Estimating the original colors of the original Star Wars trilogy — Page 10

There is a bug in the latest version of the algorithm that needs fixing, which is causing problems in the darker tones. I implemented this version last weekend. It has some clear advantages, and I think I know where the problem is. Hopefully it will be fixed by tomorrow.

Deleted

Awesome, thanks poita, that’s such a difference! The problem of the blue sheen was caused by a programming error I made while updating the algorithm. The problem has been fixed.

Here are the corrected frames based on your scan. You can notice, that even for this high quality scan, there’s some blue noise in the very dark areas. The reason for this, is that there’s actually blue noise in the scan. It’s pretty faint, but it’s the same color as the darker areas of Beru’s shirt, for which the blue is desaturated:

So, by restoring the color of Beru’s shirt you also enhance the noise. You simply can’t do one without the other.

Interestingly, the colors are now an almost perfect match to the Technicolor scans I have, including the Luke shot, aside from a slight pink shift in the Tech scan.

I also tested the algorithm on one of the The Empire Strikes Back frames you posted a while ago in your thread. After applying the algorithm, I got the following results:

Raw scan:

Automated correction:

…and after a gamma correction:

It’s interesting to note that there’s significant amount of blue and yellow noise in the print.

poita said:

The ‘noise’ is actually in the print, the blue layer is terribly grainy, and that is true for film generally, the blue layer has the most noise.
In this case, I also haven’t set the black levels, so they may be too high in the blue layer, causing you issues, the black levels should be set so that the base noise level is obscured.

Just on film grain, many people incorrectly think it is the grains are the ‘pixels’ that form the colour information. In reality the ‘fundamental particles’ that hold the colour information are the silver particles and colour dye clouds. The grain is an order of magnitude larger than these particles and is effectively laid over the image, obscuring it with noise, it is not the source of the image itself.

Thanks for the interesting info! So, is film grain actually a side effect of the print process?

I’ve been wondering why the Technicolor print scans we’ve been using and still are using as a color reference, aside from the obvious green shift, also seems to have a slight pink shift, when the common belief is, that they’re not supposed to fade. The sky on Tatooine is more often than not a shade of purple rather than blue.

Today I came across this quote:

“If you took a Technicolor dye imbibition print and you projected it many times at a drive-in theater and put a lot of light through it, the dyes do fade a little bit — particularly the cyan — although not that badly. But if you are careful with Technicolor imbibition prints, and keep them in the dark and don’t show them a lot, they don’t seem to fade at all. I’ve never seen one that has faded if properly cared for. It is a remarkably good process. Technicolor imbibition on triacetate base is very, very good.”

So, although Technicolor prints are extremely stable, they are not entirely fade proof, and the pink shift in the Star Wars print scans could have been caused by a very slight fading of the cyan layer.

I also came across this piece by famous film archivist Richard Harris called “A few words about…™ The Myth of Dye Transfer Printing”, in which he states that most Technicolor IB prints are, and I quote, “not useable as archival reference…While a normal, run of the mill, dye transfer print can usually provide a general concept of densities, it cannot be used for color”:

"Virtually every time someone makes a post on the web, I receive messages from people asking me to clarify discussions.

Possibly this thread can serve as a guide to how reference prints are considered, and used toward film restoration.

I presume that those who move back and forth between various sites, will copy and paste, thereby saving me the time and effort.

There are several types of potential reference prints, for both black & white and color productions.

The majority of prints, however, are faded, treated, burned and damaged in a myriad of ways. These are obviously of little use – not no use, but little.

Even faded direct positive photographic prints can serve a purpose, most notably as a guide to day for night and overall densities.

The single, and seemingly most confusing type of “reference print” is that produced by Technicolor via the dye transfer method, as they do not fade.

In the 1950s through early '70s, the number of prints produced for a national release could run around 300 - 400.

This would take multiple sets of printing matrices, as a matrix had a limited lifespan.

During a run of matrices – let’s arbitrarily pick 100 as an average number of prints per set – the color, densities and grain structure could change over a run of prints, as each matrix began to wear.

While the first dozen or so prints could have near perfect color, density and grain retention, the 80th, 90th or 100th, could appear different - occasionally slightly softer in resolution, and with color drifting via the three different color components.

Dye transfer prints were never sharp to begin with, due to the use of liquid metal dyes, and whatever mordant was used to make them properly imbibe to the stock.

Sharpness was more “apparent” than actual, as contrast was raised slightly to create a sharper appearance.

Where dye transfer prints shone was in their ability, as a second generation printing element, to transfer the original look and textures of large format films. In some cases, large format grain would become lost in matrix grain, and the overall image could be a silky and velvety marvel.

I’m taking the time to go through this, as there is a discussion occurring over at BD, in which someone is relating that because they viewed a dye transfer print of The Godfather multiple times in a theater back in 1972, that he has:

A. Total recall of the grain structure and color palette;

and

B. That the look and textures of the restored Godfather(s), as overseen by the filmmakers are incorrect – based upon his memory of what he recalls seeing in 1972.

This is a position that has been taken numerous times over the decades.

Which takes us back to the manufacture and distribution of dye transfer prints during that era.

Generally, when prints were produced, there would be a run of each reel in its entirety for the order, before the next reel went to process.

That means that of the 100 prints of the main title sequence, reel 1A, reel 1B and onward, that every print was slightly different from the previous.

While a reference print was always on hand, and many of these prints have been preserved, and are available as continued reference, drift of color occurred on a continuous basis.

That means that the 100th print could be two points (or more) toward cyan, yellow or magenta, up or down than the first. Re-issue prints were notorious for poor color accuracy.

After all of the prints were produced, and those too far off to be used were discarded, all reels were matched for color, unit by unit. As I recall, The Godfather was 20 units. Lawrence of Arabia was around 30, and Mad World, also around 29 or 30. That’s a great deal of matching.

Prime premieres, and major cities would receive the prints that hit their target precisely. Those up or down a point or two would go to second tier cities, etc.

This is the long way round of explaining that not only are most dye transfer prints not alike, but that the majority are not useable as archival reference.

Because I can only recall color and densities in a general sense, I do not depend on memory.

I need reference.

While a normal, run of the mill, dye transfer print can usually provide a general concept of densities, it cannot be used for color.

For The Godfather, with the cooperation of The Academy Archive, we were able to access the final approved Answer Print of the film for which cinematographer Gordon Willis had signed off. This was the print that he had screened and approved in 1972 via carbon arc projection (yet another anomaly) and which had retained its color.

During the restoration, this print was constantly accessed via 35mm projection on the same screen that shared the image of our data.

Nothing was left to chance. In the end, both director and cinematographer approved the final look of the restoration as matching the reference print screened before color work had begun, as closely as technologically possible.

We were extremely fortunate that this print had survived.

As another example, a complete pure reference print did not survive for My Fair Lady, but enough units, especially magnetic striped (which were generally produced to the highest standards) did, to allow us to get color and densities where they belonged.

There are very few dye transfer prints surviving that can be used a bona fide reference.

Which brings us back to the wonders of the web, and people innocently sharing their memories of prints viewed decades before, which may have not matched reference at that time, when they were new. Add to that the anomalies of projection: The color of the optics, the port glass, the alignment of the optical system, the cleanliness of the mirror at the rear of the arc lamp…

and of equal importance, the color the motion picture screen, which could add a couple of points of red or yellow to the image, as theaters allowed smoking at the time along with cool, refreshing air-conditioning.

Final thought. There are a few people - very few - who have color retention far better than others. One gentleman occasionally posts here. It’s a rarity.

RAH"

Here’s a cool picture of a cross section of a piece of film, showing the magenta, cyan, and yellow layers:

One thing to note, with a Technicolor print, you do not get very much added grain. That is a huge benefit in clarity. As a frequent viewer of TCM, I can tell you that they usually just telecine the prints they have on file in the archive for broadcast. You can really tell that the quality of some of the early Technicolor prints was not that great as even at 720p there is some obvious misalignment. It doesn’t diminish the fantastic colors or the fairly sharp image. Not as sharp as the high quality black and white, but sharper than some of the other color formats. Though I think with access to the original negatives and the color separations, modern computer technology could achieve a higher alignment. From what Mike Verta has done, I think you could even scan one of those old Technicolor prints and realign the color and adjust for any fading or printing issues and achive a decent (for home viewing or broadcast) HD picture quality.

For our beloved 1977 Star Wars, this does mean we can’t be 100% certain of the quality of the extant prints. Lucas obviously had the best one. They are a good reference, but the images that DrDre has been showing us tend to show that the Technicolor prints are a bit off in a variety of ways. I’m thinking, since they were shutting down production in 77, that they are probably print control issues rather than fade issues. I think if we combine information we are gathering about the color film stock prints with what we have on the Technicolor prints we will be about as close as we can get. We need to identify how the colors are different and apply that to a technicolor print scan to get as close to the original as is possible at this point.

Sorry to interject here, but I find this thread interesting despite being colorblind. I can see the differences, but not the subtle nuances discussed here. Still fun to try though. From what Ive deduced from reading this thread, there really is no 100% way to know what the true colors were without a pristine, unprojected print as a referance. Even then, time would be our enemy to some degree. Just for grins I threw together a comparison shot of a few sources to look at side by side. I had my GF (good color vision and good cinema knowledge) look at them and compare and contrast. She made an offhand remark that the top image looked best for cinema. Just her opinion though. I really have no knowledge on this subject, but I like to learn what I can by reading the threads here. One is SSE, one BD (obvious which one) and the third is something I had in a drawer. I already know the verdict on the 2011 BD (middle image), but I was curious about the top and bottom inage, one being the SSE and one I found here at home. Again, sorry to I cant really add to this thread.

Colson said:

What do you mean you had in a drawer? What is it?

Its one of those 70mm collector frames they made years ago. I had asked sometime back where they got the frames for those and was told they were “made” specifically for those collector thingees to sell. What does that mean? Did they just make faux prints from another source to look like it was cut from an actual 70mm reel? Anyways, I just wanted an excuse to scan it so I could actually make out the detail in it.
Also, being that I have difficulty discerning color anomalies, I thought Id post it in this thread for those of you who have a better eye. It helps me recognize certain things that I might not even realize without it being pointed out.

BobaJett said:

Colson said:

What do you mean you had in a drawer? What is it?

Its one of those 70mm collector frames they made years ago. I had asked sometime back where they got the frames for those and was told they were “made” specifically for those collector thingees to sell. What does that mean? Did they just make faux prints from another source to look like it was cut from an actual 70mm reel? Anyways, I just wanted an excuse to scan it so I could actually make out the detail in it.
Also, being that I have difficulty discerning color anomalies, I thought Id post it in this thread for those of you who have a better eye. It helps me recognize certain things that I might not even realize without it being pointed out. To my eyes, the top image has a brownish/slight yellowish hue to it.

It looks to me Bobbajet to have a slight lack of green, the faces just don’t quite look neutral to me, especially compared to the SSE. But the blu ray… lol what were they thinking? 😃

Here’s an interesting test case for the algorithm. In 2012 Jaws was released on bluray, after it was restored digitally at 4K resolution from the original negative:

The original negative was in a very poor state:

Although these scans are at low resolution, and rather compressed, we can use the automated color correction algorithm to attempt to restore the colors and detail. These are the results (without manual augmentation):

The colors are not as contrasty as the bluray, but they seem pretty authentic, and the skin tones are less orange, which I personally prefer.

AntcuFaalb said:

yotsuya said:

For our beloved 1977 Star Wars, this does mean we can’t be 100% certain of the quality of the extant prints. Lucas obviously had the best one.

No, Lucas had to borrow a 35mm IB Technicolor print from a collector to use as a reference for the 97SE. This particular collector was fortunate enough to receive his print back from LFL when they were done.

Serves me right for listening to what GL had to say. I was under the impression that he had a print in his personal collection that they used.

yotsuya said:

the Technicolor prints are a bit off in a variety of ways. I’m thinking, since they were shutting down production in 77, that they are probably print control issues rather than fade issues.

The Technicolor IB process introduces color changes independent of quality control. IB Technicolor prints tend to have greater contrast, saturation, and are shifted toward green. They may be fade-free, but this doesn’t mean that they’re truly accurate to their source.

Well, the early Technicolor process had many color issues. They refined it as time went on to the point where next to 70mm, a Technicolor print was probably the highest quality image that audiences could see. From what we have seen from the Technicolor scans of Star Wars, there is some green tint, and DrDre has noticed a additional red tint, but in terms of contrast and saturation, the colors seem to very much match our other sources.

AntcuFaalb said:

yotsuya said:
From what we have seen from the Technicolor scans of Star Wars, there is some green tint, and DrDre has noticed a additional red tint, but in terms of contrast and saturation, the colors seem to very much match our other sources.

The two 35mm IB Technicolor scans poita has had done for our team have some considerable differences from our '80s Eastman dupe LPP scan, even when each is corrected to their respective print. As noted earlier, the techs tend to be more contrasty and are more saturated.

Well, the clips and images I’ve seen don’t really show that. I’m also comparing them to what I’ve been doing to restore the colors to both the GOUT and BR as well as reference photos. I see the color difference, but not the contrast or saturation. And from what I’ve seen, the other prints have some pretty serious issues that arise from the photo duplication process. You may be quite correct, but I haven’t seen it.

Here are a set of automated corrections of the early scenes of reel 2 of the SSE Grindhouse. The previous corrections based on the Grindhouse were faulty (bugs in the code). These corrections were done using several scenes, and therefore should be more representative of the original theatrical color timing, although, as for the first reel, the color balance of the raw scan shifts quite a bit from scene to scene or even shot to shot: