Stabilizing footage from 35mm scan

[PointFloat]

The footage is scanned to a 2048x2048 tiff sequence and there is a lot of wobble from the scanner. The whole image shakes vertically and so far has been a pain to stabilize. I have tried the AE warp stabilizer and PFclean, yet with no great results. The problem is that most stabilizers are suited for stabilizing a single scene, yet my footage is a whole reel from a movie, 16minutes long. The wobble is consistent bad results when used across multiple scenes, as it tries to compute average stabilization data from the whole 16 minutes.
PFclean gives ok results across 3-4 different scenes, but fails when the whole 16minutes is processed. Here is an example of the footage: http://speedy.sh/wJYwm/h264Test.mp4'' . How could I get the whole movie stabilized?

[hech54]

Uh oh. My Spammy senses are tingling.

[PointFloat]

Uh oh. My Spammy senses are tingling.

And what do you mean by that? The link?

[lordsmurf]

Attach to the forum or use Dropbox. I don't download from crap download sites.

Have you tried VirtualDub deshake?

[sanlyn]

The footage is scanned to a 2048x2048 tiff sequence and there is a lot of wobble from the scanner. The whole image shakes vertically and so far has been a pain to stabilize. I have tried the AE warp stabilizer and PFclean, yet with no great results. The problem is that most stabilizers are suited for stabilizing a single scene, yet my footage is a whole reel from a movie, 16minutes long. The wobble is consistent bad results when used across multiple scenes, as it tries to compute average stabilization data from the whole 16 minutes.
PFclean gives ok results across 3-4 different scenes, but fails when the whole 16minutes is processed. Here is an example of the footage: http://speedy.sh/wJYwm/h264Test.mp4'' . How could I get the whole movie stabilized?

As this sample has been processed in another app that re-encoded the results rather than attempted in lossless format, it's difficult to say how the original looked. In any case, no 35mm movies are shot at 16:9. It would be somewhere near 1.85:1, which is wider than 16x9. So cropped and resized to 1280x720 for a 16x9 image at 24fps (24fps is not valid for BD/AVCHD), you would have about 10 or 12 pixels at the top and bottom of the image to fit the entire image in a 1.777778:1 window without distorting the image. A bitrate of 2300 Kbps is absurdly low for a quality HD image.

Besides wobble, there's geometric distortion and wrinkling within the image itself, and flicker.

[themaster1]

if you can't stabilize it with pfclean, i don't know which software will...

Have you tried with the following settings for the stabilize effect:
mode: vertical
frame window: 3
lock motion:no
sub pixel: yes
sensitivity: 100
add area: pick the whole frame

[sanlyn]

Here's an effort, certainly not perfect. Something has happened to levels and color. I didn't have a lot of time for that. Repaired a couple of "wrinkle" frames and calmed things down a bit. Could certainly use more work. Had to use 2 steps, as that big image gave my poor 32-bit OS some memory problems. Script attached -- which I'm certain could use some work in itself. There's a slight letterbox to maintain a 1.81:1 image. Still a olittle judder, but I wouldn't try feeding this into AFter Effects without the original source and working with lossless media -- and you'd need Avisynth anyway.


[Attachment 20593 - Click to enlarge]

[lordsmurf]

If the scanner sucks, and the film is yours, just get it redone somewhere reputable.
Fixing it in post is not desirable.

[smrpix]

How could I get the whole movie stabilized?

Normally one goes scene by scene.

edit: Looking at Sanlyn's sample (I won't download from a site that installs an exe either) looks like you got lucky with those rounded borders. They can be used as stabilization points and may allow you to fix bigger chunks.

[PointFloat]

Thank you for the replies!

if you can't stabilize it with pfclean, i don't know which software will...

I have not tried the settings you have specified, but I'll give it a go. For the area, I have mostly selected the top/bottom edge of the frame. However, wouldn't selecting the whole frame across 16 minutes of footage mean that it will average the stabilization across scenes?

Here's an effort, certainly not perfect. Something has happened to levels and color. I didn't have a lot of time for that. Repaired a couple of "wrinkle" frames and calmed things down a bit. Could certainly use more work. Had to use 2 steps, as that big image gave my poor 32-bit OS some memory problems. Script attached -- which I'm certain could use some work in itself. There's a slight letterbox to maintain a 1.81:1 image. Still a olittle judder, but I wouldn't try feeding this into AFter Effects without the original source and working with lossless media -- and you'd need Avisynth anyway.


[Attachment 20593 - Click to enlarge]

Thus far I have only worked with a tiff sequence. The problem with avisynth is the final output format, which in this case has to be a .mov. But I'll give deshake a try.

[sanlyn]

The problem with avisynth is the final output format, which in this case has to be a .mov. But I'll give deshake a try.

Avisynth outputs lossless, decoded, uncompressed AVI. You can encode it to whatever you want. I have AE myself, but I'd never use it for crop/resize/encode. Avisynth is better at that and at denoising to begin with. AE could offer a little better motion control after Avisynth cleans it up. You might notice in some frames that even though the borders ren't moving, you still see some shimmer and distortion. AE won't fix that. Won't fix the spots or clumpy grain, either. The grain in the uploaded sample isn't what film grain looks like.

[2Bdecided]

Nice job from Sanlyn - but I have a suspicion that if you tell VirtualDUB's deshaker to only concentrate on the borders (i.e. the transition region from frame to black), you might get an excellent result. You need the "Ignore pixels inside..." option.

DO NOT crop down to the original borders before processing. Crop most of the sea of black top and bottom, but keep a bit of it. It's the transition from black to frame that's most important.

Cheers,
David.

[sanlyn]

I last used DeShaker years ago. I see where it's improved since then. Will give it a run.

I left about 12 pixels top and bottom. You'd have to do that anyway in a 16:9 frame to keep the original aspect ratio of the image itself.

[budwzr]

Uh oh. My Spammy senses are tingling.

And what do you mean by that? The link?

Yeah. Why don't you upload it to the forum so more people can get involved. Hint: "My Files".

[sanlyn]

I had no problem with the link, except for having to look around and find out which link to download.

[poisondeathray]

Or you can use various download mangers e.g. jdownloader will allow you do download it without downloading some suspicious exe

For your convenience

[budwzr]

I had no problem with the link, except for having to look around and find out which link to download.

I'm scared.

[budwzr]

Or you can use various download mangers e.g. jdownloader will allow you do download it without downloading some suspicious exe

For your convenience

That was very considerate and helpful. Thanks a bunch.

[budwzr]

Needs color work too.

[2Bdecided]

Or you can use various download mangers e.g. jdownloader will allow you do download it without downloading some suspicious exe

OT: There was a clear direct link right at the top of the page that pulled down the .mp4 and nothing else - what are you guys messing about at?! No need to read further or click on the .exe download at the bottom.

[sanlyn]

Needs color work too.

I agree, it needs some color work, but where did you get that washed-out look?

[budwzr]

Needs color work too.

I agree, it needs some color work, but where did you get that washed-out look?

From the washed out sample. Like you get an "old photo" look when you color correct an "old photo". Or a blown highlights look when you color correct something with blown highlights.

The source IS what it IS.

Note: I used the Trench Coat to determine correct WB. And toned down the green a smidge. The eyes are overly sensitive to the green channel. That's why it looks "funky cold medina" in the orig. No other tools used. No contrast/bright/etc.

[poisondeathray]

Or you can use various download mangers e.g. jdownloader will allow you do download it without downloading some suspicious exe

OT: There was a clear direct link right at the top of the page that pulled down the .mp4 and nothing else - what are you guys messing about at?! No need to read further or click on the .exe download at the bottom.

It's not there in firefox , or maybe some security settings prevent it from being displayed . Or maybe I'm just blind

[PointFloat]

Nice job from Sanlyn - but I have a suspicion that if you tell VirtualDUB's deshaker to only concentrate on the borders (i.e. the transition region from frame to black), you might get an excellent result. You need the "Ignore pixels inside..." option.

DO NOT crop down to the original borders before processing. Crop most of the sea of black top and bottom, but keep a bit of it. It's the transition from black to frame that's most important.

Cheers,
David.

Indeed, concentrating on the borders and not cropping it fully is the guideline I have tried to follow with every program thus far, when possible. It's just that the footage at times is very dark and the transition from frame/image content to black can become ambiguous for the program.

[sanlyn]

Needs color work too.

I agree, it needs some color work, but where did you get that washed-out look?

From the washed out sample. Like you get an "old photo" look when you color correct an "old photo". Or a blown highlights look when you color correct something with blown highlights.

The source IS what it IS.

Note: I used the Trench Coat to determine correct WB. And toned down the green a smidge. The eyes are overly sensitive to the green channel. That's why it looks "funky cold medina" in the orig. No other tools used. No contrast/bright/etc.

My source "IS" not like the one you posted. Look, the darkest darks are crushed anyway, there's no detail in them. Raising blacks up to RGB70 makes them murky gray, not black -- with no detail except the crushed-dark artifacts shown in your sample in the girl's hair, artifacts that will move with every frame and look like noise. The trench coat isn't white, either. If you wnt to get a white balance, try the bright edge on the window frame, not something that's not white. There's also such a thing as a "black balance" and a "gray balance". You can see in the histogram below that you've killed off all the darks, blown out the brights, made the girl's eyeballs red, and made her hair grayed-out purple.


[Attachment 20601 - Click to enlarge]

I don't think this shot is supposed to look like a faded old photo. It has angular light coming through the window and strong, moody shadows as in the other shots (which, I agree, are too dark--and too red). I worked for the white window frame, shadows on the window and curtains, black hair, convincing flesh tones, and whitish eyeballs. I don't even think the midtones are supposed to be that bright anyway, and certainly trying to make the hair lighter is useless--there's no detail there in the source at all, so IMO it's a waste of time trying to brighten crushed darks.


[Attachment 20602 - Click to enlarge]

[budwzr]

I didn't say I "sampled" the trench coat for WB. I used my skilled eye. Everyone knows a trenchcoat is like a vanilla/tan. I didn't blow any highlights. Are you boozin' over there? I'm gonna knock "one" back too then, and see if what you say makes sense.

Your Sample#2 above is horrendousl. Her hair is a solid black blob. And your colors are desaturated too much. Like it was shot in a morgue. Cold and dead.

Maybe time to get a new eyeglass prescription I'll accept critique from Smurf only.

[sanlyn]

Well, make it whatever you want. As for the black hair, mine is a black blob and yours is a gray blob with gray artifacts, LOL!

Everyone knows that the forum's pic viewer makes images look darker than they really are, but take a look at the histogram of your sample. Luma and every color is crashing against the right hand side. I believe that describes blown-out brights, does it not? And I think there's something amiss with your monitor setup. You should be able to differentiate every distinct panel in this graphic, which has been set up for sRGB monitor display:


[Attachment 20605 - Click to enlarge]

You can interpret the scenes of a movie in any way you wish. I look at scenes that tell me the director's intent. I don't think this film is a faded photo nor a movie musical. Looks like a moody, serious piece to me. But whatever you like....

[budwzr]

If you look at the right cuff-end, you will see that I have more detail. Proof's in da puddin' Your histo reflects all the noise, and limited dynamic range of the original. Not anything "I" did.

I can cook up a jiggered histo too.

[sanlyn]

budwzr, I know not from whence you come or what you're looking at.

That histogram was from VirtualDub, copied as-is to the clipboard. Try a pixel sampler. Pixel value readings from your sample:
Darkest part of the girl's hair on the left side: RGB 58-57-58. It's also the darkest color in the sample, including the small bit of border in the corner.
Light blocky artifacts in girl's hair, left side: RGB 79=79=79. That's a low-medium gray. In fact it's just plain, unaltered gray.
Darkest part of the skin shadow, left side, left part of nose: RGB 68, 63, 57 (that's not a skin color. That's grayed medium brown, not enough red, too much blue)
On the right side, dark part of girl's hair: RGB 84-54-56. Definitely not black, and not even gray. It's dark grayed magenta.
On the right side, shadows in the "white" curtain: RGB 155-147-134. That's not gray. That's tan. Of coursed, it could be the curtains ain't white to begin with. Who knows?
On the right side, shadow on the "white" window frame: RGB 133-188-112. That's a long way from gray shadow on white wood. It's grayed-out red.
On the right side: sunlight on the window frame: RGB 223-224-228. Congratulations. It's almost real white. But a white window with red shadow? Doesn't compute.

Apparently your "skilled eye" jiggered your brain into thinking purple hair is black hair.

The main point is that you enjoy your results. I have no argument with that.

Come to think of it, I wonder what really was the color of the sunlight in the window, in the original movie? I'll bet it was slightly yellow.

[budwzr]

Nobody else weighed in on this, so I win.