Is an HD Super 8 transfer overkill?

[Nicholas Kovats]

Looking forward to these HD Super 8 film transfers, Paul.

[cinelicious]

I would hopefully like to end this very long thread with something that will be more convincing that words. May I present the first HD Spirit transfer that we are able to show online "The Journey" featuring Supermodel Elyse Taylor:

http://www.vimeo.com/14317782

Enjoy.

-Paul

[Nicholas Kovats]

Congrats to both you and your team, Paul!

Excellent work.

I suspect this is either Vison 2 o3 200T S8 negative and I hope your investment recoups faster than later re: S8 gate.

Do you have a theory why S8 labs cannot or will not control electrostatic buildup on the processed S8 frame? I can still see a few dust particles and hair(?). In Canada we only have one S8 negative lab here in Toronto and both the head and tail of my negative reels consistently exhibit this issue. But more pronounced.

[Nicholas Kovats]

Correction. I meant to type "...Vison 2 or 3 200T S8 negative..."

[cinelicious]

Congrats to both you and your team, Paul!
Do you have a theory why S8 labs cannot or will not control electrostatic buildup on the processed S8 frame? I can still see a few dust particles and hair(?). In Canada we only have one S8 negative lab here in Toronto and both the head and tail of my negative reels consistently exhibit this issue. But more pronounced.

Film loves dust. And since Super 8mm is small gauge dust particle size is magnified. We do have a DVNR 2K with the ASC-3 ME dirt/dust algorithms than can correct for this (about 95%) on the fly but many Super 8mm filmmakers don't want us to use it as they like the organic imperfections. We can also do frame-by-frame DRS / restoration work that would clean up an image pefectly. Costly though.

[Nicholas Kovats]

Excuse my technical ignorance on these matters but what are "DVNR 2K with the ASC-3 ME dirt/dust algorithms"?

Final Cut Pro based apps? Or expensive stand alone proprietary solutions?

[cinelicious]

The DVNR 2K is a top-of-the-line realtime processing hardware cards made by Digital Vision that sit in a "DVNR" hardware box used that is often utilized in film scanning and telecine. There's a picture of it on our site in the "rad video" clip that you can watch from our home page or you can learn more about it on Digital Vision's website. It's not "proprietary" because if you have $125K you can go buy one, but yes it is expensive.

That said the creative handles it gives you for fine tuning a film image in realtime are worth every penny. These types of algorithms would take 20:1 processing time in a software-based solution. Our DVNR 2K has the AGR-4ME grain reduction module (six, hardware video processing cards, as well as ASC3-ME... another 6 cards) and is one of most fully-loaded you can buy. Most DVNRs out there only go up to HD frame size and have grain-module only.

Hope that helps explain things.

-Paul

[Tscan]

I recently started doing full HD captures with the Workprinter. An HD capture is really nice, but most of all I can still downconvert to SD and my image is cleaner than my old standard SD captures. With an HD capture, you can have the best of both. Sometimes I output to SD, and others HD. Right now, my only output in HD is for Vimeo... I'm still burning DVD's in SD because the only missing link on my HD transition is a Bluray Player. I have Bluray burner, HD camera, HD PC capture, HDTV, HDMI Digital Projector... But once I finally get that last $200 component (maybe for X-mas) my scans will be up to date. But overall, YES, i like my HD and SD downconverts better than plain SD scans-

[mattias]

ok, so i'm later than late, but while this is a *very* nice looking clip i think it only proves that the spirit is a wonderful machine, and actually rather suggests that hd isn't worth it. i took the hd clip (compressed from vimeo but it doesn't matter much for sharpness), downconverted it to pal, then back up again using a naive bicubic scheme. there was definitely some grain lost in the process, but *very little if any* detail.

i like the idea that hd can capture more defined grain, especially for a film out, and if you want to watch it in hd it's obviously better to have it hd to begin with, but i've been convinced for a long time that there's very little more than pal resolution in a super 8 frame (i did the math back in the 90's using kodak's lab test specs, at that time to prove that s8 was as good as or better than dv), and even more so after this. the quality of the telecine equipmen is way more important than the resolution for the sharpness and look, both perceived and technical.

speaking of technical, can you post a clip done with technical grade so we can try correcting it ourselves? probably not, but at least i asked. ;-)

[carllooper]

I don't think it's overkill.

Ignoring the definition of Super8 film itself, the very best lenses have a native resolution limit of around 140 line pairs/mm = 280 pixels/mm. The actual resolution could extend as far as 200 line pairs/mm, (400 pixels/mm) but details into this range become harder to see due to the drop in contrast. One can still notice them due to inter-frame correlations that the brain can pick up on.

Now assuming the Super8 frame to be:

Code: Select all

5.69 mm x 4.22 mm 

Source: http://www.super8data.com/database/arti ... ndards.htm

then the best signal possible (ignoring the film) would be:

Code: Select all

1593 x 1182 pixels     (lens = 140 lp/mm)  
2276 x 1688 pixels     (lens = 200 lp/mm)  

Filmstocks, however. have a native definition somewhat less than this, not because of the particle sizes per se (which are very small) but because of the grain (which concerns how the particles are distributed/clumped).

The equations for calculating the joint resolution (film + lens) vary a little but they share the same principle. This one can be used a rule of thumb. It is used by Kodak:

Code: Select all

1/R ^ 2 (system) = 1/r ^ 2 (film) + 1/r ^ 2 (lens)

So lets assume a 140 lp/mm lens and a 100 lp/mm film then the joint resolution becomes:

Code: Select all

R = 81 line pairs/mm = 163 pixels/mm
= 774 x 704 pixels

But like lenses, filmstocks also encode information beyond their MTF resolution rating. The rating is done at 30% contrast, which is an arbitrary cut-off point. It's not as if you suddenly stop seeing details at 29% contrast.

Assuming one can see, or otherwise extract this information below the cutoff, whether subconsciously or by using super-resolution processing, one could start speaking in terms of a higher resolution, eg. 200 line pairs/mm (400 pixels/mm).

Computing for lens and film at 200 line pairs each gives:

Code: Select all

141 lp/mm = 282 pixels/mm
= 1605 x 1190 pixels

Carl Looper

[mattias]

yes, you're right. but very few lenses used on super 8 resolve 200 lines, i'd say it's closer to 50 most of the time, and lens transfer functions are just as fixed as pixels, a photon that hits the lens in the same certain spot will for all practical purposes hit the emulsion in the exact same spot. there's no randomness over time that integrates to any higher res.

the other thing is that the ccd capturing the image also will pick up the "randomness" of the emulsion through varying density of the pixel and its surrounding pixels and watching video in motion has the same effect on the brain as watching film in motion.

the third thing, which was my main point, is that scaling down and then back up doesn't increase sharpness in this clip. however, i've since read in a post by awand that the spirit doesn't use the full 2k with its super 8 gate, so that could also be another reason for that.

[carllooper]

yes, you're right. but very few lenses used on super 8 resolve 200 lines, i'd say it's closer to 50 most of the time, and lens transfer functions are just as fixed as pixels, a photon that hits the lens in the same certain spot will for all practical purposes hit the emulsion in the exact same spot. there's no randomness over time that integrates to any higher res.

Yes - the lens can't be super-resolved in the same way that the film can. It represents a practical limit (although I've read some interesting work on lens super-resolution). But for Super8 cameras with an interchangeable lens arrangement, you don't need to be limited by "Super 8 lenses".

But as mentioned, even where lenses are rated at a certain resolution it's an arbitrary cutoff point. The resolution increases beyond that rating but the drop in contrast decreases it's visibility. You can still see into that region due to inter-frame correlations the brain performs (which increase the contrast)

The contrast is also dependent on the film, (not the lens) for it's visibility, so by increasing the film's resolution, you are able to extend the visibility of the resolution (beyond the rated cut off) into the low contrast (unrated) range of the lens.

he other thing is that the ccd capturing the image also will pick up the "randomness" of the emulsion through varying density of the pixel and its surrounding pixels and watching video in motion has the same effect on the brain as watching film in motion.

This is not quite true. The grain of film is not a variation in density as much as a variation in the distribution of otherwise very small particles. A CCD, or similar digital sensor, has a fixed size pixel, that varys in intensity as function of signal noise.

But the noise in film, while similar to sensor noise at the scale of a sensor pixel, is very dissimilar at the scale of individual particles. One can increase the definition of a film scan by scanning the original film at a higher definition and performing temporal integration etc, but one can't achieve anywhere near the same thing doing that with sensor data.

the third thing, which was my main point, is that scaling down and then back up doesn't increase sharpness in this clip.

Very true.

however, i've since read in a post by awand that the spirit doesn't use the full 2k with its super 8 gate, so that could also be another reason for that.

[mattias]

This is not quite true. The grain of film is not a variation in density as much as a variation in the distribution of otherwise very small particles.

...which on digital registers as a varying density. let's say a single black grain is the size of half a pixel. if it happens to register smack in the middle of a pixel the sampled signal will be 0.5. now if the next frame the grain (even though the subject hasn't moved) is smaller and a bit further to the side, that means that the pixel will be sampled as maybe 0.3 and the neighboring pixel which was previously 0 now at 0.1. this varying density when the video is played back will integrate to a higher perceived level of detail just as with film.

[carllooper]

This is not quite true. The grain of film is not a variation in density as much as a variation in the distribution of otherwise very small particles.

...which on digital registers as a varying density. let's say a single black grain is the size of half a pixel. if it happens to register smack in the middle of a pixel the sampled signal will be 0.5. now if the next frame the grain (even though the subject hasn't moved) is smaller and a bit further to the side, that means that the pixel will be sampled as maybe 0.3 and the neighboring pixel which was previously 0 now at 0.1. this varying density when the video is played back will integrate to a higher perceived level of detail just as with film.

Hmmm. I think we're talking about slightly different things. From what I'm now understanding of what you are saying, the variation in intensity, of which you speak, was not in relation to sensor noise, but in relation to the transfer of the film noise to digital.

But what you are saying is still not quite correct - there is a perceptual integration taking place. But spatially. If you want to improve resolution (level of detail) - then that is precisely what you don't want. Spatial integration can only improve tonal resolution (think an increase in number of bits per pixel). Perhaps that is what you mean by "level of detail". In that case yes- a lower definition scan will improve the signal. But whether that takes place in the pixel or the brain/perception (or both) it's the same thing - it's at the expense of spatial resolution.

Temporal integration is also something that occurs perceptually (or enhanceable algorithmically). You can increase the spatial definition of the film scan, which otherwise degrades the tonal definition (decreases the bits per pixel so to speak - making the image noiser), but allows the brain/perceptual system to pick up the spatially finer details through temporal integration of the noise (rather than spatial integration).

[mattias]

yes i don't mean sensor noise but integration of grain. but no i don't mean that spacial integration improves sharpness, quite the contrary as you say, i mean that spacial integration preserves the temporal integration of film grain, resulting in the same kind of perceived increase in sharpness. I'm not quite sure if the level is the same or even close, but the more we discuss the closer to the truth we're likely to get. ;-)