Super8 signal processing

[sciolist]

Until I have my film transport component completed I'm only working with a handful of test frames.

I'm not sure this - http://www.jkcamera.com/digital_printer.htm- would suit your needs, but since it already exists it might be worth a look.

[carllooper]

Some very good points have been raised.

Regarding rig dynamics the one I'm using is an example. Indeed I had a look at the enlarger lens, as used by peaceman, and that sounds like a much better idea, since those lenses are specifically designed for exactly this sort of thing (abstracly speaking). So I'll be getting one of those for a production run.

Camera stress.

Regarding camera stress. I hadn't thought about that. The exposure time on my rig, per frame, is currently about 20 seconds. In a production run the capture is (or will be) computer syncronised with the film advance motor (a stepper motor). The exposure time can be brought down by using a brighter light source - but how far down may need to be considered in terms of camera stress.

A twenty second exposure, per frame, in a production run is not necessarily a problem. In computer animation projects on which I work, a single frame render takes anywhere between 20 seconds and two hours per frame. And Super8 frame acquisition I'm treating with the same respect.

Light Source.

I haven't yet tested colour film myself but peaceman has, and reports some clipping happening there. The "white" LEDs must have photons that are failing to punch through the denser areas of a particular colour dye range in the emulsion. A solution here may be to use an array of red green and blue LEDs rather than, (or in addition to) "white" LEDs.

Some questions
1. Do you foresee a Windows binary or Linux CLI interface regarding your efforts?

The result will be a standalone freely available windows binary but, at this stage, with a strictly non-commercial use license. Commercial applications are not out of the question but will require commercial agreements be made between all involved - including those programers with whom I'm collaborating directly or indirectly.

2. A media file 101 question. Is an AVI file "just" a container/wrapper for the embedded codec? i.e. could it "wrap" a RAW file? Would you want to?

An Avi file is a container of sorts. I don't have the specs for Canon's RAW file format. I believe it's closed source. I found some reverse engineered specs on the web. I haven't tested these. I'm assuming when I get around to the SDK there will be a function for parsing individual pixel data.

I'm currently just using the available camera software for saving the RAW file as a TIFF. I believe a RAW file is no more than a modified TIFF file.

I assume it's quite possible to store RAW data in an avi. I can't see why not. Would need to write a codec. Could be something of immediate value since it would allow the RAW data to pass through avisynth and virtual dub, etc.

3. What is the proposed native file system? Or will you architect options?
The end result of the algorithms will have a number of options including the usual image sequence formats. The data has to go off to the lab for printing to 35mm (in my case) so the first versions will just be whatever my lab prefers.

[carllooper]

I'm not sure this - http://www.jkcamera.com/digital_printer.htm- would suit your needs, but since it already exists it might be worth a look.

I've lined up labs here in Australia to do some 2K transfers. The only reason I'm building a rig is that I need more than a 2K scan of the film. In addition the available 2K scans use a 16:9 aspect which means either some of the Super8 must end up cropped or a full frame scan would not utilise the full 2K width.

The problem for the humble Super8 is that it is not treated in the same way as the larger formats. There are understandable commercial reaons for that. For the time being a DIY approach is the alternative.

As an example of commercial reasons, I can't get a good 2K scan done until after chistmas because the transfer facilitys are too busy during that period doing transfers of home movies to cope with specialist requirements. But I'm not in any hurry to meet any particular deadline such as Christmsas. And i'm happy to wait till downtime where I can talk to the transfer facilitys about, for example, doing a two pass scan. And to get technical details on the precise nature of the scan.

One problem with a commercial scan is if any sort of compression/digital processing is occuring during the scan. This can play havoc with the intended purpose of the scan - which is to obtain a raw unprocessed signal with which to experiment.

[aj]

Camera stress.

Regarding camera stress. I hadn't thought about that. The exposure time on my rig, per frame, is currently about 20 seconds. In a production run the capture is (or will be) computer syncronised with the film advance motor (a stepper motor). The exposure time can be brought down by using a brighter light source - but how far down may need to be considered in terms of camera stress.

A twenty second exposure, per frame, in a production run is not necessarily a problem. In computer animation projects on which I work, a single frame render takes anywhere between 20 seconds and two hours per frame. And Super8 frame acquisition I'm treating with the same respect.

20 real seconds exposure??? On a digital CCD sensor? Will there be any image at all? I thought/would expect that noise would drown the image and certainly the contrast will be gone.

20 seconds as a step in workflow may be workable but exposing is certainly not the same as rendering (if you mean CGI)

[sciolist]

I'm not sure this - http://www.jkcamera.com/digital_printer.htm- would suit your needs, but since it already exists it might be worth a look.

The only reason I'm building a rig is that I need more than a 2K scan of the film.

The link I provided is to a website showing an optical printer that will perform most if not all of the functions of the device you're planning to construct as you've described it and as I understand it. It's not a website offering telecine service (as your reply seems to suggest). These J-K optical printers can be gotten second hand, they accept pin registered interchangeable printer modules (regular 8mm, Super 8, 16mm, Super 16, and 35mm), and they can accommodate a variety of cameras for capturing an image.

[carllooper]

Yes - I understood it wasn't a service as such. It is just that the printer which had an S8 gate, was for HD scanning (ie. 2K). And it was just that I could get a 2K at a local service.

And the one set up for the EOS didn't have an S8 gate.

But revisting the link I realise that I've jumped to hasty conclusions. The printer has interchangeable configurations, ie. the images and descriptions shown are just examples.

So will definitely look into that.

cheers
Carl

[carllooper]

20 real seconds exposure??? On a digital CCD sensor? Will there be any image at all? I thought/would expect that noise would drown the image and certainly the contrast will be gone.

20 seconds as a step in workflow may be workable but exposing is certainly not the same as rendering (if you mean CGI)

The image I'm getting is just fine. The noise is no different from a fast capture and the contrast is no different. I'm not sure I understand why it would be otherwise?

The comparison with rendering was simply in terms of the time you are willing to sacrifice for a particular result. At this point in the research and dev stage I'm quite happy to sacrifice 20 seconds per frame for the result. Just as I do with the computer animation work.

But for a production run, as mentioned, I'll be bringing that down by improving the brightness of the light source.

[sciolist]

. . . will definitely look into that.

I have an earlier model of the K-107 optical printer (the K-104). It's functionally about the same, so if you've any questions about it, I'd be happy to try to answer.

[peaceman]

Carl,

three things form my end:

a) a minor clarification: I do not think that color clipping occurs due to the use of LEDs. Actually, it also happens with full spectrum lamps. The color space of some stocks (Velvia 50D might be such a candidate too) just does not fit into the CMOS's gamut. To not lose any information through clipping, one would need to shoot each frame three times with red/green/blue filters (or LEDs) and do some "chroma DRI" afterwards. This includes transitioning into the destination colorspace, unless you want to lose further information during "printing" on 35mm. In short, you need full color management.

b) The problem with LEDs is their spiky spectrum. Watch a white LED through a CD and you will get some idea of its non-linear spectrogram. This causes color to appear different (to sensors) than they are. Have you ever shot with your EOS under energy saving bulbs? Looks a lot worse and more tinted than the eye perceived it, right? That is that same effect.

c) On long exposures, you activate the camera's noise reduction algorithms and thus mess up your signal, even in RAW. Consider that your CMOS (Canon does not use CCDs!) does not work like film. Actually, the camera merges many individual shots during long exposures - kind of pre-SR
It should not be a problem to get your setup down to 1/30s or less. Why not using an old flash? That has a good spectrum, perfect color temperature to achieve highest dynamics from your CMOS and is bright and even.

Some more thoughts: I am pretty sure that not even RAW files are unprocessed. Check your camera's "custom functions" and profiles to deactivate denoising, "highlight priority" and sharpening. I am not sure if the 1000D allows all this. You might need to read the CR2 files natively to bypass any post-processing (while I think the de-bayering still happens inside the camera... haven't verified that though). Check out http://www.cybercom.net/~dcoffin/dcraw/ too -- RAWs are definitely totally different animals than TIFFs.

Also, consider your EOS' ISO-Settings. It is not the case that the lowest ISO has the least noise, and most EOS also have way more noise on intermediate steps (like 640) than on full stop steps (like 400 or 800). You might be able to Google that, if not, try it out. My 5Dmkii surprisingly has highest dynamics and lowest noise at ISO 400.

Oh, and one more thing: Activate mirror lock if the 1000D allows that, especially with exposure times like 1/5 to 1/30s. The mirror flap might make your camera shake otherwise during exposure.

[carllooper]

Good points Peaceman

Colour reproduction

Yes the three exposure/filters are the alternative to three LEDs. Basically the "white" LED doesn't emit as evenly broad a spectrum of photons as one might like. For that part of the spectrum in which the output of the LED is low in photons one would need a longer exposure - but then that can blow out (clip) the results from the other parts of the spectrum( for which there are more photons). Or vice versa - with a shorter exposure, that part of the spectrum with a low number of photons fails to express details in the denser regions of the film.

Noise Reduction.

Yes I actually have this turned off now, as it spatially blurrs the signal. I discovered that denoiser a while back. It's turned on by default. But the denoiser works against the SR algorithms.

Flash
Yep - good idea. Will incorporate that into the mounting plate.

Mirror lock.
Another good idea. With the 20 second exposure that isn't a problem but as the exposure time is reduced the potential effect of the mirror (camera vibration) would increase. However it's a pretty rock solid rig I have. But turning off the mirror will mean less stress on the camera anyway.

[carllooper]

Re. Canon RAW files.

I heard about the RAW being a version of TIFF here:

http://www.sno.phy.queensu.ca/~phil/exi ... n_raw.html

Which takes you here:

http://lclevy.free.fr/cr2/

which takes you to dcraw:

http://www.cybercom.net/~dcoffin/dcraw/

[miles&jules]

Hi Guys this is an interesting topic.

We built one of these for our nikon d70 and now our d90. But we shoot off the back of a working projector.(50watt halogen shooting from where the projector lens once was through a white opaque piece plastic (white box). We use a bellows with a reversed 35mm Nikor lens. This does a perfect frame of standard or super8.

The big problem (which has already been mentioned) is you will only get an hour of footage out of the slr. Our nikon d70 shot 130,000 (Yes we have them all on our hard drives) before it died. This wasn't done transferring super8 but shooting stop-motion animation.

So I cant see this slr technique being a pipeline for a business.

We use ours for a particular film we are making in which a segment of standard 8 has to cut with a animation that is 3-4k digital. You do get some amazing images and the raw control is awesome for neg. Keep up the experimentation! I like the sound of the digital processing!
Miles

[miles&jules]

Oh forgot to mention . We have the projector being run by a 12volt cordless drill motor. This is controlled by a adjustable speed controller($15 from china) The microswitch triggers either the mouse to capture the raw with nikon capture. Or it controls the d70s wired remote. to is cf card. Its does it at about 1 frame every 10 seconds. But ccd heat up so I only ever do it for a hour at a time then let the sensor cool. Otherwise hot pixels. I don't think this happens on cmos of the d90. Haven't done much experimenting with it yet.
Miles

[peaceman]

We use a bellows with a reversed 35mm Nikor lens

This is good, but most likely not as good as an enlarger lens -- since your regular Nikor is calculated "for infinity", not near field. Even reversed, the focal area is probably not totally flat and distortion also higher than with an enlarger lens.

On your dying D70, did you let the mirror move too or did that stay locked up and you just used the shutter? (Not sure if the D70 allows that...)

[VideoFred]

Check on the works of member videofred http://www.super-8.be before reinventing the wheel. It has all be done.

@aj: Not "all" has been done yet, not even by Freddy

But both Frank and Christoph are working with better machine vision camera's then mine

Frank has even made custom acquire software:
http://www.cine2digits.co.uk/

And with his RGB light source it will be possible to make multiple RGB capturings and even several capturings with different shutter speed settings. This can only be done with a machine vision camera with trigger.

The biggest weakness I see in avisynth would thus be that it only handles 8 bit per color channel, at least as far as I know. The color dynamics of e.g. Ektachrome 100D are often challenging CMOS sensors, even the 5Dmkii sometimes runs into clipping with that stock (and its dynamic range is enormous).

Yes, the 8 bit limit is an Avisynth weakness. But in 90% of the films I have transfered (and believe me, I have transfered a LOT of films by now) there is no "magical " dynamic range.

The dynamic range of film is very easy to check (while capturing, with my machine vision camera) by blowing out the whites in preview and then see what becomes visible in the darker parts of the film. In 90% of the cases I only see darkness.. In 10% of the cases I see indeed the limits of my 8 bits machine vision camera. This can be fixed with a double or even triple exposure of every film frame. And then a smart software merging of the frames.

Fred.