8mm Scanner

[MovieStuff]

Interesting project. I am certain that scanners will eventually get good enough to do this.. Just some observations: I noticed in the sample movie that there is some fairly obvious pulsing of the image exposure, as if it were shot off of a projector with the shutter out of phase of the video. What is the cause of that pulsing?

The other inherent problem with scanning long strips is that the scanner sees the entire strip as "one image", instead of dealing with each frame individually, and tries to average the exposure for the entire length of the strip, which may have a lighter scene at one end of the strip than the other. I have never been able to reconcile in my mind how this is really workable, since the next strip may very well end darker than where it starts, thereby making the "joint" between the two strips fairly obvious due to sudden changes in density. Is this the cause of the pulsing that is seen in the sample? Or is that pulsing built into the original movie? Just curious.

Roger
http://www.moviestuff.tv

[MovieStuff]

if it says 3000 dpi you should get that, shouldn't you?

You'd think so but I have not found that to really be true at all, especially when you are dealing with really tiny images like slides and the such. It seems that the scanner industry plays games with the term "resolution". I really don't understand it, frankly.

Roger
http://www.moviestuff.tv

[sophocle]

scanner industry plays games with the term "resolution"

I pay about $50 to scan a 4x5 on a drum scanner with 1/3 of the resolution of my epson photo flatbed.

I can assure you the $ is worth it, there is simply no comparison.

[jimfcarroll]

oh, i didn't think it would produce as good images, but if it says 3000 dpi you should get that, shouldn't you?

How about if it says 4800? Then should you get 3200?

I wrote the following about the quality of these cheap scanners a while ago; when I was working with an earlier Epson model:

http://www.jiminger.com/s8/early.html

Jim

[jimfcarroll]

Interesting project. I am certain that scanners will eventually get good enough to do this.. Just some observations: I noticed in the sample movie that there is some fairly obvious pulsing of the image exposure, as if it were shot off of a projector with the shutter out of phase of the video. What is the cause of that pulsing?

GREAT observation. Luckily my parents didn't notice

It's caused because the scanner lighting is not planar and varies along the length of a single scan. The frequency of the pulsing is 14 frames in the original (the length of a single scan - actually, the lighting goes through two complete periods over the length of a single scan so the frequency is more like 7-8 frames). I've was working out the software way to correct for this. I need to base it off solid samples across the scan region at various shades. I was going to expose a bunch of 35mm photo film and cut it up to get my samples.

The other inherent problem with scanning long strips is that the scanner sees the entire strip as "one image", instead of dealing with each frame individually, and tries to average the exposure for the entire length of the strip, which may have a lighter scene at one end of the strip than the other. I have never been able to reconcile in my mind how this is really workable, since the next strip may very well end darker than where it starts, thereby making the "joint" between the two strips fairly obvious due to sudden changes in density. Is this the cause of the pulsing that is seen in the sample? Or is that pulsing built into the original movie? Just curious.

The scanner doesn't do any exposure correction. The same settings are used throughout (unless I manually change them at some point). I was hoping to get a better dynamic range and come up with software to do the exposure correction. The scanner is ADVERTISED to do 48 bpp (16 bits per channel - though I'm sure it get's nowhere near that - and when I have it set that way it runs twice as slow).

As a matter of fact one of my main reasons for wanting to try a real film scanner is to get a real 12 bpp dynamic range from which I could digitally correct the exposure frame by frame.

Jim

[jimfcarroll]

Roger [...] http://www.moviestuff.tv

Roger,

I didn't realize who I was talking to. I've been reading about your products. You have some great stuff available. How open is a WorkPrinter? Could I do my own processing of the raw scans? What is the resolution per frame and dynamic range per pixel of the CCD? 1 fps certainly blows away 15 frames every two minutes. Do you think my approach might rival that of your CineMate line (with more work and perhaps a better scanner)?

Great products Roger,
Jim

[jimfcarroll]

OK. Per request I added a DivX clip that's a piece of the original.

http://www.jiminger.com/s8/clip1_DivX.avi

[VideoFred]

OK. Per request I added a DivX clip that's a piece of the original.

http://www.jiminger.com/s8/clip1_DivX.avi

Thank you!

Again, this is the first time I see someone producing good results with a scanner.. Sharpness is very good, but a little overexposure on the whites...

Altough this is a very nice project with many interesting technical challenges, I do think you should take a closer look at my system or Rogers Snipers and/or Workprinters.

But you already have done all this nice work... Have you tried even higher resolutions?

Fred.

[Petteri]

This is cool...very nice job.

I think you are not Homer kind of person like me?

"Forty seconds? But I want it now!"
- Homer Simpson

Petteri

[jimfcarroll]

Again, this is the first time I see someone producing good results with a scanner.. Sharpness is very good, but a little overexposure on the whites...

Yes. Getting the contrast right is the most difficult part of this. I've been trying to figure out the right way to do this but I haven't come up with anything great yet.

Altough this is a very nice project with many interesting technical challenges, I do think you should take a closer look at my system or Rogers Snipers and/or Workprinters.

Snipers are a bit on the pricey side for me (no doubt worth every penny though - I considered some of his stuff before I even began this project). I have been considering his WorkPrinter series though.

Following your link I just noticed your site. Very nice. But as the photo's on my website demonstrate - it's WELL beyond my capabilities to fashion. Really nice job though.

But you already have done all this nice work... Have you tried even higher resolutions?

I experimented with 4800 dpi. I actually have not scanned any large portions at that resolution but I think I will go back and try it. 4800 is the max (advertised) optical resolution of my scanner - (no doubt there is a lot of pixel overlapping though). I might be able to do better at 4800 dpi and then a digital reduction to 500 lines per frame. Thanks for the suggestion, I'll give it a try.

Jim

[VideoFred]

Yes. Getting the contrast right is the most difficult part of this. I've been trying to figure out the right way to do this but I haven't come up with anything great yet.Jim

We all have this problem, you know... no matter what system. The best solution is: exposing on the highlights (if possible with your scanner) and do some preprocessing, using gamma amplification to get the dark parts back to normal.

Fred.

[jimfcarroll]

Edit: I took a better look at your site, I'm very impressed about your knowledge both general and about writing software. How do you deal with the framerate pulldown? I think you can help me with my Avisynth script to do the pulldown. (See my 'auto-processing' topic)
Fred.

This is actually funny. I've been tracking your other thread hoping to get your pulldown script for AVISynth. I have extensive software experience (20 years in the industry) but my understanding of Video (since it's not related to the industry I'm in) is no where near as extensive. My former computer vision/image processing work ended in the early 90s - way before the digital video revolution got going.

Right now I don't deal with framerate pulldown. I let the encoding software (TMPGEnc) handle the conversion from 16fps MJPEG AVI to NTSC and I'm not sure how it even does it ... (though I've been wondering myself). It's for this reason I've been watching your other thread on "Auto-processing AVI-files."

Jim

[reedsturtevant]

OK. Per request I added a DivX clip that's a piece of the original.

http://www.jiminger.com/s8/clip1_DivX.avi

Very interesting to examine the artifacts from this approach - thanks for the clip. Does anyone else see an overall waviness? I see it most noticeably in the grass on the long shot where you see the carriage from behind. Looks more "rippling" than "breathing".

Jim, I wonder if this is from non-linearity in the sensor (like the exposure cycling) or from the scaling & rotation of individual frames.

Question: do you locate each sprocket hole and transform the adjacent frame independently? As a thought experiment, if you calculated the rotation, e.g., of the whole strip then used that to transform each frame then maybe there would be less frame-to-frame variability. Tons of tradeoffs of course.

Anyway, great piece of programming and not bad for under $300

[jimfcarroll]

Very interesting to examine the artifacts from this approach - thanks for the clip. Does anyone else see an overall waviness? I see it most noticeably in the grass on the long shot where you see the carriage from behind. Looks more "rippling" than "breathing".

Jim, I wonder if this is from non-linearity in the sensor (like the exposure cycling) or from the scaling & rotation of individual frames.

I think this is an artifact of a "nearest neighbor" rotational correction. I minimized this (somewhat) in later scans by correcting the film guide (the "track" that the film fits in) in the scanner. A software approach would also be straightforward (using bilinear interpolation rather than nearest neighbor).

Question: do you locate each sprocket hole and transform the adjacent frame independently? As a thought experiment, if you calculated the rotation, e.g., of the whole strip then used that to transform each frame then maybe there would be less frame-to-frame variability. Tons of tradeoffs of course.

In my early experimenting I didn't use a "guide" for the film in the scanner and the rotation actually varied from one end of the strip to the other. My software was written based on these early scans and as a result I have kind of a hybrid of what you are suggesting and a purely independent approach. I actually use the film edges (mostly, the edge opposite the sprocket hole) taken piecewise around each frame, as well as the sprocket hole, to align the frame.

Thanks for the complements.
Jim

[reedsturtevant]

..."rippling"...

I think this is an artifact of a "nearest neighbor" rotational correction.

That makes sense. It does looks like "swarming" as the pixels are influenced by their neighbors. 8O