Spaced Ranger

There's nothing like watching a rock-solid video after having viewed it jittering about! A major improvement (that everyone takes for granted).  :)

After a comparative review (the SOTS-R3-LUT to the SOTS-R3-LUT-PFC-all), I would recommend doing the clean-up in discreet steps, one at a time, with a full render after each step for critical inspection. This will confirm that each process is doing it's job completely, without introducing additional artifacts.

As a first step, to get started on the right track, re-render the uncorrected clip and confirm the rendering is sufficient adjusted reproduce the source identically. If you can't, look for another renderer that can do the job. I bring this up because, when checking out PFClean's "dust removal" (a spacial repair, pictured here), I noticed what looked like macro-blocking in nearby areas. It could be from the rendering ... or it could be from the clean-up ... or from one provoking the other.

In this magnified section, the top is the LUT-capture and the bottom is post-PFClean.

For large debris, like that yellow blob on this single frame, I would overlay a matching replacement area from the previous or next frame (a temporal repair). There are probably many ways to do this same thing in PFClean or maybe they hide all that from you. If you have no control, fixes like this partial clean, with additional smearing, just trades one bit of garbage for another ... needlessly. If so, you should consider using outside programs (there are Avisynth plug-ins that do this sort of this automatically) where PFClean fails to be the perfect solution.

Also, do not necessarily assume that each category of fix must be a one-shot application. Breaking it down to multiple, sub-category fixes might do a better job.

I previously emphasize the word "automatically" because the only "manual" clean-up should be in setting parameters for the program to do the work (otherwise this may end up being your life's work).

Regarding workflow, I've edited and re-ordered what I think would be a good (non-interfering) sequence of fixes:
¤ capture
¤ luminance stabilize
¤ frame motion stabilize
¤ de-spot (single-frame debris)
¤ de-scratch (predictable multi-frame debris)
¤ de-noise
¤ align (non-squared framing)
¤ color correction (film fade)
¤ sharpen (non-haloing)
¤ mask/resize/frame (HD, DVD)

ww12345 said:

Antcufaalb and I  have been working on cleaning up that Reel 3 scene. I hope to have the clip finished (stabilized, dirt+scratches removed, etc.) by tonight ...

From AntcuFaalb's post, I went looking at PFClean and had a Keanu Reeves "woah" moment.  :)  Then I saw the price and my wallet had a Keanu Reeves "woah" moment.  :O  Great things should come of that (the PFClean, not the woah)!  :D

If Zip Doodah has full access to the 35mm files, perhaps he can run an Avisynth script to sample a single frame inside each shot change. That would produce color correction targets for the entire film. I saw such a script once. I'll see if I can track it down, unless you know or have one.

ww12345 said:

Interesting. I'd like to see what you could do with the scenes in motion, too, if you have that ability.

I use VirtualDub to apply my tests to the video (see my workflow, above) with the plug-in Graduation Curves. It was the only one I found to do the job of using the results of the paint program tests, and with interactivity, too. It was supposed to be usable in Avisynth (my first choice) but it wouldn't work -- just as well, as Avisynth wouldn't be interactive with this plug-in anyway.

This way, I can give you a handful of numbers, which you can use/tweak on your end. If you use other software, I'm sure there must be an easy way to also use those results. Let me know if you're having trouble getting it into your work-stream.

poita said:

Remember these are all compressed 8 bit images, the capture happens in log12bit or 16bit linear, so you have a lot more control when colour correcting. ... by getting the colour range closer to the desired output at the capture phase, and eke as much colour in each channel out as possible.

That is so awesome! And, of course, your results are way beyond the poor, faded source. A so-much-better starting point for this project! BTW, I think all our computer stuff is stuck at 8-bit R-G-B. Any suggestions of what you do to make use of all those extra colors?  :)

poita said:

As for extending the range to the full width, one has to be careful doing that on any given scene. ... You want to stretch the range, but not too far, otherwise it all ends up flat and lifeless with no dynamics...

Yes, that was my bad. On a still frame, I push adjustments to the still's max. Ideally, a run of the entire film would be made to get it's min-max value, and then a global-range-extending capture could be made that wouldn't crush/blowout the picture anywhere.

BTW, as long as you don't mind my make-work suggestions, I'd appreciate your continued knowledgeable explanations and any references to in-depth information (especially regarding the hardware, it's capabilities, and settings).

I was curious on what values were adjusted for the LUT (Look-Up Table). Not knowing how this is implemented, my hope was that it was analog in nature (dimmer/brighter light source with shorter/longer exposure) and not digital in nature (like a paint program's rounding-error-cascading integer math).

                               FLAT CAPTURE

First was to stretch out each R-G-B spectrum from it's fade-compressed range to a full range (0 thru 255). This I expect to be the proper approach for correcting faded film, which loses it's contrast in becoming ever more transparent. Note: the sample should have a full range of color for greatest accuracy of color correction -- which I ignored for this test.

                      R-G-B NORMALIZED
                  RED      12-166  to  0-255
                  GREEN   4-176  to  0-255
                  BLUE     4-170  to  0-255                                                           LUT CAPTURE

Clearly, normalization wasn't the approach use for the LUT capture. The LUT capture didn't have the white cartoon teeth that normalization produced. A quick search revealed that gamma was the adjustment for look-up tables in graphics. So, next, the range was left untouched in favor of gamma adjustment. This came close but it also required midtone adjustment for a better match:

          R-G-B GAMMA & MIDTONE ADJUSTMENT
                  RED       1.5  &  -15
                  GREEN   1.3  &   -4
                  BLUE      1.2  &   -3                                                                      LUT CAPTURE

Of course, I don't know how you determined what settings to use for the LUT to pull as much color out of the faded film as possible. Was it from a known movie reference or by a general standard of adjustment?

Here's something interesting. Clinging to my film fade contrast approach, what if normalization was a missing factor needed with the gamma & midtone adjustments? Combining the two sets of values produced this:

I'll test this idea on the coach scene with real people. If this works well, the master color correction might be performed in the analog capture (if possible to make such adjustments), with only minor touch-ups made digitally thereafter.

poita said:

Those are looking promising. I decided to develop a new LUT for the capture process to try and get as much colour out of the faded print as possible ...

Excellent!

BTW, do you (still?) have any of those download's in their "no special settings" capture -- for my ongoing "practical research" -- on how the capture is affected? Also, can you go into more detail about the "LUT" settings (or point to documentation describing it for this system)?

After a careful reading of the DVDBeaver comparative review of the 2001: A Space Odyssey video releases http://www.dvdbeaver.com/film/DVDReview/2001.htm, I looked into their claim that "The first unrestored release from Warner (not shown here) is identical to the MGM edition in every way except packaging" (earlier in this thread). This claim is wrong. It is the MGM 1998 DVD only that has the animated, anamorphic menus. The Warner 1999 DVD (mentioned but not shown) has those same animated menus, but cropped to fullscreen:
                                      MGM 1998 DVD                                                WARNER 1999 (and 2001) DVD

This is also true of the later Warner 2001 DVD menus (now as a still frame -- not animated), regarding which they have correctly stated. Just FYI for anyone doing a project to include the best from all the releases.

SilverWook said:

I haven't watched either disc since before I got a widescreen tv, so any aspect ratio difference in the menus would not have been seen by me.

It's the player sending the image to the TV that determines how it's displayed. In your settings, you can force your player to behave as if the TV is widescreen. Of course, if your TV isn't the 16x9 ratio, you will see the full anamorphic picture, but still squeezed (whereas a widescreen TV would fit it across it's widescreen, effectively unsqueezed).

With the ideas of a proper Smilebox® (but a "proper" one would no longer be Smilebox®), and stereoscopic 3D viewing, and an "in theatre" experience, I looked around for a good, and cheap (free?), and easy to use, and continuously interactive, and powerful 3D program. I know, I'm an optimist. Fortunately, optimism pays off and I came across the once awesome program from Caligari, sold to Microsoft for their Virtual Earth online program, later discontinued by Microsoft when it moved on, and now freeware with the blessing of both the original author and Microsoft. How's that for optimism? Sometimes I amaze even myself.  :D

It'll probably be quite some time before I could get around to this. So anyone with an inclination towards 3D design and rendering might like to give it a shot. Start at the Flat2D website to get the now completely free program with links to more resources (tutorials, plug-ins, and such):

Flat2D - trueSpace 7.61
http://www.flat2d.com/Truespace_en.aspx
trueSpace 7.61 is a powerful 3D modeling and animation [program], with several years of tradition. Its first version was launched in 1994, and the program has been refined over time, slowly becoming one of the most respectable of its area. In 2008, the production company was acquired by Microsoft, and then the software is now distributed free.

Start at United3DArtists forum's "Links that you might need one day using TrueSpace" and trueSpace:

 Major Geeks - trueSpace 7.61
http://www.majorgeeks.com/files/details/truespace.html
trueSpace is a 3D modeling software for creating models, photo-realistic objects, and landscapes.
Work in real-time giving you the best possible modeling environment, and the ability to render an image in seconds rather than hours, or an animation in minutes rather than days.

Just in case there isn't a shift from 16mm to 35mm preservation, I wanted to demo the process using Zip Doodah's 35mm clip (the ride on the coach) as a color target. The wider the color spectrum in the target, the better the overall correction ... and one setting might suffice as a final or master setting.

In a paint program's histogram function, the individual R-G-B spectrums are filled out using low and high. Then gamma shifts the bulk of the spectrum lower or higher as needed. Usually that's enough. If the R-G-B's seem too weak or too strong, then midtones can expand or compress (dilute or concentrate) the center area of the spectrums. This usually requires some back and forth adjustments before the combined coloring falls into place:

histogram   lo    gamma    hi    mid
R                19   1.4         162  25
G                  6   2.5         234  20
B                  7   2.0         187  25

I find this process still can leave the saturation and/or contrast weak, but I haven't quite worked out an easy joined-R-G-B movement. Until then, a second process of a joined-RGB graph is added to compensate:

curves    1 I/O       2 I/O       3 I/O         4 I/O
RGB       0 : 0       64 : 24   128 : 128  255 : 255

The end result is a nice improvement that can be use in multiple shots:

Finally, for video processing, I use VirtualDub with the Gradation Curves plug-in. One must manually duplicate the paint program's graph to attain the same correction (this required resizing the paint program's rectangular graph to make it 1-to-1 to the plug-in's square graph):

I did this quick & sloppy, but even at that, it looks almost spot-on:

It still needs tweaking (pulling out the purple scarf and the green leaves is proving difficult as the graphs must remain smooth in the process). If Zip Doodah could post a 35mm grab (full-sized & raw) of a full color spectrum shot and then ww12345 post the same 16mm frame (again, full-sized & raw), that might just work for an overall color correction.

Well done teaser, Harmy. Even more interesting than George's effort. Can hardly wait to see the full making of featurette!

I did notice a glaring omission (but expecting to see it be in the final release) was the nod to my invaluable contributions. It needn't be grandiose ... "without who's help this project would not have been possible" ... something like that. But more important is getting the name spelled right. Thanks buddy!  ;)

Before DeSpecialized! Before Specialized! At last ... the original Star Wars!

(but too big for the small screen)
http://www.ufunk.net/wp-content/uploads/2010/07/la-guerre-des-etoiles.gif

Zip Doodah said:

What does 'CA' mean?

Chromatic aberration is color fringing caused by photography. Different light frequencies "bend" through the lens at miniscule different angles (coated-lenses attempt to minimize this effect):

As shown in the diagram, the capture-lens focus is thereby imprecise across the R-G-B color slices that compose the complete picture:

Colors slightly out-of-focus are slightly different sized from the one color in-focus. Because the capture lens always is axis centered, simple resizing (which is also axis centered) keeps picture alignment the same. However, for this Song Of The South frame, the axis was at the bottom left corner (!) indicating the frame had been previously misaligned. Enlarging only the Red separation by a pixel or two (height & width) and shifting from the bottom left corner (instead of center-picture) did a good job correcting the problem:

poita said:

I think the issue is ... trying to create a 2D texture map, that would look correct if projected onto an actual Cinerama screen?

Sitting in a good seat at Cinerama, the image doesn't look compressed ...

That's right, and such a projection is correct in both directions.

However, my last (this) approximation is also inaccurate in that the projection lines from the curved screen to the flat screen (see that top-view diagram) are parallel, as if to an infinite focus. The focus should be somewhere in the audience. That ups the level of manual labor. (I was trying shortcuts for quick mock-ups, but each time I'd think about it and come back to throw more work into it. Somebody stop me!)

Of course, the best projection would be to fill it upto the widescreen (16:9) area while maximizing the normal-proportioned center to it's cinemascope (2.35x1) boundary. Or, if putting it in a cinerama theatre setting, do it proportionally smaller to fit.

Are you still thinking of trying a stereoscopic representation of a 3D cinerama screen? With or without the theater?  :)

Asaki said:

Heck, while you're at it, you can turn it into a real game ... Imagine the multiplayer possibilities, as you and your friends debate the best plan of action against the "noobs" behind you, kicking your seat and throwing popcorn ... maybe I've gone too far @_@

Well, maybe just a little.  :D

But we have to give Oldfan, and all of us for that matter, the original experience of 2001. To that end, I was thinking of additional theatre-centric ambience (no popcorn):
 * curtains opening & closing at beginning & end
 * dimming & lighting (the theater) at movie start, intermission, movie exit
 * theater reflection-lightness in sync with movie lightness (?)
 * the slight concert-hall-sized hollow/echo sound

Sorry to make even more work for ... whoever wants to pick up this gauntlet.  :)

... once characterized by someone as the book that ripped off everyone and everything. Now that's what I call good reading and a bargain at only $1.50 (book-cover price)! Just don't tell'em you bought it for the pictures (96 pages worth).  ;)  Too bad the pictures are all B&W.  :(

And too bad that the one really promising source for color-correction-worthy-material had the plug pulled on it:

2001: Beyond the Infinite - The Making of a Masterpiece - Trailer
http://www.youtube.com/watch?v=EspoBBvhKdQ
Douglas Trumbull, director of Silent Running and Special Effects Supervisor for 2001: A Space Odyssey, is readying a documentary about his time with Stanley Kubrick. The detail in this undertaking is amazing. The filmmakers are literally crawling into never-before-seen-images.

mikeaz123 said:

http://imgur.com/a/3TEhj#0

Awesome!

I hope, as things move along, we'll get more work-in-progress samples with technical information on it's processing ... for a more definitive statement on Star Wars original color.

No, don't worry about doing some ultimate preservation. Just look at all the Star Wars "preservations" here over the past decades. There is no such a thing.

Just do the one you started doing -- fixing up a particular source, adding other source extras (maybe new extras, too), and wrapping it up into a Blu-ray extravaganza far better than the present Blu-ray ... and yours would be the first.  ;)

Oh ... you must mean this:

IMDb - THX 1138 (1971) - Trivia
http://www.imdb.com/title/tt0066434/trivia

"A scene in which THX falls into a garbage compactor and fends off a mutated rodent was cut because the monster did not look realistic. This situation was later reused in Star Wars."

Of course, now that George can make "realistic" monsters, expect to see yet another release -- THX 1138: The George Lucas Guaranteed Original Director's Cut.  :P

Zip Doodah said:

Here are some still comparisons from a 35mm IB Technicolor HD transfer of the film.

Nice captures! Those do give superior color targeting to fix up (as far as can be expected) other faded film. BTW, are some caps not coming through? It looks like there are spaces in your post where something more should be.

ww12345 said:

edit: Also, it could just be my eyes, but it looks a little hot, like somebody boosted the saturation. What does everybody else think?

Yes, they are over-saturated. A test reduction in the saturation (-25 in the paint program) brought things down enough to look gloriously more natural:

Back to the cap's original state, closer inspection reveals color fringing characteristic of lens chromatic aberration. (The same thing showed up in the THX 1138 restoration thread capture but was fairly easy to fix.) Also, the R-G-B spectrums show slightly crushed blacks:

Anyway, I was enthused to see if I could quickly match those colors into this project. The first approach, using histogram, wasn't going too quick. Then I tried the second approach, using curves, to map RGB low/middle/high neutral points from the new still to the original capture. That went quick and looked good -- it could use tweaking (or perhaps additional specific-color points), but it's almost there and an improvement over the previous unguided corrections:

poita said:

There are plenty of 2001 film prints available as a colour reference.

Unless we could get the vaulted master separations (B&W), we would be stuck with (any-kind-of)-fade color stock. From what I've read, only Criterion's 1989 "special edition" laserdisc was color corrected to director Stanley Kubrick's personal specifications, http://www.lddb.com/laserdisc/00997/CC1160L/2001:-A-Space-Odyssey:-Special-Edition-%281968%29
which, by that release, should be the only other permanently non-fading record of the movie.

I would be interested to know whether that particular edition was released, intact, as one of the successive DVD releases.

Oldfan said:

I think considering that 3D is doing somewhat well out there, ... this was a missed opportunity by the studio to use 3D to create the true curved feel of the Cinerama format. They could have mapped out the "smilebox" type of screen, but in 3D, requiring the viewer to use their 3D glasses.

That is an excellent idea, Oldfan! Long Live 3D!
But let's not stop there. I say, as the Cinerama screen of necessity reduces the movie's size, fill in all that new black space with (for example) ... the Vinterpalatset Cinerama (Stockholm) or some other opulent cinema of your choice, viewing from the seat of your choice:

[16x9 sized example would fill your widescreen TV]

Well ...

... when I hear arguments of "original intent", it's mostly from self-appointed champions and not the authors themselves. Except Ray Harryhausen ... and he actually wanted his old black & white movies colorized (for example).

On the other hand, Smilebox, whether produced after intense study or on a whim, is a poor simulation of the curved screen Cinerama (for the reasons previously stated). Not that I'm a specialist ... I simply can believe my eyes.

This demonstration should've been done in a 3D rendering program but I didn't have one set up. Using a paint program took a little longer than expected but it was only some high school math and allot of pixel counting.

In 3D space, the flat picture to be Cinerama-ized is wrapped to the screen's curve. This means that the viewer watching from a distance sees the center of the picture in it's proper proportion but the edges as ever more thinner, as in this top-view diagram:

So, with 2001: A Space Odyssey, for example, we start with the image's full frame, which includes top and bottom space due to it's cinemascope size within the widescreen TV frame:

The picture is then cut into equal slices. This follows the diagram's rough (for this demonstration) 10° spacing, except for the leftover 6° that I placed in the center of the picture for symmetry sake (making it easier for me). The gaps are for ease of viewing what's going on with the slices:

The slices are horizontally re-sized for their proper widths when seen in perspective. Notice from the diagram that full sized 3D slices show much thinner in narrow angles. That makes for an overall narrower picture:

Now, to remove those slice gaps ... ah, that's better:

That is the real view ... horizontally. Next, as this is a real world representation, there is perspective in the vertical direction as well. As the screen curves towards the viewer, both top and bottom show that perspective:

Each slice is then vertically re-sized to it's screen length:

This looks crude, of course. In an automated production, the finish result would be perfectly smooth as each slice would've been processed on the pixel level. As for me, it's shortcuts when I'm doing it by hand. Fortunately the paint program has a flat-perspective-angling tool, which I applied to each slice in the direction of the curve (with the bottom mirroring the top):

There, that did it quite well for this rough rendering. And here is how it would look on your widescreen TV:

Finally, the comparison of the flat cinemascope format to the same picture curved in 3D space for Cinerama. :

Doctor M said:

Edit: I've always liked the way the fft3dfilter/fft3dgpu filters work for denoising.  Very tweakable.

I've seen this before but never looked into it. I generally keep away from smearing filters (or use one as a last resort). But this has a name that indicates targeting specific anomalies. Interesting.

poita said:

If you scan in mono you get RGB separations that have more resolution per channel. ... It just means that in theory you get more colour detail if you scan each channel separately in mono, ...

Just a personal curiosity about your system:
With film fade taking a toll more on one color (usually Red) than on the others (which are also affected), can you independently set each channel's capture lows and highs (and gamma?) to fill out each of their spectrums (0-255)?

The last 10% that I previously mentioned was to get the blue shirt colors the same. It came only close (the remaining 5%). Perhaps they were unevenly playing around with the color? So I capped an earlier source (videotape). It was pan & scan but, fortunately, this scene had both parts of the shot (at different times) that could be stitched together. I also re-sized it to better approximate your capture:

If you have access, you might consider capturing or getting sample screen grabs from the laserdisc for a better representation of what the color once looked like. That may be the best guide to "fixing" the Blu-ray.

I wanted to determine, with de-noising the picture as an example, whether working in RGB (if you're considering it) is worthwhile ... practically speaking. I wouldn't to steer you into more of work if the lesser is just fine.  :)

MVDeNoise() is a temporal de-noising script for Avisynth that overlaps a specified number of frames to make the "noise", which is random by any practical measure, cancel itself out. To take into account any picture movement, to minimize temporal smearing, it includes picture tracking. Pretty cool.

Because of HD picture size, I worked up a template to visualize cropped strips of the picture that would be displayed at full pixel size, including intermediate processing and final result -- 5 sections: "raw", "R", "G", "B", & "processed" -- all within a DVD-sized picture:

The first run is the raw image, it's RGB components (to show where the degree of noise resides), and the processed image. For this, MVDeNoise() was applied to the raw image only for the processed result:

The second run is the raw image, it's RGB components (MVDeNoise() was individually applied only to these channels), and the processed result:

From the first set, it is obvious the noise is greatest in Blue and least in Red. Therefore, in the second set, I had MVDeNoise() process +/- 1 frames in Red, +/-2 frames in Green, and +/-4 frames in Blue. This way, the settings produced approximately an equally de-noised result in each of the channels.

Comparing the two approaches, at this point in the restoration, doesn't show allot of difference ... at least in a still picture. Of course, the "de-noiser" in your software may show a greater difference depending on it's type of de-noising and how it implements it. Just FYI.

The original was shrunken to the posting size, but it was easy enough to again grab your posted picture and plug the numbers into a paint program's histogram function (which I use for quick tests):

You may have some other program or plug-in to apply this to your video, but I've used the VirtualDub plug-in Gradation Curves and it's just as interactive as a paint program, with improved features. It's supposed to be Avisynth compatible but I couldn't get that to work (Avisynth is not interactive anyway and interactivity may not show there). See my post in the THX 1138 preservations thread http://originaltrilogy.com/forum/topic.cfm/THX-1138-preservations-Italian-Cut-available-see-1st-post/post/589761/#TopicPost589761 for more information and examples.

The webpage link is now in my previous post -- it is the Hirschler page.

Too bad ... a nice sampling of video sources. That screen-cap looks good though.

I came across someone's Song Of The South DVD project, coincidentally entitled ...
Song of the South DVD Project - Converting and preserving classic animation
http://atlas.kennesaw.edu/~dhirschl/song_of_the_south/

Lots of pictures and lots of detail (now that's my kind of webpage). Of particular note is the British TV broadcast entry:

This might be a useful reference source ... including his finished (?) project.