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

[nolesrule]

Let me add thanks to all those who participated in the restoration. I know you didn't do it alone-- couldn't do it alone. I certainly didn't intend for my post to come across that way.

I enjoyed the experience so much that I wanted to personally thank someone who was directly involved. Your name was the one I recognized because of your participation here and at the Digital Bits (and was at the top of the credit list).

 

[Robert Harris]

To Mike Maloney...

All of the technical gargon aside, I'm still confused.

From what you've explained, it sounds as if all long films need to suffer on DVD, especially if they also contain a DTS track.

I haven't seen this as being the case.

The high frequency image information in Cold Mountain looks to have been unceremoniously lopped off by someone or some piece of equipment or software at your post facility.

If other long films can make it to DVD without the removal of this high frequency information (which controls detail), why was it necessary to do it on Cold Mountain?

Is this a standard procedure for films of a certain length?

Cannot the film and the inherent bitstream be massaged into a higher end final result.

RAH

 

[TedD]

Mike Maloney:

I'd just like to add one more point here.

If, in fact the high frequency detail were simply rolled off, then I could easily use the DScaler sharpness filter to restore much of the hidden detail and produce a very enjoyable image.

It is the addition of gross amounts of Edge Enhancement, Ringing, Halos, or whatever you choose to call the effect that make the disc almost unwatchable on any kind of large screen display system.

This Edge Enhancement, Ringing, Halos, or whatever you choose to call the effect seems to be a standard part of almost every recent DVD transfer to come out of Walt Disney studios in recent months.

"The Happiest Millionaire"
"A Far Off Place"
"Cold Mountain"

are just a few of the recent Disney owned films that I have suffered through because of being plagued by severe Edge Enhancement, Ringing, Halos, or whatever you choose to call the effect.

Actually I can't count "The Happiest Millionaire", because 5 minutes into the DVD, I shut it off and went back to the anchor bay release which is as close to being a reference DVD as the Disney DVD is to being another "Highlander".

It would seem to me that 1 hour with the executives of Walt Disney using a large screen front projector and the DVD of the remake of "Moulin Rouge" and any recent Disney 2.40:1 DVD in an A-B comparison would be enough to convince someone in charge that something is clearly wrong with the quality of the DVD releases coming out of the Disney studio and that the current decision of "Content over Quality" needs to be reversed.

Heck, I'll even fly out there and do the demo myself.

I shudder to think of what Mary Poppins is going to look like under the current direction Disney is taking.

Well, I had to say it.

Moderator, you may ban me now.


Ted

 

[Mike Maloney]

Thanks to everyone who has participated so far.
A couple of quick notes before I get to the meat of today's post. The comment about the flame-proof suit was made because I'm a veteran of usenet news groups.

Also, I haven't said anything about edge enhancement because it really has me stumped. We truly do not have any EE gear in the plant. I'm doing some research and I may need to do a poll to find some common factor.

We do take this matter seriously and I am able to do some experiments to figure out what's going on.

So far I have been able to compare layer 0 of C.M. using the
following:
1) The original bitstream used in the DVD. It turns out that this bitstream had a 4.5 MHz low pass filter used. My bad, I apologize. However, see below.
2) A new encoding using the original encoder at the original bitrates and absolutely no filtering.
3) A new encoding using a different encoder at the original bitrates and absolutely no filtering.
4) A new encoding using the original encoder, no filtering and with bitrates set as if there would be no DTS track.

Tomorrow I intend to make an encoding using the original encoder and no filtering, but splitting last 35 minutes of the movie to the second disc. This will the effect of giving a lower peak rate than #4 but an ever higher average rate.

I haven't had a chance to thoroughly evaluate case #4 yet, but the first three cases are remarkably similar. Which, I admit surprises me. However, it is gratifying because it proves empirically what the theory suggests. That is: high frequency detail suffers as the encoder works harder. It's also a relief to me that two encoders behave similarly.

As to the question of why all movies of certain length don't show the same problems, if I understand the math correctly it is not just a function of the bitrate but rather the interaction of the bitrate and the spatial frequencies.
C.M. has a definite texture to it. It is that texture that is causing the encoders fits. As I said at the end of my last post, Last Tango should behave in a similar fashion.

I need to let the data digest for a while or I'm going to get a brain blister.

Cheers,
Mike

 

[Robert Harris]

To Mike Maloney...

I don't quite understand your comparison to Last Tango, which may be a joke that I'm not getting.

I'll toss something else out to you for digestion.

It is a known that a film element (sprocketed) that has been based upon or derived from a digital source will not survive as many generations of analogue duplication as a similarly produced analogue element.

Film reproduced to five or six generations based upon an optical film element can still look quite acceptable when projected after going through a continuous IP / IN set of stages.

Take the same element which has been derived from a 2k source, and it will begin to self-desruct several generations earlier.

By this, I mean that there is a perceived quality in digitally based film, which is more of an illusion than reality.

There was a similar situation with dye transfer Technicolor which appeared much sharper on screen than it acutally was, based upon its levels of contast.

As a project which came from digital files and went to film as a DI, mightt Cold Mountain be more of a problematic image than a film-based title?

Just an odd thought.

By the way, allow me to add that your experimentation in attempting to find a solution to this problem is appreciated by all at HTF and sets a most positive tone for your facility.

RAH

 

[TedD]

Mike:

Thank you for taking the time to research the issues that have been raised in this thread.

I have a possible theory on the EE issues, based on some research I did earlier to try to understand why my impression of the quality of transfer on the new "Happiest Millionaire' was radically different from RAH's impression.

In the course of testing, I discovered that interlaced displays tend to mask EE that is applied on the vertical axis to a large degree. While HM was totally unwatchable on my 720P scaled FP, it was almost watchable on a 27" interlaced TV set.

Just as a level set, and so we don't go down the path of blaming the our display devices, both DVE (Digital Video Essentials) and many of my DVD's show absolutely NO EE, or anything vaguely resembling it. On my FP / HTPC combination, the resolution test patterns on DVE are clearly rendered out to the limits of the signals recorded on the disc with NO sign of EE or ringing present on the transitions.


Is it just barely possible that all the monitors used for the transfer, mastering and QA just happen to be interlaced displays?????


If so, could you please use a non-interlaced display, preferably an HTPC feeding a large 720P (or 1080P) front projection display to take another look at the CM (or HM) DVD (and the mastering elements as well) and tell us what you see?

Thanks again for the attention.

Ted

 

[Mike Maloney]

Hey Robert,

While I had the HD Master of C.M. I made a down conversion of my own so I could rule it out. The downconversion I made was identical to the compression master we received from the client.

The reference to Last Tango is that it has a distinct texture to it as does C.M..

I'll try to keep the jargon to minimum but it's difficult to eliminate it entirely.

To Ted:

I think that the displays we are using (on the consumer side) is a huge variable. From my experience, CRT based displays should be more forgiving in general, due to the Gaussian response curve (sorry 'bout that).

I really shouldn't write about this before coffee.

You can help by performing an experiment for me. How does HM look if you don't scale it on your FP but display it at it's native resolution? Do you see the same artifacts, but at a different scale? At the native res. does it look similar to the CRT, or is there something about the DLP or LCD in your FP unit that is interacting in unexpected ways with the DVD?

The questions come from sampling theory, which since I still haven't made it to the coffee pot, I don't want to go into.

Cheers,
Mike

 

[TedD]

Mike:



Yes, but I am not asking you about the consumer side, I am specifically asking about the monitors used in the telecine bays, the disk authoring workstations and the QC area. If they are interlaced, then they are hiding many of the defects that eventually wind up on our DVD's.

OK, Here we go on the captures:

(Note: All these captures were taken from the output of the NVidia FWMM V4 decoder, acknowledged by all who have used it to be the ultimate in DVD PC decoder software. The signal was then fed into the FFDShow Directshow filter and either scaled or not, then captured by the capture module in FFDShow.)

Here is the DVD Resolution pattern at 2x scaling:

http://webpages.charter.net/tvdias/DVE_res.jpg

First, for reference purposes, the Anchor Bay transfer at 2x scaling with Lanczos 4 with luma sharpening set to .5:

http://webpages.charter.net/tvdias/AB_2X_scale.jpg

Next, the Disney transfer using the same scaling and sharpening parameters:

http://webpages.charter.net/tvdias/WD_2X_scale.jpg

Now, the Disney transfer rendered at the native 480 x 720 resolution:

http://webpages.charter.net/tvdias/WD_no_scale.jpg

Note that upon first glance it looks better than the scaled version, but if you save it to your PC and zoom in on it to give the same size image as the scaled images above, you will see it's pretty poor as well.

On all the captures taken from the Disney DVD, notice the three distinct lines at the top of the image just below the letterbox bars. On a transfer with no Edge Enhancement, Ringing, Halos, or whatever you choose to call the effect the letterbox bars will have a clean edge with no additional lines.

Here's another example of the EE that seems to defy all attempts at identification and resolution and that plagues many DVD's:

http://webpages.charter.net/tvdias/WD_lookatthebox.jpg

This capture was also taken at native DVD resolution, so you may want to save this one as well and zoom in on the box. Pay particular attention to the cover of the box which has at leat 4 distinct lines that don't belong there, possibly created by filtering somewhere in the signal path between the film element and the final product.

One last capture, I apologise for the incorrect AR

http://webpages.charter.net/tvdias/AB_lookatthebox.jpg

Now check out the same box in the AB transfer.

I hope this helps you to identify the issues, although it seems like it may be too late to rescue the DVD's that have already been released.

As you can see, I don't need to even begin to bring my DLP into the mix to see the issues, and I also have a CRT which shows the same issues. I never scale on a solid state projector, I always run a pixel-perfect resolution out of the video card. In this case, it's 720x1280 to the DLP.

See my setup here:

http://webpages.charter.net/tvdias

Thanks again for taking an interest in our little problem here.

Ted

 

[Robert Harris]

I'm not certain that HP ever came up in the thread.

If anyone (Ted?) has similar captures of Cold Mountain, they could be most interesting.

RAH

 

[TedD]

RAH:

I'll do some Cold Mountain captures tomorrow and post them here.

Mike specifically requested that I use it for these tests.

Ted

 

[TedD]

Mike:

I am going to apologize in advance, but I am calling this one the way I see it.

OK, Here's the promised capture for Cold Mountain:

http://webpages.charter.net/tvdias/CM_cap.png

I am only providing this one capture, because it is apparent from this one capture that it is pointless to spend more time with this title.

Here is my assessment of the transfer:

It is, without question, a transfer totally lacking in any detail, of possibly marginally better than VHS quality. This is no doubt due in large part to the 4.5 Meg LP filtering applied to the video.

There is so little detail present, that no amount of post-processing on the PC can salvage this transfer to make it suitable for viewing at a distance of any less than 5 to 6 times the width of the image. This makes it totally unusable in any respectable home theater environment.

My feeling after working with this DVD is that I might as well have bought the VHS release, because that's what I got, a VHS image quality release.

Ted

 

[Robert Harris]

Ted...

Thank you for posting.

This is precisely what I'm seeing on four (count 'em), four different monitors.

I've viewed samples of Cold Mountain on an ancient, but still worty 20" Sony XBR, which I use in my office; a 40" Sony LCD rear projection; via a Runco on a 110" Stewart 1.3 gain screen as played by a Pioneer Elite, and lastly on my new favorite, a Sony 34" CRT XBR910. I began with the 910 and played through a Pioneer Elite.

The Elite was a six year old unit, and thought that something might be unfair or amiss. For the this test I checked out virtually every moderately priced (non-entry level) player out there, and decided to bring in the Integra 8.3 to see if there might be any difference.

There was.

The image was even more accurately reproduced.

A sampling of other titles ---

Gladiator...

The Haunting...

Master and Commander...

and then back again to Cold Mountain.

Which in comparison to those above...

looked precisely like the sample frame kindly (and efficiently) reproduced by Ted D.

What I'd love to see in comparison would be the same frame derived from the digital files, followed by samples through equipment other than that used for the specific compression.

Might we get test samples from your post facility or directly from Miramax post or Disney?

I'm looking at this problem as a non-partison situation, in the hope that the industry can do better.

Now that you're saying that you perceive a problem, can we not find a solution which can be a new guidepost for studio post people via which they can more accurately determine what can and cannot be included on a DVD, and the best methodology for getting there while still holding the quality of the cinematic experience...

which in the case of Cold Mountain, was not exceptionally high. Please keep in mind that film frames had already be resolved to 2k as either a printing negative or an IP. Precisely how much information is lost generationally going from the film Oneg to a 2k file and then out to another 2k file, and then finally to film is a most interesting question when compared to an Oneg to an IP to a dupe to a print.

In the end does it matter on screen?

It may depend upon the quality of the projection lens and the alignment, steadiness and optical system of the individual projector.

RAH

 

[Michael Caicedo]

This confirms my assessment of the CM dvd. I have discussed this transfer with friends prior to reading this thread. Just think about watching this on 119" screen at 1.3x seating distance.

 

[Max Leung]

Ted, I cannot access the screenshots from your webhosting site through Mozilla. I can view it with Internet Explorer. Can you tell your webhosts to not deny access to non-Microsoft-monoply browsers? Um. Unless there is some sort of popup ad/cookie/Javascript or other weird thing causing the security on Mozilla to freak out?

Anyways, this is a very informative thread. Mike, I hope you get to the bottom of this and find a solution. I have not bought any Miramax disc, except Kill Bill and a couple of MGM DVDs. I now have a 720p front projector and have put all Miramax and Disney purchases (except the CG movies) on hold until this issue resolved.

 

[george kaplan]

Max,

I'm using Mozilla and access them fine, so that's not the problem.

 

[TedD]

Max, my web site is charter@home and I have no influence over what or how they provide web hosting.

That said, it sounds like it may be a different issue for you, since at least one other Mozilla user has access to them.

Sorry for this little hitch,

Ted

 

[David Grove]

I wonder if the low-pass filter may have been implemented as a "brick wall" filter in the frequency domain (in other words, by just zeroing out all DCT coefficients above the cutoff). Implementing "perfect" filters in that manner does knock out the intended frequencies, but has the notorious side effect of introducing ringing artifacts (due to Gibbs phenomenon).

On the other hand, this is so very well understood in the field of digital signal/image processing, that I'm almost embarrassed to mention it. The filtering almost certainly would not have been implemented this way. But, just in the interest of getting ideas on the table, even if they are highly unlikely, I mention it with all honorable intentions.

Regards,

DG

 

[Mike Maloney]

David,

Thanks for that post; it's suggested another round of experiments.

A few days ago I wrote that I might need to poll the forum to gather data and that time has come. I need data on the edge enhancement issue. So would the group give me examples of movies that have the least edge enhancement and those that have the most enhancement. To limit the number of titles, please limit your list to a maximum of five titles in each category. Also, please limit your comments to edge enhancement. (My local rental house is gonna love this.)

Thanks,
Mike

 

[TedD]

Least EE:

Moulin Rouge (Remake) (My reference transfer, spectacular detail and NO EE ) Fox
North by Northwest WB
Parent Trap, The (Original) Disney
Pleasantville New Line

Most EE:

Sound Of Music Fox
What Dreams May Come Polygram
Pearl Harbor (either 2 or 4 disc set)
Ted

 

[Eric Stewart]

I'm afraid a lot of this discussion's a bit over my head, but I'm extremely interested in what Mr. Harris says, in his latest article, about recent Miramax DVDs' inadequacies.

As these are apparently synonymous terms Mr. H. mentions, can anyone fill me in on exactly what the "post house" or "post facility" does?

Is it the video transfer? The "softening" of the image, removing grain? The "electronic enhancement" or edge enhancement? Is it video format conversion, as downrezzing from a 2K transfer to 480p or 480i? Is it MPEG-2 compression? DVD authoring? Or perhaps all of the above?

In other words, does a single facility do all of these things, or is there a transfer of responsibility between the earlier and later steps? If the latter, where is the line drawn, and on which side is the "post house"/"post facility"?

Thanks in advance to anyone who can enlighten me on this.

And thanks to Robert Harris for continuing to educate us all.