A technical analysis of Manga's
Wings of Honneamise DVD
DANGER! Long load
In order to show the problems with this DVD I have had to include full-size high-quality JPEGs and lots of them.
This page is over 2 megs and will take a while to load.
A few days ago I posted a message in the Anime On DVD forum regarding the quality issues with the Wings of Honneamise DVD. After a brief flurry of replies, it became obvious that I needed to go a bit deeper into the details of the problems that plague this release. To make things a bit clearer, I have included as many images as possible to illustrate the issues.
DISCLAIMER: I am no expert. I am simply a programmer and DVD enthusiast. I obtained this information using tools that are probably illegal to use in the USA (I myself am not a US citizen and do not reside in the USA).
First of all, I watched the disc on my Sony 27" Wega with a Pioneer 525 DVD player. The video quality wasn't as bad as I have heard some people describe it, however, it was nowhere near DVD quality. Still, it was watchable.
Next, I pulled off the original VOB files and proceeded to walk frame-by frame through various spots in the movie. Right off, some things were readily apparent.
Problems with the source:
A lot of people have speculated that this DVD was created from a laserdisc master. After viewing a dozen different sequences on the disc, it is quite obvious that whatever the source material for this release was, it most certainly was NOT any kind of original film (24 fps). What we see on frame-by-frame analysis is that the entire movie is interlaced. BADLY interlaced. The source was clearly either VHS or laserdisc-based (30 fps). There are obvious telecine artifacts all over the place. Worse, this source was captured incorrectly. So now you've got an interlaced capture of an interlaced source.
To illustrate this, I picked a sequence of 5 frames from the first WOH VOB file that represented the kind of problems that this stream has. Here they are,unmodified from the original DVD (no resizing or deinterlacing filters applied. Reference iDCT was used).
From the above, we can see that frames 2-5 all display some degree of interlacing. Frames 2 and 4 are particularly bad while frame 3 and 5 show interlacing around the pockets of the jacket and in the hair. At first glance a lot of DVDs look like this if you do a frame-by-frame analysis of them. Interlacing is perfectly normal for 30 frame-per-second playback. To a somewhat more experienced eye, however, this interlacing looks a bit odd. For one thing there's an awful lot of it -- for this kind of video only 2 out of 5 frames should be interlaced. To understand why you need to have a basic understanding of a process called...
Telecine (or 3:2 pulldown)
When a normal movie is shown in the theatres it is shot and played back on 35mm film at a rate of 24 frames per second. Our NTSC TVs, however, play back video at 30 frames per second. In order to watch 24fps movies on our 30fps NTSC TVs a conversion must be performed. This method is called Telecining (televised cinema) or 3:2 pulldown. The process is fairly straightforward, but can be a bit weird to explain. Lets start out with the math:
24fps / 6 = a sequence of 4 frames.
We need to get to 30 fps, so
30fps / 6 = a sequence of 5 frames.
Okay, so to go from 24fps to 30fps, we need to come up with an extra frame every 4 frames. There are a number of possibilities here, but the movie industry set on the Telecine standard. Heres what happens. Certain fields are repeated a couple times so that we get an extra frame out of this 5 frame sequence. There are a number of nasty method of doing this, but it is most simply explained the following way:
Youve got these 4 frames from the original movie:
A B C D
Now we split the odd and even lines apart so that we can make 2 fields out of each frame but theres a twist. Frames B and D are going to get repeated one more time than the others.
A A B B B C C D D D
Now we take each of the resulting 2 fields and combine them into single frames.
AA BB BC CD DD
So what were left with are 5 frames. 3 of which (AA, BB & DD) are exactly the same as the original source, but 2 (BC & CD) are mangled versions of the master "C" frame. This works out fine on our TVs but can make dealing with certain video sources rather interesting. This is just one way of handling pulldown. Although the basic concept is the same, many companies will vary the order that they do it in. Also, its not uncommon to find that some movies will have HALF the field in one frame and half in another
Do that 6 more times, and youve got your 30 frames-per-second-NTSC-happy video clip. If you didnt understand the above then read it again. Keep on reading it until in sinks in. It took me a few times to get it, too.
So, at some point in its life WOH was telecined so that it could be displayed on TVs. With DVDs this is usually done through the setting of special flags that are read and interpreted by our DVD players. This allows the original video to be encoded at 24fps on the DVD but played back properly at 30fps on our TVs. For quality reasons, it's better to do it this way, although it is also acceptable to perform the telecine in analog and then encode the resultant 30fps stream on the disc. I had assumed that this is what had been done to WOH. It turned out that I was quite wrong, however...
Just as you can perform telecine on a DVD, you can also reverse the process. This is commonly called Inverse Telecine or IVTC. The IVTC process attempts to find the original 24 frames that are interlaced into the 30 fps stream. I decided to perform an IVTC on WOH using a program called TMPGEnc. TMPGEnc's IVTC function is unique in that it lets you preview the selected frames before you commit to encoding them. After running the above five frames through TMPGEnc's IVTC algorithm we got the following 4 frames:
At first glance it looks like the inverse telecine worked. All of the obvious interlaced frames are gone, replaced with nice progressive images. But look closer. His hair is still interlaced. The jacket pockets and folds of his coat also exhibit strong interlacing artifacts. So what's going on here?
I have seen similar results of an IVTC before. Slayers The Motion Picture, Evangelion 0:1 and a few others (notably, all by ADV) all exhibit the same kind of interlacing patterns when the telecine is reversed. The cause for this is simple. Instead of starting with the original film print, Manga has taken some kind of video source that was already interlaced to begin with. This may have been an old VHS master or even (as some have speculated) a capture of the original laserdisc. There is no way to know for sure what the source was, but I highly doubt that it came directly from Gainax. Anyway, this original source was then captured as 30fps interlaced. So now we've got another level of interlacing added to the already-interlaced source. Since an IVTC only removes the original interlace, we're left with artifacts of Manga's capture.
Okay, so now we know that WOH was captured from a less-than-optimal source. Why then does it look so much worse than Slayers or Eva? Well, although Manga may have blown it with their source materials the house that authored this disc blew it in an even bigger way.
Problems with the master:
As mentioned before, there are two ways of encoding video onto a DVD. The first method is standard 30fps interlaced. This is what most OAVs and TVs shows are captured and played back at. This is a requirement of NTSC televisions. The second method is 24fps progressive encoding. Most (99.9%) of all movies (by movies, I mean widescreen movies shot on 35mm film and designed to be shown in theatres) are encoded as 24fps progressive with some special flags set on the stream so that it plays back at the required 30fps on our TVs.
So which is WOH? Well, it's both. Or neither depending on your point of view. An analysis of the MPEG2 stream shows that the movie is encoded as 30fps progressive. Here is a shot of the original DVD being analyzed by DVD2AVI. The key points to notice are the frame rate (29.970fps) and the frame type (Progressive). Also, make note of the very obvious interlacing in the background -- this on a frame that is marked "progressive" -- we should never see this kind of an artifact in a progressive frame!
So what does this mean and why should you care?
Well, there's no such thing as 29.970fps progressive (note: for the purposes of this discussion, 30fps and 29.97fps can be considered the same thing). To understand why this disc plays back so horribly, you need to understand a bit bit about how NTSC displays work... An interlaced 30 frames-per-second video stream isn't played back the way one might think. Your DVD player doesn't just display one frame after another at a rate of 30 per second. Instead, it breaks each frame up into 2 fields. A field is made up of either the odd or even lines of a particular frame. 30fps x 2 = 60 fields per second (or 60hz which is the standard for NTSC video). Now when you have a properly encoded 30fps interlaced stream, the dvd player is told to first display all the odd lines of frame 1, then all the even lines of frame 1. Then it goes on to frame 2 and repeats the process. Actually, sometimes it starts with all the even lines first. The process for determining which lines get processed in which order is called "field order" and is encoded as part of the MPEG-2 stream as well.
Below, I have taken our 5 sample frames and converted them into 10 fields and performed the resizing/downsampling from 16:9 to 4:3. This is the way that they should be played back on your TV.
Frame 1 - Field 1
Frame 1 - Field 2
Frame 2 - Field 3
Frame 2 - Field 4
Frame 3 - Field 5
Frame 3 - Field 6
Frame 4 - Field 7
Frame 5 - Field 9
Frame 5 - Field 10
Note that only one of the 10 new frames shows any ghosting. Actually, there shouldn't be any ghosting at this point, but that is an artifact of the original crappy transfer. Still, most of the interlacing problems are gone. My real point behind showing these 10 frames is that no interlacing artifacts are visible at all. If WOH had been mastered properly as Interlaced rather than Progressive, this is what we would have seen on our TVs and there would have a been a lot less bitching from all of us. >:>
Just to be clear on this point, let me say this again in a different way. The above clip is already on your DVD. If the MPEG2 had been authored with the Interlaced flag set (as mandated by the spec) the DVD would play back properly with 90% of all the interlacing issues fixed. The majority of the quality issues that we are all experiencing with this disc are NOT due to a crappy transfer, but a botched DVD mastering process.
Still with me? I hope so, because it gets a bit more complicated now. If you've ever looked at the 30fps interlaced output of a DVD in a program like FlasK MPEG or MPEG2AVi or DVD2AVI, you'll notice that most of the frames have interlaced artifacts in them. Namely, you see half of two different pictures in a single frame. Why then don't you see this on your TV? Because of the playback method described above. The odd and even lines are extracted separately and played separately. So you never see the entire interlaced picture. If WOH had been encoded this way, there probably would have been WAY less complaints. The problem is that that the whole 30fps stream was encoded as progressive.
Again, there's no such playback standard. So instead of weaving the interlaced stream together properly and playing it back at 60 fields per second, your DVD player happily displays all 30 frames in their ORIGINAL INTERLACY GORINESS. For those of us not gifted with 16:9 screens, our DVD players resize the image to 4:3 for us and instead of resizing 60 solid fields, it resizes 30 progressive (but really interlaced) frames. When you reduce the vertical size from 480 to 360 you lose the sharp interlaced edges and end up with ghosts all over the place. Also, because the stream is progressive there is no field order encoded into it, so some sequences are overly-jerky as your DVD player hasn't been told in which order to play things.
Many people have commented on the fact that Apex DVD players don't handle downsampling all that well. The reason for this is that the Apex uses a much slower video chip for its processing and therefore uses a method of resizing that requires less muscle. The Apex method works well enough for progressive sources but looks godawful on interlaced ones. Here are our 5 favorite frames resized using the Nearest Neighbor algorithm (the downsampling method that best approximates what the Apex 600a does):
Ugh. I'm glad that I don't own one of these players. >:>
Most other players use a bilinear (yes, bilinear, not bicubic) resize. You still get some interlacing artifacts after resizing, but nowhere near as severe:
Note how the interlaced lines are blended together to give us that oh-so-enjoyable ghosting that we've been seeing on this disc. This is why it's happening.
This DVD is broke. I can live with the fact that Manga had a bad master, but the fact that this DVD was authored out-of-spec really bothers me. This is a VERY amateur mistake and can be relatively easily corrected. Let me say this a little more clearly -- "THIS DISC IS DEFECTIVE." It is just as defective as the missing subs in BGC or the out-of-phase audio problem with the Slayers box set. Sure, there are some setups out there that will mask the problem (ie: HDTVs won't introduce the ghosting issues because the image doesn't have to be resized) but the simple fact remains that this disc should probably be recalled, remastered and reissued.
I hope that this page clears up some of the confusion about this DVD. If you have any further input, please post to the discussion General discussion forum on Anime on DVD and I'll do my best to address it.
- Inwards (2000/12/13)