software

[jcupitt]

Hi @jackw01, I came across your interesting repo on reddit.

I was wondering what software you were using for processing these scans? I made a tool you might find useful:

https://github.com/libvips/nip2

It's a bit like a spreadsheet, but for image processing, so cells can contain whole images. You can enter formula that describe the processing you want (eg. 1 - 1 / log A1 or whatever). Whenever you change anything, perhaps replacing an image in a cell or editing a formula, everything updates. It's built into most linuxes, or there are windows binaries in the "releases" tab.

I tried (very quickly!) processing your comparison shot:

That does the following:

• extract the shot from your nice lightbox
• split the colour bands to RGB, then for each band
• estimate the minimum density (max pixel value) by taking the average pixel value of an area of film near the centre edge
• max density by doing a gaussian blur and taking the min (to avoid camera noise)
• rescaling the grey values to 10 .. 240 (ie. avoid the extreme edges)
• recombine the corrected greyscales to RGB
• invert in density space (ie. exp (1 / log X))
• adjust brightness

It makes this output, which looks OK-ish:

It's just a quick demo, it needs some obvious improvements:

• in your blue band, the film is brighter than the sprocket holes, I'm not sure why (hence the cyan squares)
• flatfield correction of the lightbox (your box is quite a bit brighter in the centre)
• you could estimate the actual transfer function for your film + camera + lightbox combo from a shot of a greyscale, like a macbeth chart
• once you've found the transfer function, you could use a macbeth shot to compute the optimal 3x3 colour correction matrix, which would be great fun to try

[jcupitt]

... that workspace, if you're interested:

http://www.rollthepotato.net/~john/lightbox.ws