Star Wars GOUT in HD using super resolution algorithm (* unfinished project *) — Page 43

DrDre said:

No that's not Harmy's color correction. Here's Harmy versus the corrected Harmy (when I say corrected I mean to the reference frame):

http://screenshotcomparison.com/comparison/138320

 Thanks. Shows that I really can't gauge these things or remember how a source looked without direct comparison. His is perhaps in the same direction, but extremely restrained compared to the print.

Atrocious or not they are actually a pretty close match to the Technicolor IB print that is viewed as a reliable color reference, right down to the greenish color of the Death Star walls:

http://screenshotcomparison.com/comparison/138323

http://screenshotcomparison.com/comparison/138324

http://screenshotcomparison.com/comparison/138325

http://screenshotcomparison.com/comparison/138326

http://screenshotcomparison.com/comparison/138327

http://screenshotcomparison.com/comparison/138328

http://screenshotcomparison.com/comparison/138329

http://screenshotcomparison.com/comparison/138331

It would be interesting to see how such a color correction would look on a true hi-def source, but from these caps I would say it is definitely possible to do a pretty accurate color correction from a single frame. I mean the frame a used as a reference wasn't even from the same print.

Update: I noticed some of the frames in the  Technicolor IB are not exactly the same, but I guess I got my point across.

I see you caught it before I switched my wording to hopefully be less insulting. :P

Anyway, it would be interesting to watch the whole film like this, especially on a projector. It's jarring to me at first, especially in comparisons. But I find that I adjust to a film's color/contrast rather quickly.

Hahaha, no insults here. My main point was to show off my single frame color mapping methodology, but it also raises some interesting points with regards to the actual colors of the film.  

These corrections look totally wrong. The Technicolor caps too, they're burnt out.

It takes some getting used to, but this print is seen as being in pristine shape. It was screened at the now closed Senator Theatre a number of years back. Here are more photos from this screening:

http://photos.petergaultney.com/Movies/Star-Wars-at-The-Senator/i-LfFw9q9

In any case, I was simply trying to match the print. Whether the print has the correct colors is a matter for debate, but I think it would be a bit presumptuous to call them totally wrong, as it is one of the few prints from that era that has survived in good shape. Technicolor prints are known for their saturated colors, and their ability to retain their colors virtually unchanged for decades.

DrDre said:

It takes some getting used to, but this print is seen as being in pristine shape. 

 Where did I talk about the print?

Sorry, I misunderstood. Could be that the photos are not actually representative of the actual print. 

I think what he's trying to say is that just because the print is in great shape does not mean the scan is representative of it. Hence the Harmy/Mike issue I mentioned, and the work that some of our 35mm experts here do post-scan, matching the scan to the print.

That may certainly be true. On the other hand these photos have been the basis for a large number of color correction projects on this forum. However, it would of course be better to compare directly to scans of the print. 

The way the color mapping works, is that you first take a test frame and a reference frame, and you match the cumulative histograms with a technique called histogram matching:

https://en.wikipedia.org/wiki/Histogram_matching

From the histogram matching you obtain a mapping between the R, G, and B colors in the test frame, and the reference frame, that can then be used to correct the colors of other frames. This should work as long as the color fading or color alterations are similar across frames. 

I used MATLAB for the color correction.

DrDre said:

That may certainly be true. On the other hand these photos have been the basis for a large number of color correction projects on this forum. However, it would of course be better to compare directly to scans of the print. 

You mean people used those photos taken in the theater as reference? What a horrible idea. I don't know what they were thinking.

Even film scans go wrong many times. Photos about film frames? Oh, my.

Well, they were probably assuming Michael Kaminski, who atttended the screening, was right, when he said these photos are the best reference for how the film is supposed to look:

http://fd.noneinc.com/savestarwarscom/savestarwars.com/technicoloribscreening.html

Of course they are not absolutely accurate for a number of reasons, but they are said to to be in the ballpark. 

Yeah, I have seen these pictures on that site many years ago already. But I never thought they'd be useful color correcting, or that anyone would use them for any practical work.

There are a million ways to get the caps wrong. It counts a lot how the film is projected to begin with. Any accidental auto-setting on the camera they take the photos with. And so on. It's crazy.

Anyway, the photos could be completely inaccurate, but this was mostly about testing the method. The reference frame I used from a different print is pretty close to the photos, as are the color corrections that are based on the reference. It may be a coincidence, but it may also indicate the photos are a reasonably good representation of the Technicolor print. I suppose Mike Verta would know the answer to that one.

DrDre said:

it may also indicate the photos are a reasonably good representation of the Technicolor print.

It would (in theory) only indicate that the photos are consistent, and not that the color balance - or anything else for that matter - is right.

However, I disagree that the similarity/difference between your caps and the IB caps would be particularly consistent. Some caps are totally different. But it is to be expected, it's not due to the histogram matching method.

One correction for the whole movie would probably never match the GOUT to an existing print color-wise. And of course, I didn't criticize the experimenting in any way. I've only stated that the photos are not reliable, no matter how much good intent is behind them.

@ zee944

It's just a good debate and critisism is good so keep them coming.

@ Laserdisc Master

Thanks! Missing you here buddy!

Maybe you can ask Mike Verta for his color corrected version of that frame and use that for your color mapping.

SR only scales by factors of two, so 720x480 to 1440x960 to 2880x1920 etc. Any other ratio is done by using a spline. I'm wondering if I can answer the question.