Athiril
Established
I know many fellow film shooters are not rich enough to afford high end scans and sometimes feel sad.
So I thought I would share this.
A bit of background information as to why this works.. the technique is known as superresolution, basically a single 'point' as deifned by the lens (an airy disk perse) will be of a certain size (traditional resolution limit), but of course they are not necessarily arranged in a grid with equidistant spacing like an image sensor.
Basically, the point spacing can be smaller than the size of the point itself, they can overlap, and they can move micro amounts between shots.
Between multiple shots, it is possible to calculate their position to a finer degree of accuracy than the size of the point of detail itself.
This has limits of usefulness in photography where you need to take multiple shots, as with any multiple-shot technique.
OF course, when we are scanning film.. our scene (the piece of film) is fixed and unchanging. So in scanning, it can be more useful.
Examples of Superresolution
http://www.ok.ctrl.titech.ac.jp/res/CSR/MTSR/index.html
Now here is my example, the first scan is a 3200 dpi scan on a Epson 4990, the second is from multiple scans combined using superresolution and sharpened (note that it increases resolution, but in order to see it you need to raise the acutance, or the level of which it can be more perceived easily) after downsampling back to 3200dpi (superresolution uses an enlargement factor, in this case 2x to account for the cases of effective use of the pixel space already, so you need more for the resolution increase in those particular cases).
And here is the result (I should also mention this still has the benefit of improving SNR/reducing noise):
Highly improved.
The scan is from Velvia 50.
I used off the shelf software called PhotoAcute for this, it needs a camera profile, but you can specify one manually, I specified the 5D Mk III and 24-105 iirc.
Newer algorithms work more effectively iirc, and there are multiple open source examples, I'm looking at trying to incorporate these into a working program for scanners using libtiff.
As it should also be more effective with the variables configured properly for a scanners performance.
Just for kicks, here is a 4000 dpi sample of a nearby area of the same piece of film scanned with a 4990, Plustek 8200i and Flextight 949. The superresolution sample appears similar to the 8200i, which is funnily enough rated at 3200dpi.. I should have downsampled the superresolution sample to 4000dpi and see what I could get with that, but I wasn't actually planning on comparing it with the below images in the first place.
How to do it:
Prepare your scan, hit scan multiple times using the same settings, the more source scans you use the more precision can be calculated (to a limit with this piece of software though). There also seems to be a pixel dimension limit before the software craps out (so if using MF film, a 2400 dpi source scan might be best).
Drop them into the photoacute window, go to step two, hit increase resolution and remove chromatic abs if you want that, then specify a camera profile and lens setup. Go to step 3. Wait. Highlight/click result, hit save, change to tiff, drag into Photoshop or other, down sample it, and find a sharpening setting that matches the resolution frequency that bumps it up.
So I thought I would share this.
A bit of background information as to why this works.. the technique is known as superresolution, basically a single 'point' as deifned by the lens (an airy disk perse) will be of a certain size (traditional resolution limit), but of course they are not necessarily arranged in a grid with equidistant spacing like an image sensor.
Basically, the point spacing can be smaller than the size of the point itself, they can overlap, and they can move micro amounts between shots.
Between multiple shots, it is possible to calculate their position to a finer degree of accuracy than the size of the point of detail itself.
This has limits of usefulness in photography where you need to take multiple shots, as with any multiple-shot technique.
OF course, when we are scanning film.. our scene (the piece of film) is fixed and unchanging. So in scanning, it can be more useful.
Examples of Superresolution
http://www.ok.ctrl.titech.ac.jp/res/CSR/MTSR/index.html
Now here is my example, the first scan is a 3200 dpi scan on a Epson 4990, the second is from multiple scans combined using superresolution and sharpened (note that it increases resolution, but in order to see it you need to raise the acutance, or the level of which it can be more perceived easily) after downsampling back to 3200dpi (superresolution uses an enlargement factor, in this case 2x to account for the cases of effective use of the pixel space already, so you need more for the resolution increase in those particular cases).
And here is the result (I should also mention this still has the benefit of improving SNR/reducing noise):
Highly improved.
The scan is from Velvia 50.
I used off the shelf software called PhotoAcute for this, it needs a camera profile, but you can specify one manually, I specified the 5D Mk III and 24-105 iirc.
Newer algorithms work more effectively iirc, and there are multiple open source examples, I'm looking at trying to incorporate these into a working program for scanners using libtiff.
As it should also be more effective with the variables configured properly for a scanners performance.
Just for kicks, here is a 4000 dpi sample of a nearby area of the same piece of film scanned with a 4990, Plustek 8200i and Flextight 949. The superresolution sample appears similar to the 8200i, which is funnily enough rated at 3200dpi.. I should have downsampled the superresolution sample to 4000dpi and see what I could get with that, but I wasn't actually planning on comparing it with the below images in the first place.
How to do it:
Prepare your scan, hit scan multiple times using the same settings, the more source scans you use the more precision can be calculated (to a limit with this piece of software though). There also seems to be a pixel dimension limit before the software craps out (so if using MF film, a 2400 dpi source scan might be best).
Drop them into the photoacute window, go to step two, hit increase resolution and remove chromatic abs if you want that, then specify a camera profile and lens setup. Go to step 3. Wait. Highlight/click result, hit save, change to tiff, drag into Photoshop or other, down sample it, and find a sharpening setting that matches the resolution frequency that bumps it up.
ColSebastianMoran
( IRL Richard Karash )
Article on this in current issue of Digital Photo Pro. See page 2 of the article "Stacking" and comments on PhotoAcute.
Crux of the article is: PhotoAcute takes multiple images of same scene and creates one better image. Article is about multiple shots of the scene in your digicam. Interesting, Athiril, to use it for multiple scans of a film original.
Thanks for posting this.
Crux of the article is: PhotoAcute takes multiple images of same scene and creates one better image. Article is about multiple shots of the scene in your digicam. Interesting, Athiril, to use it for multiple scans of a film original.
Thanks for posting this.
Athiril
Established
I should have posted the results much sooner! No matter, still should be useful for some.
Athiril
Established
My cut film is already very flat, but it may help as well as optimising the focus distance since we are pushing the resolution up. Would be interesting to try, but I already have access to a Flextight 949 at work I can use 
Thanks, Daniel, for sharing this information, interesting post.
Regards,
Brett
Regards,
Brett
