Sign In

Color matching and prediction: color correction tool v1.3 released! — Page 11

Author
Time
 (Edited)

The artifacts in C3PO's leg are greatly reduced, although there's always going to be some artifacts, because of crushed whites. No method at this point can predict what isn't there, at least not without introducig problems in other areas.

Bluray:

Bluray regraded (using old reference, not Tech IB print):

Author
Time

What about this reference...

http://www.youtube.com/watch?v=dprxDRIq23I

seems like it's closer to the Blu-ray to me than this color matching program. Not saying it's bad but to me I think there is something you are failing to take into account... Common sense on skin colors and so fourth. Over saturated Blues and Reds should be very apparent to you by now in your output... You can see that it's just not a natural range of color.

Author
Time
 (Edited)

Not sure if this is applicable, and it's way before your time here, but there was a user named snicker who had an interesting idea about clipping (with actual good results!).  Basically because not all channels were boosted equally, not all channels clipped equally.  For example, on a lot of clipped colored lights, the red channel is clipped, or possibly the blue channel, but the green channel retains some detail, which could be used in restoring the shot.

Anyway, not sure if what he did was exactly scriptable, but he produced some impressive results, where previously clipped/crushed data seemed to be restored to bright red lights and I believe also the black "pit" in the death star hangar.  Anyway, you might find someone here with more technical detail, but I thought since you're messing with clipping anyway, it couldn't hurt to mention it...

Project Threepio (Star Wars OOT subtitles)

Author
Time

Ronster said:

What about this reference...

http://www.youtube.com/watch?v=dprxDRIq23I

seems like it's closer to the Blu-ray to me than this color matching program. Not saying it's bad but to me I think there is something you are failing to take into account... Common sense on skin colors and so fourth. Over saturated Blues and Reds should be very apparent to you by now in your output... You can see that it's just not a natural range of color.

 The fact that it's closer to the bluray is not surprising. It is a showing of a new 35 mm print, probably based on the same master as the bluray, so it's colors will be pretty much identical. So, it's not a very good reference for how the film looked in 1981. The reference I'm using is from an actual 1981 print. Whether it's accurate is of course questionable, but the colors are a lot more saturated than the bluray, and closer to what comes out of the color matching algorithm. 

Author
Time

How accurate are the colors on odd film cells being sold on eBay, added to collector's BD editions, etc? Does anyone know the process of making them?

I'm asking, because I imagine that those generally come from specifically struck prints containing only one, or a couple of frames replicated over and over again. Are those prints made from the internegative used for theatrical prints (or rather, parts of it), or is the process different in any way?

DrDre's software is a big deal, because in theory buying only a couple of frames and scanning them would give us a very good approximation of original colors. The possibilities would be endless.

Fanrestore - Fan Restoration Forum: https://fanrestore.com

Author
Time
 (Edited)

Ronster raises a point that I have been thinking about too.  What it looked like in theaters and what it looks like in real life or what the intention of the look was supposed to be are often two different things.  It's almost like there needs to be two versions of color fixing for a film.  One is to make it look like it did in the theater (good or bad as it was), and the other is to make it look correct as if it happened right in front of you.  I hate all of the orange and teal they are doing these days.  Yes, that is what it looks like in the new Star Wars trailers (orange and teal in at least the first one that came out), but looking at the trading cards in comparison which have the original colors, I much prefer the look of the trading cards.   

http://www.starwars.com/films/star-wars-the-force-awakens-digital-trading-cards

vs.

http://www.youtube.com/watch?v=erLk59H86ww

at 44 seconds.  His suit is fluffy baby pink blanket instead of orange, and where did his 5 o'clock shadow go?

My personal vote is to make it look correct in terms of skin color, flag color, etc. taking into account the lighting that affects this.

CatBus's comment is cool too.  This can get into another issue where HDR would be required in order to have all of the color properly.  To keep all of the detail and both the high and low end, the colors need to be "compressed" into the Rec.709 colorspace which is a small subset of the real world visible color range.  Your eyes deal with the wide range by opening and closing the iris. When you are outside on a bright sunny day and walk into a relatively dark room, you can hardly see any thing (sort of like black crush).  After your eyes adjust, you can see a lot better.  If someone filmed from inside the room and there was a window that could see the outside, most likely the window would have mostly white blown out details.  If they then went outside and filmed a scene with a window to the inside, it would be mostly black. This is because the camera has an iris too.  If the camera was HDR capable, it would most likely preserve more of the details and on an HDR display they could be shown properly.

Also other factors play into it.  The bulb in the projector degrades over time and will produce a different look as it is projected.  Scanning a 35mm film doesn't mean the result is what it looked like.  I know this first hand from my trailer scanning project.  You need to scan at different settings with multiple scans to capture all of the dynamic range.  Any one of those scans looks different than the others, so which is correct?

Lots of rambling, but in the end, my take is that the color correction should strive to reproduce the lighting and color of the scene as it looked in the real world as much as possible, unless they purposely altered the look of a particular scene to add more atmosphere to it such as for a scary movie.

Author
Time
 (Edited)

I guess I should add that while snicker used the RGB channel differences to reconstruct clipped highlights, I think in the Death Star hangar example he uncrushed the blacks by pulling usable detail out of the BTB data, and I'm not sure if you've been making use of what unseen detail is lurking in BTB and WTW.

Project Threepio (Star Wars OOT subtitles)

Author
Time

CatBus said:

I guess I should add that while snicker used the RGB channel differences to reconstruct clipped highlights, I think in the Death Star hangar example he uncrushed the blacks by pulling usable detail out of the BTB data, and I'm not sure if you've been making use of what unseen detail is lurking in BTB and WTW.

There is definitely a great deal of data hiding in the superwhites or WTW in Star Wars, and it can be pulled out if you don't clip the detail through a colorspace conversion, but how would it be possible to pull useful data out of the superblacks, since most of the data is random noise within a point or two of RGB 0,0,0?

You probably don’t recognize me because of the red arm.
Episode 9 Rewrite, The Starlight Project (Released!) and ANH Technicolor Project (Released!)

Author
Time
 (Edited)

I think I might have come up with a better solution for the crushed blacks/whites. Going back to the The Empire Strikes Back frame with a color correction model I calibrated on a frame of Star Wars, I had the following problem giving the frame a darker grading:

Semi-Specialised Edition V2.2:

This is what I got:

The source frame suffers from crushed whites, so the model could not correctly estimate the colors in those areas, resulting in artifacts.

This time I decided to upgrade the image with a new algorithm I developed to estimate what an image with a higher color depth would look like, and use that for the color correction as a source frame. The source frame looks identical on the surface, but when I color correct it now I get this:

I think this looks much better.

Author
Time

Multi color space model predictions are now just as robust as single color space models. Since both the single and multi color space model are able to match the Raiders reference frame, their predictions should be virtually identical.

Single color space model prediction:

Prediction of multi color space model:

Author
Time

Is this due to your highlight compensation?

Preferred Saga:
PT: Hal9000
4/5: Adywan
6: NSP
7/8: Hal9000

Author
Time

nightstalkerpoet said:

Is this due to your highlight compensation?

 No, one of the stabilization parameters I use was set a little too high, resulting in a lower accuracy.

Author
Time
 (Edited)

That's awesome you figured it out. New release coming soon then?

Small request - could you include your release number in the Program Name at the top of the tool, and keep this thread title updated with the current release.

ie DrDre Color v1.2 and Color matching between two sources - GUI v1.2 released

also, would this help you set up video input support? http://stackoverflow.com/questions/15538730/reading-and-displaying-video-file-frame-by-frame

Preferred Saga:
PT: Hal9000
4/5: Adywan
6: NSP
7/8: Hal9000

Author
Time

nightstalkerpoet said:

That's awesome you figured it out. New release coming soon then?

Small request - could you include your release number in the Program Name at the top of the tool, and keep this thread title updated with the current release.

ie DrDre Color v1.2 and Color matching between two sources - GUI v1.2 released

also, would this help you set up video input support? http://stackoverflow.com/questions/15538730/reading-and-displaying-video-file-frame-by-frame

 Sure, on that point, how do you change the title of a thread?

Author
Time
 (Edited)

On your original post when you started your thread, edit it and change "Post Title" section.

I'd also suggest you put a direct link to the current GUI download in your signature (edited under Profile tab at the top of the page). Since it's not copyrighted film material, it doesn't fall under the no direct downloads rule.

Preferred Saga:
PT: Hal9000
4/5: Adywan
6: NSP
7/8: Hal9000

Author
Time
 (Edited)

After being convinced by a few members that the Raiders regrade is too saturated, I decided to use a different reference, namely a 1981 publicity still:

Bluray:

Bluray matched to reference (slightly corrected for blue shift): 

The sky is now a natural blue, and the colors are very balanced. Interestingly, predicting the frame that was matched to the former reference frame leads to a result that is close to the former reference, without being oversaturated:

35 mm frame:

Bluray:

Bluray corrected:

More in the Raiders of the Lost Ark regrade thread:

http://originaltrilogy.com/forum/topic.cfm/Raiders-of-the-Lost-Ark-35-mm-regrade/

Author
Time

NeverarGreat said:

CatBus said:

I guess I should add that while snicker used the RGB channel differences to reconstruct clipped highlights, I think in the Death Star hangar example he uncrushed the blacks by pulling usable detail out of the BTB data, and I'm not sure if you've been making use of what unseen detail is lurking in BTB and WTW.

There is definitely a great deal of data hiding in the superwhites or WTW in Star Wars, and it can be pulled out if you don't clip the detail through a colorspace conversion, but how would it be possible to pull useful data out of the superblacks, since most of the data is random noise within a point or two of RGB 0,0,0?

Well, I don't know the technical details behind snicker's methods, but there was clearly something usable there for that one shot.  Either it was specific to that shot, or maybe the channel separation trick provided some benefit there too, i.e. separating signal from noise.

Project Threepio (Star Wars OOT subtitles)

Author
Time

DrDre, those latest Raiders pics are spot on beautiful and perfect!  Please don't tweak them further.  They look gorgeous!

Author
Time
 (Edited)

Here's another interesting color match/prediction. This is what you get when you match the Star Wars bluray to Team Negative1's scan of the 35 mm print of the 1997 Special Edition:

Team Negative1 35 mm scan:

Bluray: 

Bluray matched to 35 mm:

Bluray:

Predicted correction using above reference frame:

Author
Time

CatBus said:

... there was a user named snicker who had an interesting idea about clipping (with actual good results!).  Basically because not all channels were boosted equally, not all channels clipped equally.  For example, on a lot of clipped colored lights, the red channel is clipped, or possibly the blue channel, but the green channel retains some detail, which could be used in restoring the shot.

An excellent and creative use of a film's own resource. It might have been an original idea snicker had:

http://originaltrilogy.com/FORUM/topic.cfm/New-colormatching-script/topic/15002/

.. but I also saw the technique referenced in some movie restoration documentary (was it The Wizard Of Oz  Blu-ray restoration?).

His version wasn't 100% effective because he didn't take the idea far enough -- to realize that the clipped channel (if only one or two of the R-G-B layers) first must be compressed to make room for information re-purposed from a remaining, non-blownout channel.

Author
Time

As with many Technicolor IB prints, there appears to be a slight green shift in the print, as is evident from this frame:

When we compare this frame to a 1977 publicity shot we see this:

Of course we can correct the Tech IB matched frame, such that we get this:

Author
Time
 (Edited)

Yes, I think poita said that some of the IB prints had a green shift. (I'm assuming this would not be due to fading - since IB prints don't fade - it would be due to the color getting shifted in the original printing process itself.)

And again, be very careful with publicity stills - they weren't timed to match the theatrical color timing. (FWIW, the 1979 book "The Movie Brats", in a reference to the Nazi imagery of the villains, said "Tarkin lives in a gray-green world with gray-green uniforms", so what people would have seen in theaters - not just on IB prints, since those were only in the UK - has to have had at least a smidge of green in it.)

Great exercise, though - this color script is absolutely amazing.

Author
Time

I'd be very curious to see the prediction results from that, since it touches on all of the main colors. 

Preferred Saga:
PT: Hal9000
4/5: Adywan
6: NSP
7/8: Hal9000

Author
Time

nightstalkerpoet said:

I'd be very curious to see the prediction results from that, since it touches on all of the main colors. 

 It looks pretty awful, most of the other scenes don't have the same green hue, so it looks like it was delibirately shot this way. 

Author
Time
 (Edited)

Moved from the SR thread:

To give you an idea here's a comparison for frame 8228 between the bluray, Harmy's Despecialized Edition 2.5, and the bluray matched to the Technicolor IB print scans I have.

Bluray:

Harmy:

Tech IB:

and let's throw in the bluray matched to the GOUT for good measure. Note the reduced contrast, leading to some artifacts on Vaders's cape, caused by the crushed blacks in the bluray: