Estimating the original colors of the original Star Wars trilogy — Page 9

Dreamaster said:

Secondly, Poita has a full scan of the movie that he has done independently that is red faded and Dre is working on obtaining it?

He has multiple scans, including Technicolor and the LPP. But I think he does have an Eastman one, too, which would be faded. Do you just mean to test Dre’s algorithm?

Poita has done cleanup work on the THX film with high dollar software… so would his scan be “cleaned” like his THX sample or will it be “raw/grindhouse”?

Eventually, that would of course be ideal.

ScruffyNerfHerder said:

Hey DrDre,
This looks amazing! I make it a point not to swear, but holy shit your work impresses me. 😛 I did have two questions. You use Matlab to create your algorithms, correct? As there is a Mac version of Matlab, is it also possible to create a Mac version of your completed applications? I don’t want to add to your hassles if the process is complicated, but if it’s a matter of hitting a button and spitting out a Mac app, I for one would appreciate it. I do have Bootcamp, and quite a few old photos and videos that I’d love to put through your process either way.

Keep up the amazing work! 😃

Thanks! Sadly, to compile a Mac version of the tool requires that you have a Mac version of MATLAB, and a Mac. Unfortunately, I have neither. An alternative would be to obtain a Mac version of MATLAB, and run the tool from there.

I would have posted this a couple of days ago, but my computer has been out of service (network issues - had to replace my router and a switch). But today I had a chance to import DrDre’s reel 1 into Vegas and compare it to my GOUT color correction. It is so very close on some of the shots. You can see the same colors in both, but the color in the GOUT has a bit larger pallet. Not surpising since the prints they were scanned from are separated by several generations. I’m going to continue my BR color correction following my belief that the GOUT is a good guide to match to.

And while I can understand the desire to recreate that original 1977 experience, that is not my goal. What I want is an authentic scan of the original negative with the original color timing. The Tech IB prints are the closest we can get our hands on, but they all seem to have a noticeable tint. And pretty much no-one doing a major color restoration is relying on the Tech IB print colors. Add to that worrying about bulb color is kind of pointless. We need to know what the original colors of the print were first and even then I have no interest in a version based on any given bulb. I want the original colors captured on film - what the 3 color separation master would reveal if they would bother to have it restored (it would create a digital master that is superior to anything we have ever seen). In 1997 restoring 3 color masters that have shrinkage was impossible. Today it is commonplace. Until they do that, we will have to use our best judgement. This project is the closest anyone has yet come because it is revealing the original colors of a faded print. It is unfortunately a copy of another print, but it is still as close as we have gotten and it shows a close relationship to the GOUT colors (an inferior image, but good color information hidden in there).

This might not be a dream for much longer, if Dre has anything to say about it.

I’m working on the second reel of the SSE Grindhouse, which is in much better condition than the first reel. Here’s one of the Obi-Wan meets Luke shots before and after the automated color correction.

Before:

After:

Like on the Senator print photos, the automated correction says there should be a blue sky in this shot.

Senator:

To be continued…

What happens to the uncorrected 35mm Rancor shots with your process? I know Harmy had a tough time dealing with the very different brownish home video versions and blueish film versions.

towne32 said:

What happens to the uncorrected 35mm Rancor shots with your process? I know Harmy had a tough time dealing with the very different brownish home video versions and blueish film versions.

I would need to have access to a large portion of the first reel to make an accurate estimate of the colors. The correction is in principle always done in relation to other shots to preserve the original theatrical color timing.

Here are some shots from the dinner scene at the Lars homestead, corrected using the algorithm. I adjusted the brightness somewhat, as the print is very dark for these shots.

Could you please always write the source of each picture? They look nice, but way too bright btw 😃

Here are the unadjusted results directly from the algorithm.

That’s fantastic, Willarob. Thanks very much for continuing the work on the SSE!

I think there might be a little too much pink/red in the SSE 1.5 version. That frame might be a little dark and funky in the raw print, so the correction is probably not very accurate, but after matching the brightness to the SSE 1.5 I get:

and the other two frames:

While I agree the Luke frame has some problems, that need to be addressed specifically, rather than applying a global correction for the reel as was done in this case, I think, aside from being a bit murky, the colors of the other two frames are very accurate.

Raw scan:

Automated color correction + brightness adjustment:

Senator print photos:

It’s interesting to go back to the comparisons for this shot we did a while ago in the bluray regrade thread. Both NeverarGreat and myself presented our regrades based on the available references at the time.

Bluray regrade by NeverarGreat:

Bluray regrade by DrDre:

Automated color correction of LPP:

Senator print photo:

The most important question for this thread is, whether the automated correction provides colors that are accurate enough to be used as a reference? Considering that the automated regrade closely resembles the senator print photos, the Technicolor print scans, and in terms of color appears no less accurate than NeverarGreat’s manual correction, and my own correction of the bluray, I would say it is.

clutchins said:

There’s just that sharp blue sheen in the automated correction and I don’t know where it would come from or why.

There’s a dark purple edge around the raw frames, that is interpreted as color by the algorithm, and therefore also “corrected” to a brighter more obvious blue streak. I didn’t bother to crop the frames before I processed them.

Probably my attempt can’t be used for a reference but I like the high temperature on these frames

http://i.imgur.com/4sc4pAt.png

http://i.imgur.com/txMUqic.png

http://i.imgur.com/FgJIj8J.png

Yeah, all the dinner scene “corrections” look totally wrong, with blacks turning into strong blues. It’s especially obvious with the Senator next to it, which doesn’t show this blue sheen at all.

clutchins said:

Ah, so it just needs to be properly matted to theatrical AR.

Yes, indeed.

Laserschwert said:

Yeah, all the dinner scene “corrections” look totally wrong, with blacks turning into strong blues. It’s especially obvious with the Senator next to it, which doesn’t show this blue sheen at all.

The problem is that in order to make a comparison with the Senator print photos I had to increase the brightness substantially, so a very slight black imbalance [0 0 1] in RGB now becomes [0 0 7], which leads to the blue sheen. I actually know where the imbalance comes from. I updated the algorithm, and forgot to include the black balance. However, it’s easily fixed. I will debug the algorithm, and update the frames.

The thing I keep coming up with when looking at correcting ANH is that you can’t go too bright. I feel strongly that some of the crushed blacks in the DVD/BR are supposed to be. That is why few of us remember seeing the garbage mattes or Darth Vader’s red eyes until we saw it on TV. Those things are artifacts of a presentation that is too bright. Funny how my attempts to color correct Star Wars are leading to me looking at other movies differently. I’m not just comparing it to on set photographs, but to other movies. I think that because ANH is filled with sand and artificial corridors (either in white or gray) we tend to think of it as a fairly low color movie, but R2, Threepio, the many lights, laser blasts, explosion, and light sabers, as well as some props and costumes, are actually quite colorful and a lot of times those colors are diminished too much, I think in reaction to the almost neon tone of the DVD/BR releases. I think doing a sample correction to compare it to a particular color reference is great, but those brightened versions are way too bright. I prefer the darker version done just with the algorithm.