Need help determining the nature of a DVD (also: strange aliasing)

[Chainmax]

I am trying to rip my recently bought "Robot Chicken - Season 1" DVDs. For all of you that don't know, it's a comedic take on pop culture made in stop-motion with action figures. The problem is that there's very little motion and lots of duplicate frames, and upon examination it looks progressive. I seem to recall a similar post I made a long time ago in which IIRC I was told to look inside the d2v file and look for some numbers in it but I can't find it. So, is there a way to know if the source is progressive/telecined or interlaced that does not involve manual examination?

[Skelsgard]

DGIndex has a good detection for interlaced vs. telecined streams.
Load the VOB/MPEG-2 and use the Play function. As the streams plays it will detect the video type (more accurate the longer it plays).
http://img92.imageshack.us/img92/3748/filmvx5.jpghttp://img187.imageshack.us/img187/7837/interlacedqs7.jpg

One more thing, if it comes from USA TV (NTSC) it´s most likely to be Interlaced.
Cheers.

[Chainmax]

DGIndex does say "Interlaced", but wasn't that prone to many different interpretations? Besides, I don't think this particular case was recorded on tape as it's stop-motion-like animation.

[Skelsgard]

DGIndex does say "Interlaced", but wasn't that prone to many different interpretations?
In what way?

You can use "Force Film" and check how many interlaced frames you get. Force Film will give you a progressive output the closer the movie is to Film and will show abundance of interlaced frames the closer it is to Interlaced. In most Telecined movies, Force Film gives you clean progressive output.
For example, in my Final Fantasy VII DVD, the main movie is Film Telecined, while the credits are Interlaced.

Cheers.

[Blue_MiSfit]

Try MeGUI's analysis on a d2v that you created using 'honor pulldown flags' mode.

If you don't use MeGUI, you can do it manually. Make an AviSynth script that's just MPEG2Source, and then have a look in virtual dub. Get to a scene with some motion, and you should see some combing. Okay, now step through it one frame at a time. Do you see a 3/2/3/2/3/2 pattern of interlaced/progressive/interlaced/progressive frames? If so, then your video is telecined and should be treated as such (force film /ivtc). If it's all interlaced, then it's interlaced and should be treated as such (deinterlace)

~MiSfit

[Chainmax]

Skelsgard: IIRC, according to the IVTC tutorial, Interlaced could mean anything from actual interlacing to hard pulldown [3:2 pulled down field pattern but encoded as interlaced (pulldown was applied prior to MPEG encoding)].

Blue_MiSfit: I might try MeGUI's analyzer, thanks for the advice. As for doing it manually, it's not an option. Like I already said, the show has very little motion, which couple to the frequent duplicate frames makes it very hard to spot any combing at all.


In any case, I have found interlacing, but most of it seems to occur on red-colored things. Maybe it was shot on a DV camera and/or has the chroma upsampling bug. I am uploading three short samples and will request this to be moved to the Avisynth forum.


[edit]Here are the clips:

http://rapidshare.com/files/7792275/Sample_1.demuxed.m2v.html
http://rapidshare.com/files/7793111/Sample_2.demuxed.m2v.html
http://rapidshare.com/files/7793392/Sample_3.demuxed.m2v.html

[Terranigma]

Have you tried Didée's method from your anti-aliasing post for those clips?

I was only able to download the first sample because rapidshare had a waiting period. But it seemed to have worked very well from what I can tell.

I'm not of much help when it comes to being able to tell if a source is either interlaced or progressive/telecined :p

[Chainmax]

It could work, but that script was meant to improve the NASTY deinterlacing job that had been done to the source. In this case, I need to know wether this is interlaced, progressive or just has some form of the chroma upsampling bug (first time it happens here though).


P.S: I am not able to tell if it's TFF or BFF either, but SecureBob().SelectEven() reduces the artifacting a great deal, though not completely.

[setarip_old]

@ChainmaxI am trying to rip my recently bought "Robot Chicken - Season 1" DVDs.If all you are doing is ripping a copy of the DVD, what difference does it make if it's interlaced or not?

BTW - Just curious to know why someone with 2,400 posts is posting a question in the "Newbies" thread?

[Chainmax]

What I meant was that I mean to encode the episodes to AVC. As to why this thread is here, I consider the OP's question to be of newbie content. In post 6, however, I changed the content of the thread which is why I asked for a relocation to Avisynth Usage.


P.S: not being able to sleep sucks :(.

[neuron2]

In post 6, however, I changed the content of the thread which is why I asked for a relocation to Avisynth Usage. But you didn't change the noob title!

I'm wondering why a guy with over 2400+ posts doesn't know the simple process for determining whether content is interlaced or progressive. :)

[Chainmax]

I didn't change the title because I still want the OP's question to be answered. As for your other question: yeah, I don't know. Why don't you take a look at the clips (they are small, less than 7MB each) and see why this particular source is giving me trouble?

[neuron2]

I din't change the title because I still want the OP's question to be answered. As for your other question: yeah, I don't know. Why don't you take a look at the clips (they are small, less than 7MB each) and see why this particular source is giving me trouble?
I looked at the first one. It's progressive but has bad aliasing, not interlacing. Probably it is because someone deinterlaced that segment.

So, is there a way to know if the source is progressive/telecined or interlaced that does not involve manual examination?
Manual examination of the fields is the only fully reliable method. It takes about 2 minutes. Why are you averse to that?

[Chainmax]

I looked at the first one. It's progressive but has bad aliasing, not interlacing. Probably it is because someone deinterlaced that segment.

That doesn't seem logical, deinterlacing would imply a waste of money on the distributor's end. Maybe it was always progressive and this sparse artifacting was introduced during the disc's creation?


Manual examination of the fields is the only fully reliable method. It takes about 2 minutes. Why are you averse to that?

I already did that:

The problem is that there's very little motion and lots of duplicate frames, and upon examination it looks progressive.
I have found interlacing, but most of it seems to occur on red-colored things.


It's not me being lazy, I simply can't determine wether the source is interlaced, pure progressive or telecined. Field order examination makes it look like TFF and BFF are equal so that would imply progressive content, but again, the low amount of motion and the duplicate frames make me not be sure if that's the case.

[jel]

moved to avisynth usage as per user request.

[Mug Funky]

That doesn't seem logical, deinterlacing would imply a waste of money on the distributor's end. Maybe it was always progressive and this sparse artifacting was introduced during the disc's creation?

what they waste on machine time (negligible considering it's a small link in no doubt a bigger chain), they save by employing junior editors at minimum wage who've not yet learnt to watch their stuff on the big shiny monitor TV next to them, and not just on the Mac Studio panel.

as for being able to objectively tell if something's interlaced or not, i'm working on it :)

so far my best bet (which is unfortunately unacceptibly slow) involves both a spatial and temporal check.

a (reasonably) reliable spatial check involves applying a comb mask to both field-separated and weaved versions of the source, then subtracting the results. the difference is (almost) always negative when the frames are progressive, so if there's an amount of positive difference over a threshold, the frame is declared interlaced.

a temporal technique i'm messing with (that still has problems) is simply taking the difference between frame 0, bottom field and frame 1 top field (adjacent fields on different frames), and comparing that to the difference between frame 0, top field and frame 1 bottom field. the difference in interlaced (natural) video should ideally be twice as much compared to progressive which should return the same numbers for both comparisons.

both techniques can be fooled quite easily... and combining them makes it twice as slow :(

unfortunately i think the answer lies in motion-estimation which will make things kinda slow. when watching interlaced video, human eyes can see "smooth" motion way below the thresholds of any metrics i've been able to come up with. i'm sure someone else around here could handle the challenge.

the results of falsely detecting progressive (or film) frames can be disastrous on an interlaced monitor when they look perfectly fine on a computer screen.

[Chainmax]

So it's not an easy task then. Could you please download the three samples and tell me what you make of them?

[Mug Funky]

well, as implied by my post, human eyes will do a better job than anything i've scripted.

so probably just watch it through and suss it out that way.

my guess is if some scenes are randomly deinterlaced, and some scenes are natively progressive, then chances are some other scenes will be interlaced - it'll have been put together from lots of pieces, and potentially they'll all have been treated differently.

i'm at work now, so haven't the time to check it all out myself. besides which, you can do it just as well.

[neuron2]

Field order examination makes it look like TFF and BFF are equal so that would imply progressive content, but again, the low amount of motion and the duplicate frames make me not be sure if that's the case. You have to examine more than the field order, which is irrelevant for progressive video anyway.

The simple test is whether you can find frames that have different pictures (that is, from different temporal moments) in the top and bottom fields. If so, then the frame is interlaced; if not, it is progressive. A exception to watch out for is one-field-phase-shifted progressive. I've described the process here so many times that I'm too bored to do it again. But I did find this link:

http://forum.doom9.org/showthread.php?p=878536#post878536

[Chainmax]

Ok, so the pattern is like this:

[aa][aa][aa][aa][bb][bb][cc][cc][cc][cc][dd][dd], and so on
so that means it's indeed progressive. I noticed that in the 4-frame runs the first and third repeat are almost the same but with very slight differences (one might some some very little extra noise or so) and that the second and fourth ones are slightly shifted upwards. Could that be because I only used SeparateFields()? If so, could it be field shifted progressive then?

[neuron2]

When you separate fields, you sill see slight bobbing. You can ignore it. If it bothers you, use Bob() instead.

It is not phase-shifted as it doesn't have the signature that I described in the linked post.

[Chainmax]

I figured that out when fieldmatching didn't eliminate the aliasing. I didn't catch the part where you explained the signature common to phase shifts, sorry about that :o.
Could the aliasing be solved by any other way than blurring the heck out of the source with AAA() and/or EEDI2()?

[neuron2]

Could the aliasing be solved by any other way than blurring the heck out of the source with AAA() and/or EEDI2()? You can try to develop a method whereby EEDI2 is applied only to the aliased portions of the video. It would be tricky to detect them but doable. Please don't ask me to do it. :)

[Chainmax]

Ok, this is the current script:

MPEG2Source("C:\rrr\RCS1E1.d2v",info=3)

ColorMatrix(hints=true)

a = last
b = a.DeGrainMedian().VagueDenoiser(threshold=0.8, method=1, nsteps=6, chromaT=0.8)
SeeSaw(a,b, NRlimit=6, NRlimit2=7, Sstr=1.5, Slimit=5, Spower=5, SdampLo=6, Szp=16)

AAA()

EEDI2().TurnRight().EEDI2().TurnLeft()

BicubicResize(720,480,0,0.5)

AAA()

dull=last
sharp=dull.LimitedSharpenFaster(SMode=4)
Soothe(sharp,dull,25)

Tweak(sat=1.2)


And here are some comparison shots:

http://img452.imageshack.us/img452/6946/before1ve5.png (http://imageshack.us)
http://img338.imageshack.us/img338/8190/after1ip6.png (http://imageshack.us)

http://img134.imageshack.us/img134/7865/before2qg4.png (http://imageshack.us)
http://img452.imageshack.us/img452/9162/after2lz1.png (http://imageshack.us)

See how the dots on Seacrest's vest become lines? That caused by AAA. By the way, there's a scene where there definitely seems to be some interlacing, I can upload a small sample for you guys to see.

[Chainmax]

Wow, the source is really ****** in some spots. I'd have to do frame-by-frame examination with manual replacement of the trouble frames. What a shame that such a great show was handled with such sloppyness :(.