection of the picture quality and adjustment of the encoding parameters at every step, combined with a willingness to seek out the best performing equipment, and the admission that the same equipment will not treat every source element equally just as much as every piece of equipment will not treat the same source element equally. This is what the studios and mastering houses must be made to do, and what they have largely so far failed to do, resulting in the "hit-and-miss" quality aspect of DVDs. There seems to be no equivalent in DVD for the consistency and quality which THX certification was supposed to provide for LaserDiscs; at least, while THX certification does exist for DVD its effect is debatable. One may wind up with an encode chain which passes THX reference signals better than good-looking pictures!

The other option would be to return to a LaserDisc type format which preserves all the original detail at the cost of being, in the general opinion, expensive and inconvenient. I am certain that with blue lasers and proper signal-encoding techniques a 30 cm optical disc could be made to store 30min or perhaps even 75 of uncompromised 1080i video per side; but would anybody buy it? And such a thing would still be at the mercy of the processing and encoding chain. No, I doubt that we will be seeing a larger disc or anything like it any time soon. We must concentrate our efforts on improving the performance of the current standard, so that all discs approach the level of th best discs as much as they can.