Super8 100D - Latest SuperResolution Tests

[carllooper]

Just got back a scan of Super8 100D, done at 2.5K.

Testing super resolution on such. This test is bit of a cheat because the SR image isn't the result of any optical flow calculation. I stabilised what was otherwise a tripod shot and then merged 64 frames together. A deconvolution aperture was used to recover some focus lost in the transfer.

1920 wide stills (downscaled from 2560 wide) of the test can be seen here:

1. An original frame
2. Super Resolution + Deconvolution

Carl

[super8man]

Reminds me of the resolution I got when doing an 8K scan of this strip:

[luuude]

What is your setup carl? VERY nice, crips images you got there.

Best regards

Ludvig

[luuude]

super8man your webpage link seems to be down? That strip looks great! Neg scanner?

Best Regards

Ludvig

[Nicholas Kovats]

Carl,

Thanks for posting your results. But I find that number 2 doesn't "look" like film. It has evolved into something else. Whereas with number 1 I have an immediate instinctive reaction. Thats film.

But it's a bit of a quandry. I like how no. 2 has removed the "dirt speckles or clumps". As I personally endeavor to push the film medium my biggest complaint with it's "traditions/roots" ...is the classic magnified clumps of hair, dirt, dust, etc in camera gates, processing labs and projection. Especially Super 8.

I am in contact with a gentleman in the States who is selling off individual stand alone wet or dry gates for R8, S8 and 16mm but he wants a small fortune. If a better and more cost effective design could be drawn up and popularized there would be less need for remove such artifacts digitally.

I guess my point being I am starting to head in the direction of trying to provide as close to an originating "pure" analog signal for post-processing as possible.

Hence my attraction to the removable (for cleaning), the precision and the interchangeable aspects of the Bolex camera system as opposed to the closed system of Super 8 camera design. In fact the Bolex camera engineering with it's by design interchangeability allows all current implementations of small formats, i.e. R8, DS8, UP8, SD16, S16 and U16.

Excuse the bit of rambling. This is yet another attempt to tie in your work with a greater systems approach. Great work once again.

[super8man]

I fixed it...my link to my website that is...

[etimh]

...I find that number 2 doesn't "look" like film. It has evolved into something else. Whereas with number 1 I have an immediate instinctive reaction. Thats film.

What he said. And what's the deal with the white vehicle in number two? Is that blur an effect of whatever you're doing to the image? It distracts and betrays the manipulation and fiddling that is going on with the image.

I don't know what you guys are talking about or what you're doing when you go on about all this, and I have absolutely no idea what you're actually trying to achieve. Is it to make it look less like film? I don't get it.

I personally like "classic magnified clumps of hair, dirt, dust, etc." 8)

Tim

[luuude]

The "superresolution" is several frames blended together. On each frame the grains of photosensitive material makes a sample and since film is organic and random these samples are at different places at each frame, unlike a digital sensor where its a grid. When many frames are combined you get more samples of the same area in the image and therefore higher resolution. The car looks strange because it was moving so its in different places on all those avarage frames a bit like long exposure photography. I use some algorithms like these mainly on my old familys films it just makes things clearer and my relatives prefer to see what their now dead aunt really looked like rather than the nice grainstructure that I myself quite like.

Best regards

Ludvig

[carllooper]

Yes - there's a bit too much sharpening in that second frame.

I was pumping up the sharpening to see where the lens limit might be. Analysis of the sharpened image showed the lens was resolving at the very least 180 lines/mm. The test indicates the loss in focus may be occuring during the transfer stage rather than in the original camera stage. If so, I may be able to get better results by implementing the deconvolution (sharpening) prior to the SR stage.

The blurred car in the second frame is a function of the fact that the car is moving. Motion Blur. The frame is the equivalant of a 2.5 second long exposure photograph, but created from merging together 64 sequential exposures. For any motion picture use of SR, this will be not any good of course, which is why this particular SR test is bit of a cheat. The complete SR solution involves using optical flow to merge frames, so that one doesn't have things like motion blurred cars.



Here's what the test looks like immediately after the SR has been done, but before any sharpening is done.
SuperResolution8X, No Deconvolution

And here is the same after some less aggressive sharpening is done (plus some colour balancing):
SuperResolution8X, Minor Sharpening

[carllooper]

What is your setup carl? VERY nice, crips images you got there.Ludvig

Thanks Ludvig. The transfer was done in Italy by Roberto Pirodda, using his own frame by frame setup, with a 2.5K 3 chip camera. The frames were saved as TIFFs. The digital processing is done my end using software I've been writing over the last year (when I get a chance). The code is written in C++, using a software development kit called Qt. I also test and use code developed for more scientific applications such as medical imaging (eg. CAT scan visualisation)

I'm a big fan of the realist filmmakers, and their theories (Andre Bazin etc). I'm interested in working within (or pushing) that tradition in the digital age. The realist theory I'm pursuing has it's origin in Ancient Stoic philosophy. Plato's parable of the cave, in which cave dwellers believe the shadows on the wall are real, is one of the oldest critiques of that theory. In Stoic theory (against which Plato is arguing) the shadows on the cave wall are real - a reality in their own right. A fundamental reality beyond which, or outside of which (ie. outside the cave) there is no other reality. Absurd as that might sound there is some really interesting ideas in it. It's the complete opposite of the platonism otherwise informing The Matrix, Dark City, and The Truman Show. In Stoic philosophy there is no other reality, such as aliens, or reality tv show producers, who are otherwise creating the shadows.

The philosopher, Deleuze, in his Cinema books, pushes Stoic philosophy to suggest a world outside the cave that is not much different from the world inside the cave - just a bigger cave. And a bigger image. And a bigger consciousness co-existant with such. Or something like that.

In platonism the world outside the cave is constituted very differently - the outside becomes a template, or formula for what is otherwise inside the cave. The equations of computer generated images can be regarded as platonic in this regard. The fundamental nature of the universe, not as an image, but as a formula for generating the images. The Stoics see it the other way around - the image as a fundamental reality, and the formulas being a secondary function (or echo) of those images.

Now while I'm using formulas and equations, it's in a way that is the opposite of platonic realism. Instead of a mathematical universe which would otherwise produce an image, one takes the images as already a reality in their own right, and use a kind of inside-out maths (statistics) to focus the bigger image. Now in the test I've only reconstructed what Deleuze would call a "photo-de-pose". This is the first type of image that Deleuze identifies in the Cinema books. It corresponds to classical philosophy. The posed image. But here it is reverse engineered from a motion picture film. The next step is to reincorporate that classical image (the posed or reposed) back into the "snapshot", and reconstitute the motion otherwise lost (as motion blur) in the creation of this particular classical image.

Carl

[PyrodsTechnology]

[quote="carllooper"]

The test indicates the loss in focus may be occuring during the transfer stage rather than in the original camera stage.

hello Carl
you can't imagine how much work and time involved in this setup, then i go wrong with focusing ???
You can achieve better sharpness if you switch to a Bolex/Switar Double Super8 camera. You will be amazed how much better it looks vs cartridge S8. About your software: It works great, but to me it is too heavy effect. It is suitable if you want to get high quality still image from a S8 sequence. When in playback, the motion pictures are already averaged. By my experimentation, the best averaging motion pictures must not exceed 2 (two) frames. Everybody can do this simple trick on NLE software like FCP or Premiere :
put in the first timeline track a video (PROGRESSIVE) footage, then in the upper track the same video but SHIFTED of ONE frame. The upper video level must be set at 50%, so the two video tracks are mixed. Ofcourse motion blur occurs if there are fast moving objects or fast panoramic, but this trick can give an idea of what i mean.
http://www.imagehost.it/di/YN9R/timeline-bis.jpg
Best regards.
Roberto

[luuude]

Carl, I always enjoy reading your philosophical rants! I dont have even a retort but please keep them coming! I hope you dont find this bad form Roberto but might I ask you what camera you are using? I am not running a facility and am only scanning for myself and close friends. I have been looking at lots of different cameras from point grey, alliedvisiontec, ids and others. I am looking for a camera that can resolve 8mm and in the future also 16mm and preferably has very high dynamic range. Either via multiple exposures or advanced sensors.

Best Regards

Ludvig

[PyrodsTechnology]

Hello Ludvig
for SD i use a Sony DXC 990P and AJA uncompressed digital converter, for HD (2560X2048) use a SONY DKC ST-5 scientific camera 2/3" 3CCD chip uncompressed TIFF output, 15MB per frame (!). This one i an old (1997) but still very good camera that i found new/old stock in a German reseller for a ridicolous price. The "problem" with this camera is that it stores frames first in its internal memory, then download to hard drive, and this slow down the telecine processing. Sadly there is no way to go directly to hard drive unless to rebuild totally the software that drives the camera.
Rob

[carllooper]

might I ask you what camera you are using?

I assume you mean Roberto's transfer camera, but otherwise the camera used to shoot the film was a Leicina Special Super8 camera, with a Leitz Optivaron telephoto lens, set at wideangle (6mm).

The film stock was E100. Processing was done by Richard Toohey of NanoLabs in Melbourne, Australia.

The street scene was the last shot done on the cartridge. The shot has picked up some dust and hairs as a consequence. There was also some flutter in the camera gate, on one or two frames, which I assume is due to the diminished tension in the feed reel, as the cartridge comes to it's conclusion.

I always enjoy reading your philosophical rants!

cheers. I find I can't get motivated to do anything unless I've got some sort of philosophical force to drive it, or to make it comprehendable, if only tentatively. I'm afflicted with "why bother" and "what's the point" types of questions. So having something (anything) around which to work helps to alleviate such afflictions. And if nothing else provides some copy to go with the images.

The next step is running the transfer through the optical flow algorithms.

Philosophically this relates to what Deleuze calls "liquid perception". Deleuze identifies this type of image in the pre-war French School, (Jean Renoir etc). He asks what is it about water that seems to correspond to the French School? He suggests that " water is the most perfect environment in which movement can be extracted from the the thing moved". In water, for example, one can identify the wave which propagates through water. The water itself moves up and down while the wave moves through the water. In optical flow computation the same sort of thing is being attempted - to extract the abstract movement that otherwise appears to occur between adjacent frames. A liquid perception.

I'll post some more on this soon.

Carl

[carllooper]

you can't imagine how much work and time involved in this setup, then i go wrong with focusing ???

I'm not sure you have gone wrong with focusing. If there was any loss in focus during transfer it was extremely minor, and it wasn't necessarily anything that would have been physically correctable anyway. The lens could be completely in focus and the loss could still occur. Digital sharpening (deconvolution) can fine tune the focus after the fact.

There were losses in the original camera lens as well. Despite the fact that somewhere, in the image, the focus must have been already correct, deconvolution of the camera signal improved this otherwise already 'correct' focus.

The transfer was great - everything I hoped it would be.

cheers
Carl