Chroma bleed on a IVTC process. 'The Thief and the Cobbler' Restoration Project

[Gerry62]

Hi,

Working on a project from multiple sources that all have different issues.

Ok before I go into details here is the clip
http://www.2shared.com/video/1CRxZUTS/Thief_JAP_DVD.html

Its from a Japanese DVD and ive used womble to cut out a section so there is no re-compression.

Now I'm no expert when it comes to getting animation back to 23.9FPS, in fact i don't know much at all.

Now here is where it might get tricky for those who are saying 'Ahah, I can fix that'!
I want the resulting file to be combined from both fields to preserve detail, so no deinterlacing (if possible), the other thing I want to do is to stretch the image so it fills a 4x3 image as im working with a frame by frame restoration process to clean each frame, so having both of those fields is kinda important if you get my drift! But if not possible then so be it

So what did i try?
AnimeIVTC ver 2
Srestore
and the best I could get it to do is this

AnimeIVTC(mode=4,aa=0,omode=1) was the simple script i used, noting fancy as I don;t think it has a AA issue at all. I diddnt correct any Luma/Chroma or sharpen it in any way, as I want to get this part sorted first

http://i281.photobucket.com/albums/kk219/ade42_2008/Poo444440015.jpg

as you can see there is the ghost of a chroma field still present on all the frame (look at faces)?. This of course happens to a degree on all versions, more noticeable here.
Is there a way to eliminate this without having to deinterlace or cut frames away? I don;t want to clean it away with any RemoveGrain option as that will adversely effect the animation itself, am going to clean it by hand the old fashioned way! and why not, it deserves it, But of course if anyone thinks they have a 'perfect' script for dealing with any of the various elements below Id be very greatfull too see it!

for comparison here is the same scene on 2 other copies..

The Workprint
http://www.2shared.com/video/mcOK69qE/Thief_Workprint_DV.html

The Japanese LaserDisc
http://www.2shared.com/video/fznrrhm0/Theif_Jap_LD.html

[Gerry62]

Here's Tygerbug to tell you more about the project and what we are trying to fix,

Hi everyone. I'm the editor and supervising director of this restoration project. I brought Gerry in based on his sterling work on archive British television. He's used to PAL and is getting up to speed figuring out NTSC with its 3:2 pulldowns.

The sources we're using vary wildly in quality, from high-def 35mm film [not enough of it, sadly], to DVD to laserdisc to PAL VHS to nth generation NTSC VHS from hell.

I restored this film originally in 2006, and I had to deal with the interlaced chroma smudging issue Gerry's looking at now.

I was converting the M2V material to PhotoJPEG for editing, using MPEG Streamclip. If I didn't check the Reinterlace Chroma option, I got some nice clean frames:

http://orangecow.org/thief/thiefedited1-6.jpeg
http://orangecow.org/thief/thiefedited1-4.jpeg

But I also got that smearing of the chroma on all the previously-interlaced frames.

If I did click the Reinterlace Chroma option, there was no smearing but the chroma wasn't as high resolution and was clearly a little bit blocky:

http://orangecow.org/thief/thiefedited1-3.jpeg
http://orangecow.org/thief/thiefedited1-5.jpeg

I did both, and edited the best frames from each version together in Final Cut Pro, just copy-pasting my way through the entire film. There was a cut on nearly every frame as I merged the two versions together.

This time round I'm hoping for a better result, maybe through some automatic AviSynth script that'll save Gerry the trouble of what I did back in 2006.

I'd also like the image to be cropped and resized to fill the entire 4x3 frame, so that Gerry has more resolution to deal with when sharpening it and hand-removing dirt and everything.


If you can, please download the three clips Gerry has just posted; they're from different versions of the film and all have their own challenges.

The JAPANESE DVD we've been looking at is a bit soft, and high-contrast - the whites are blown out entirely in many scenes, losing a lot of highlight detail.

We also have an NTSC LASERDISC [US, not Japanese] which is terrible quality but retains that luminance detail. I'm hoping that both sources can be improved enough that they could be composited together for a better picture overall.

The laserdisc was transferred to DVD with a 3:2 pulldown. It's sort of a smeary mess of fields, chromawise. It's not a good quality bit of video to begin with, and has all sort of NTSC dot crawl and analog artifacts. It's also too blue. We'll need help sorting out the 3:2 pulldown without all sorts of nasty interlacing smearing.

But if you think that's low quality, try this on for size. Here's a grab-bag of really poor quality shots from a rare and stupid version of the film we only have as an nth-generation VHS. The CALVERT WIP.
http://depositfiles.com/files/icce3sqpq

Once, this was probably a 3:2 pulldown. Now the fields are merged together a bit and it's just a low quality mess. I'd like to see what we can do with this.


There's also the WORKPRINT. This is an okay quality, one of a kind PAL VHS. It's a 25fps treatment of a 24 fps film, and has an unusual pulldown. A field is added every half second. So every half second it switches from progressive to interlaced or back again. Now technically we could create two versions of the film and edit them together every half second, but a script would be much appreciated for Gerry's sake, so he can have a nice progressive version of the whole film without that nonsense, and get to work restoring and cleaning things up.

I actually have a few more clips up from the workprint, so here they are:

The Thief at the polo game, workprint clip.
http://depositfiles.com/files/gi27lnx5a

The Thief throws everything away and gets a new coat, workprint clip:
http://depositfiles.com/files/dr85e0lf6

The March of the One Eye War Machine, workprint clip.
http://depositfiles.com/files/w1syb3773

The beginning of the destruction of the war machine, workprint clip.
http://depositfiles.com/files/jb60jhy35


http://orangecow.org/thief/thiefedited1-6.jpeg[/img
orangecow.org


Posted on Behalf of Tygerbug By Gerry (Ade!)

[mp3dom]

It's a 25fps treatment of a 24 fps film, and has an unusual pulldown. A field is added every half second. So every half second it switches from progressive to interlaced or back again. Now technically we could create two versions of the film and edit them together every half second, but a script would be much appreciated for Gerry's sake, so he can have a nice progressive version of the whole film without that nonsense, and get to work restoring and cleaning things up.

About this, it's "euro" pulldown.
From Wikipedia:

This pulldown method[4] is sometimes used in order to convert 24 frames/s material to 25 frames/s. Usually, this involves a film to PAL transfer without the aforementioned 4% speedup. For film at 24 frames/s, there are 24 frames of film for every 25 frames of PAL video. In order to accommodate this mismatch in frame rate, 24 frames of film have to be distributed over 50 PAL fields. This can be accomplished by inserting a pulldown field every 12 frames, thus effectively spreading 12 frames of film over 25 fields (or “12.5 frames”) of PAL video. The method used is 2:2:2:2:2:2:2:2:2:2:2:3 (Euro) pulldown.


You can restore the original 24p easily using TFM and setting TDecimate to remove 1 repeated frame every 25 frames. If the pattern is fixed (doesn't change across scene changes) you can do it with selectevery too.

[Emulgator]

Hi, Gerry62 and Tygerbug!

Here is my Selectevery() procedure for a fixed pattern of Euro Pulldown.
Developed for Rota-Prism-Scans where almost every frame is interlaced,
e.g. in almost every case field content stems from different original frames and times,
occurring to both sides of a vertically rolling time border.
Therefore these fields can not be weave-restored,
instead have to be resized to frames and the redundant ones be discarded.

#[*#__________________Start of Manual Hard Euro-Pulldown-Removal 25i -> 24p (25 pattern offset variations)_________________
#SwapFields
#ComplementParity # ComplementParity only changes the assumed field order, not the physical field order !
#ReverseFieldDominance(shiftup=true)#(PAL Only ! YUY2 oder RGB only) ...Like CCE's shift lines by 1
#RevFieldDom: PAL DV is expected to be sampled bottom field first.
#In case of a device sampling top field first this filter can reverse field dominance
#by simply shifting each line up (or down) by one line and duplicating the bottom (respectively top) line.
#AssumeTFF()
SeparateFields()
#srctopbottomcrop=0#easy way to get rid of noisy borders, but it is better to repair borders using ExInpaint before
#Lanczos4Resize(width, height*2, 0, srctopbottomcrop, -0, -srctopbottomcrop)
#Spline36Resize(width, height*2)
#eedi2(field=2)#(eedi2 kontrolliert wohl am schärfsten und besten, aber nicht positionsecht!...)
nnedi2_rpow2(rfactor=2,cshift="Spline36resize", fwidth=width*1)
# Now choose any resizer and find one of the following pattern offsets...
#
SelectEvery(25, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24)#Decimation=0+1
#SelectEvery(25, 0, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23)#Decimation=1+2
#SelectEvery(25, 1, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24)#Decimation=2+3
#SelectEvery(25, 0, 2, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23)#Decinmation=3+4
#SelectEvery(25, 1, 3, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24)#Decimation=4+5
#SelectEvery(25, 0, 2, 4, 7, 9, 11, 13, 15, 17, 19, 21, 23)#Decimation=5+6
#SelectEvery(25, 1, 3, 5, 8, 10, 12, 14, 16, 18, 20, 22, 24)#Decimation=6+7
#SelectEvery(25, 0, 2, 4, 6, 9, 11, 13, 15, 17, 19, 21, 23)#Decimation=7+8
#SelectEvery(25, 1, 3, 5, 7, 10, 12, 14, 16, 18, 20, 22, 24)#Decimation=8+9
#SelectEvery(25, 0, 2, 4, 6, 8, 11, 13, 15, 17, 19, 21, 23)#Decimation=9+10
#SelectEvery(25, 1, 3, 5, 7, 9, 12, 14, 16, 18, 20, 22, 24)#Decimation=10+11
#SelectEvery(25, 0, 2, 4, 6, 8, 10, 13, 15, 17, 19, 21, 23)#Decimation=11+12
#SelectEvery(25, 1, 3, 5, 7, 9, 11, 14, 16, 18, 20, 22, 24)#Decimation=12+13
#SelectEvery(25, 0, 2, 4, 6, 8, 10, 12, 15, 17, 19, 21, 23)#Decimation=13+14
#SelectEvery(25, 1, 3, 5, 7, 9, 11, 13, 16, 18, 20, 22, 24)#Decimation=14+15
#SelectEvery(25, 0, 2, 4, 6, 8, 10, 12, 14, 17, 19, 21, 23)#Decimation=15+16
#SelectEvery(25, 1, 3, 5, 7, 9, 11, 13, 15, 18, 20, 22, 24)#Decimation=16+17
#SelectEvery(25, 0, 2, 4, 6, 8, 10, 12, 14, 16, 19, 21, 23)#Decimation=17+18
#SelectEvery(25, 1, 3, 5, 7, 9, 11, 13, 15, 17, 20, 22, 24)#Decimation=18+19
#SelectEvery(25, 0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 21, 23)#Decimation=19+20
#SelectEvery(25, 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 22, 24)#Decimation=20+21
#SelectEvery(25, 0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 23)#Decimation=21+22
#SelectEvery(25, 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 24)#Decimation=22+23
#SelectEvery(25, 0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22)#Decimation=23+24
#SelectEvery(25, 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23)#Decimation=24+0
#*]#_________________End of Manual Hard Euro-Pulldown-Removal 25i -> 24p (25 pattern offset variations)______________________

Unfortunately a fixed pattern rarely applies.
I often found variable distances between dominance shifts in my cases,
so for these cases I had to abandon the following solution and went on to AnimeIVTC.

Edit: As I just found out, your "Thief at the polo game" sample is not of that kind.

[Emulgator]

Next candidate:
The DVD sample is just hard pulldowned to 29.97i and can easily be restored to 23.976p.

The following script brings back full frames without any leaking from adjacent fields,
at least for the DVD sample's length:

MPEG2Source("D:\Thief JAP DVD.d2v")
SeparateFields()
SelectEvery(5, 1, 2, 3, 4)#Hard NTSC 29.97i to Film 23.976p Pulldown
Weave()


Well, some dehaloing and sharpening might follow.
But the mosquito noise from encoding (around edges)
prevented me at first attempt from applying serious sharpening.

[Emulgator]

And happy to announce, the "Thief at the polo game" sample can be weave-restored from its pure progressive or pure interlaced frames.

Fortunately the pattern seems fixed throughout the sample.

Edit: No, The last scenecut exhibits at least an offset. But still feasible.

[tygerbug]

Hey, does anyone want to help with this project? There's a lot to do, and I know Gerry has been having some personal issues taking up some of his time, so additional restoration crew would be helpful, just to speed up the work a bit.

The lower-quality material needs some Avisynth filtering and the higher quality stuff [some of which is now in HD!] needs manual dirt and dust removal. If anyone wants to help I could certainly send you something that Gerry hasn't touched yet.

This is a classic, legendary unfinished animated film from Richard Williams, who animated Who Framed Roger Rabbit. He spent 25 years on the film and intended it to be his masterpiece, but it was never quite finished due to Hollywood politics. You can watch it on Youtube at thethiefarchive ...
http://www.youtube.com/playlist?list=PL18B0CA620B61D076&feature=plcp


Peter C in England had some 35mm film transferred as HD, and posted it to Youtube ...
http://www.youtube.com/watch?v=ASziVj9xcPg
http://www.youtube.com/watch?v=9Dmd3d2WZyE

It needs dirt removal, though I'm still waiting for the actual clean Blu-Ray files from him.

Email me at gilchristgarrett [at] gmail [dot] com, in case I'm not checking this thread.

[Lyris]

Great to see you guys posting here. What happened to The Thief was very sad.

Those 35mm segments, do you only have them on BD, not as a higher-up source? No DPX files or an HDCAM SR tape or anything?

[tygerbug]

[Edit: Actually, Peter does have HDCAM versions.] I don't actually have the Blu-Ray myself yet, just what's on Youtube. But I'm sure Peter will send that to me before long.

The previous 50 minutes of 35mm footage were transferred as SD, but stretched vertically to fill the frame. That was a few years ago. There were issues with JPEG compression too, though smart filtering should do wonders with that.

[Emulgator]

I would be in for such restoration. Will send you a PM.