Frame correction for LEANING

[EuropeanMan]

DEi was a manufacturer of many Bollywood DVD9s in the 80s/90s timeframe. The problem with them is that MANY of their prints were messed up.

I wish I had a DVD9 to take a screenshot from. (Currently away from home) Basically the problem is this:

If I were to take a picture of someone. You'd see them standing up straight. With DEi, that same person would lean one way about 5 to 10 degrees. So this LEANING was consistent ALL the way through the movie. EROS was primarily another manufacturer who hired DEi to make prints of their catalogue. Some of those same films released THROUGH EROS had those same people standing up straight. HOW did DEi mess this up? Why do their prints LEAN while EROS doesn't? On top of it, DEi was known to heavily crop their films from all 4 sides. VERY rarely did they release a movie where it wasn't cropped and had the original A/R intact.

Is there ANY way to fix this, to get the frames BACK to lean the OPPOSITE way? I'd rather reAUTHOR my DEi titles because they are PROGRESSIVE, while EROS was always interlaced & badly on top of that...

Thanks in advance...

[Ghitulescu]

Probably a picture from one of their movies would explain better what you mean by "lean"

[jmac698]

http://avisynth.org/mediawiki/SkewRows
Or even try reform
http://avisynth.org/vcmohan/Reform/Reform.html

[Bloax]

Even though it's obvious now, I'll go ahead anyway.
Probably a picture from one of their movies would explain better what you mean by "lean"
I think he means something like this (http://img545.imageshack.us/img545/1883/screenshotdoom201010052.png). Except way less extreme.

[IanB]

@Bloax,

That is a classic example of row skewing. The right and left sides of the image being stitched 1 pixel further across as you scan from bottom to top is the key hint. SkewRows(1) is designed to fix exactly this. The true width should have been 1367 not 1366. Unfortunately the last 768 pixels from the end of the top row will be missing.
____________________

I suspect EuropeanMan's problem is going to be skewing due to a fast shutter speed being used with a slow vertical blind style shutter, i.e. shutter is made from 2 blinds, 1st blind starts a down sweep exposing the film maybe taking 1/30th second to complete, the second blind follows close behind the first, maybe 1/250th second later, ending the exposure. So although the exposure time is 1/250th second the bottom of the negative is exposed 1/30th second after the top so things that move horizontally during this time become skewed in the final image. This shutter design was common for 35mm SLR type of still cameras, but I am not aware of this design being used in movie cameras. I guess when EuropeanMan posts a still we can stop guessing.

SkewRows can quickly adjust for only integer pixel skewing. Reform will more likely be required for the fractional shift needed. The hard bit is going to be the motion tracking, still image elements won't be skewed.

[Ghitulescu]

Why don't we all wait for a sample instead of assuming things?

[EuropeanMan]

Arghhhh I had totally forgotten about my thread...so so sorry...

here's an example from one particular DEi title: Khal Nayak (1993)

http://i167.photobucket.com/albums/u148/FarooqBhai/1-88.png

[EuropeanMan]

http://www.zulm.net/sengh_15/China%20Gate/CG%20(7).jpg

^ this is a PRIME example...

[EuropeanMan]

I also realise that IF there was a way to fix this, I'd have to crop on both sides just to get the main part of the frame "perpendicular" so to speak. Each side's corner/triangle would then be gone...

[manono]

If you just want to rotate it a little bit clockwise, then Fizick's Rotate filter is what you want:

http://avisynth.org/warpenterprises/

Rotate(1.5) rotates it one and a half degrees clockwise. And as you said, you'll have to crop and resize.

[jmac698]

It's not a rotation, you need reform.

[EuropeanMan]

^ could you please elaborate on REFORM? thanks

[manono]

^ could you please elaborate on REFORM? thanks

He already linked to it earlier:

http://avisynth.org/vcmohan/Reform/Reform.html

[poisondeathray]

Another option is vdub's perspective filter , or use batch image processing as stills like photoshop gimp's shear filter . You can fix it quite easily in after effects as well with shear and bezier warp tools

There might be something similar in avisynth, use search terms like warped frame, perspective, shear, and you might have better search results than "leaning" :)

[Asmodian]

Reform looks very impressive, it should work if it is the entire frame. From the screen shots I think the entire frame might be distorted but I cannot be sure. If it is only moving objects then it could be much more complicated.

[jmac698]

It does look suspicious. The background doesn't seem to have the problem, in which case it could be like shutter roll. In that case it's related to the movement, so it's not easy to fix.

[StainlessS]

Some of those same films released THROUGH EROS had those same people standing up straight.


Might be advantageous if we could see similar frames of those invididuals that lean one way or another, compared to the fine upstanding folk.

Edit: The above would seem to be post production and rule out anything related to movement during capture.

[poisondeathray]

It does look suspicious. The background doesn't seem to have the problem, in which case it could be like shutter roll. In that case it's related to the movement, so it's not easy to fix.

The background does have problem. If you look at the palm trees behind the girl's right in the 1st picture and the ship anchor in the 2nd picture, both are skewed . I would use the ship anchor as a vertical reference object to determine how much to correct

Also, he said it's consistent all the way through the movie in the 1st post , which implies a transfer issue

Pretty easy to fix with the vdub perspective plugin, or at least improve dramatically if it's constant. Otherwise you could keyframe the correction in something like after effects .

[StainlessS]

the ship anchor in the 2nd picture, both are skewed . I would use the ship anchor as a vertical reference object to determine how much to correct


You cannot assume anything from the angle of the ships anchor,
a little strain on an anchor chain is not unusual.

EDIT: One thing to note is the disparity in angle between the two samson posts (masts) on the ship.

[poisondeathray]

You cannot assume anything from the angle of the ships anchor,
a little strain on an anchor chain is not unusual.


True, but correlate that observation with the angle of metal pillar beside the guy with the dark blue shirt (he's cut off in the background). Unless the ground is uneven there.... To me, the angle is very similar . IMO, this rough angle correction isn't worse that the original . I'm sure he can tweak it better, & crop & resize if he spends more time

http://i52.tinypic.com/zx19on.png



The vdub plugin only allows for 4 point correction. It's not a mesh filter that allows for disproporational distortions. AE has several warp filters that can correct for more complex distortions and affect different parts of the frame, but I think you can get away with the vdub plugin only in this case, but he would have to correleate with other objects and other scenes to fine tune the correction

[StainlessS]

True, the angles of the chain and 'lump of metal' are quite similar, but I would not want to bet my life on it.
The edit in previous post might bear consideration (about the masts).

EDIT: The 'lump of metal' could well be supported upon a couple of spars of wood to allow chains/slings to be
slung around, my experience of (years in) ship yards would not allow me to assume that the chunk of metal
would have been laid down with the aid of a spirit level.

[poisondeathray]

I'm not sure how the masts are supposed to look like . Analyzing other objects and scenes would help

[jmac698]

actually video would help, then we could see the motion parts.

[StainlessS]

I would expect the masts to be a little more parallel rather than splayed (fish eye?). A pic of a plumb line would be ideal,
but we may have to make do with alternative renderings of the two clips mentioned by EuropeanMan.

[wonkey_monkey]

Wouldn't the credits be the best thing to look at to determine the correct geometry?

David

[jmac698]

good point.
bunwarpj plugin in Fiji could do any correction, and it's free.
reform is like perspective.
and the speculation is getting a little silly :)

[poisondeathray]

Wouldn't the credits be the best thing to look at to determine the correct geometry?

David

aren't credits usually done separately to the main transfer ?

[jmac698]

I was thinking that too, I don't know why?
I know sometimes they have to be done separately because the writing is so tiny on film that you can't read it in video, especially for widescreen. It could be different for translation purposes, but otherwise I dunno why to think that.

[poisondeathray]

They might do things differently in Bollywood. You see folks complain all the time about this or that crappy DVD :)

[Gavino]

The background does have problem. If you look at the palm trees behind the girl's right in the 1st picture and the ship anchor in the 2nd picture, both are skewed .
You cannot assume anything from the angle of the ships anchor, a little strain on an anchor chain is not unusual.
Just to stir the pot a bit more - I see palm trees every day, and not many grow exactly vertical...
:D

[EuropeanMan]

Give me some time today folks...been busy with work & other things recently so haven't had a chance to catch up with all the replies FOR WHICH I THANK YOU ALL GREATLY.

I'll pull out some of my DEi titles and upload some samples and see what I can come up with...thanks for being patient.

[jmac698]

Since we're too lazy to do it, here ya go:
http://screenshotcomparison.com/comparison/90114/picture:1

dir="D:\project001a\leaning\"
fn1="1-88.png"
fn2="CG (7).jpg"
src1=imagesource(dir+fn1)
src2=imagesource(dir+fn2)
bg=blankclip(src2)
src2
deskew(last, bg,ltopx=[<"ltopx", 4, 848, 75>],ltopy=[<"ltopy", 4, 848, 62>],lbotx=[<"lbotx", 4, 717, 130>],lboty=[<"lboty", 4, 848, 484>], \
rtopx=[<"rtopx", 4, 848, 673>],rtopy=[<"rtopy", 4, 848, 70>],rbotx=[<"rbotx", 4, 848, 716>],rboty=[<"rboty", 4, 848, 458>])

[EuropeanMan]

^ Thanks for the efforts. Too bad there are so many jaggies :(

[Didée]

Just to have mentioned it:

ffdshow -> "perspective correction"

[jmac698]

Don't worry about jaggies; I didn't enable high quality mode. It's only a quick test; to see if the full image has constant deform or else it's motion related. I still can't tell.