Sorry for the late response. I’m working on something similar for faded film prints. You can find it in my other thread. Is this what you mean, or do you have something else in mind?
Here the video sample with the correct colors:
https://drive.google.com/file/d/0B8_LYKyZDiajZzByMFdrUjhNVHc/view?usp=sharing
Before:
After:
Here’s a video sample of the SSE Grindhouse color corrected using the automated color correction algorithm:
Video deleted
Edit: sadly something went wrong converting the file from raw to mp4, as this is how the video looks:
While this is the shot in the raw version as it’s supposed to look:
Dumb question: is it really fading that’s the issue in blue frames like this? It’s not the scan or preliminary correction or something that happened in the duplication of the print?
I’m not sure. The scan quality improves significantly directly following the Stardestroyer shot, and is pretty consistent, aside from the really dark shots, which have no shadow detail. The color cast also changes from blue/purple to mostly blue. I guess we’ll find out, once williarob recieves the scan poita made.
Darth Lucas said:
Crazy what can be pulled out of it with a little work though.
Here’s another challenging shot. The print is pretty bad during these early shots, with pretty uneven fading. Love that blue color on R2 though…
DrDre, is it possible to create another algorithm that could estimate what the previous generation print would have looked like? We have the GOUT, the Tech IB scan, and the TN1 scans. I’m just wondering if it is possible to estimate what both a first run theatrical print and a studio presentation print would have looked like based on what we have available.
I think it should definitely be possible.
Dre,
Just wonderful colors that you’re digging out of that!
Is it possible for you to post a shot of the opening star destroyer?..around the 2:20 time marker perhaps?
Great work as usual Dr!
That shot is in pretty bad shape, but here’s one of the frames:
The first few minutes of the scan has the worst fading, with the left side of the frames more purple, and the right side more blue.
towne32 said:
But my approach was definitely not an unbiased one. Your most recent ones look great, and much better than some of your other recent tests IMO. Might need some adjustments in contrast (your comparison to Mike’s makes that clear) or maybe saturation, but it’s a starting point that I would’ve thought impossible, and honestly needs little extra work done anyway.
It’s kind of an interesting dilemma. I personally belief the print originally was pretty contrasty, being a dupe print, and all. If the aim is to preserve the print, should we preserve these imperfections, or should we strive for the ultimate home video experience?
Take the example of this pink shifted 70mm print of Chitty Chitty Bang Bang:
This is the type of fading the algorithm can easily handle:
As you can see the colors look like it was printed yesterday. The algorithm attempts to restore the colors to their original state. If the print originally was contrasty, the results will reflect that. If not, same story.
Ugh these colors are just gorgeous. They honestly are my favorite color palette done so far. Harmys 2.5 is too yellow and I thought 2.7 fixed everything but for some reason the colors here are better looking to my eyes. Do you know if there’s any color differences between like those tatooine shots versus 2.7 colors?
The yellow color in Harmy’s 2.5 is probably the result of using the Technicolor print scans that have been our main reference for the color of these shots. I’ve suspected for a while that those colors were not accurate, and I think these results confirm those suspicions. The reason I’m so sure is that I calibrated the model on the Tantive IV scenes only, and these colors came out naturally when I corrected the rest of the reel. So, I believe these colors are far more representative of what people saw, when they went to see Star Wars in theatres in 1977. I think towne32 was able to undo most of the color issues of 2.5 for 2.7, so it should be closer at least. It would be nice to do a gallery of comparisons at some point.
The early shots of the reel require a separate correction, so the video sample will take a little longer to prepare. Here are a few more processed Tatooine shots:
You are a marvel, Doc! Would you be able to recover colour from a nitrate yellowed film too?
Probably, I guess we’d have to try.
Deleted
Why’d you remove the tidbit about removing digital color casts?
It wasn’t really relevant for the thread.
So I am an idiot when it comes to these things so I’d like to first say that your colors simply look amazing6! Great work. And now onto my idiotic question
How is this actually used as an algorithm. Does it integrate into Adobe CS6 (my personal current choice of creative software) or is it a standalone thing. Again I’m sorry for probably such a stupid question
There are no stupid questions. At this point it’s just spaghetti code, but it will be integrated into the color correct tool at some point, which is a standalone tool.
… and here’s another funny experiment. We know there are faded 70mm prints of Star Wars in existence. I found a very low resolution set of film cells online, and wanted to see if it’s possible to salvage the colors:
As it turns out, the colors can be retrieved:
Here’s a little experiment on a severely pink shifted frame of Star Wars that you would expect hardly contains any color information, but looks can deceive:
Deleted
Deleted
I would be interested in seeing what your color matching tool does matching the blu-ray to these 35mm adjustments.
Me too, so that’s definitely on the agenda.
This is also an interesting comparison.
Star Wars Legacy Edition (preliminary):
Automated color correction:
Double post
Here are a few more examples:
I’m preparing a video sample, but while we wait, here are three more corrected frames for reel 1:
If I well understand, neutral color means equality between red, green and blue in the shot. So, how do you know that the majority of the Tantive IV scenes have neutral color ?
If there’s a little green in the walls, red lights, blue shirts etc., what allow us to suppose the balance is perfectly grey ?
Neutral color for a single frame has many definitions, but in general it means no delibirate color cast, not necessarily equality between color channels at each intensity. For single frames you have to set the strength of the amount of color balancing, to allow for the fact that colors for single frames do not have to be perfectly balanced at each intensity. If there are few colors in a frame, the quality of the result will be highly dependent on the color balancing strength you use, and the frame you’re attempting to correct. However, if there are enough different colors in a frame, the result will generally be very good, and far less sensitive to an increase of the color balancing strength, once a color cast has been removed. However, the method remains most suitable for color correcting multiple frames simultaneously, where the assumption that colors average to gray is generally an accurate approximation.
One of the challenges involved with correcting an entire reel with one model, is that color fading may vary a bit across scenes. For example the Vader/commander shot still has a slight teal cast:
The question then is, whether the shot originally had a slight color imbalance, or whether the shot was slightly more faded than other shots. However, to make it more appealing, you can apply a second conservative correction with the algorithm for the specific shot or sequence of shots (keeping in mind the points raised by g-force, and the assumptions underlying the algorithm):
two things doc…
This is worthy of a Nobel Prize. Not joking, millions could benefit from this technology.
I think you would end up removing any intentional color timing from the film. Think about it. Take a photo, adjust the colors without clipping any of them too badly. Take the original photo and do it again. Do it in whatever color space you want. Now hit the button on each of the differently timed photos. They all get back to being pretty neutral, right? I’m not complaining, as this is an amazing tool in color recovery, but I don’t think you can say these were the colors on opening day.
-G
I agree you would, if you would correct shots individually. However, the idea with correcting a film reel, is that you calibrate the model on shots you know to have a neutral color (or no delibirate color cast), like the majority of the Tantive IV scenes. If you include enough shots, the average distributions of the red, green, and blue channels should converge for an unfaded print, allowing you to accurately estimate the color shifts for each color channel, and for each color intensity. You then apply this correction to the entire reel, which is what I did in the examples I showed for Star Wars. All the color relations between shots remain intact, since you apply a uniform correction across scenes. Unless the shots you chose originally had an intentional color cast or an unintentional color imbalance (and even in then it would have to be an imbalance that persists across scenes or an entire reel), the final result should be an accurate representation of the original color timing.
