Integrating spheres are used in commercial scanners, as the truly diffuse light source minimises the appearance of shallow base scratches appearing on the final scan.
They are however, large compared to a standard light source, so positioning could be difficult.
You can make your own quite readily, the main problem with they DIY ones is that they are great at diffusing, but lose some of the spectrum as they are not so great at reflecting the full spectrum of white light.
However you can improve on the standard DIY spheres by using Barium Sulfate as the interior surface, it has great reflective properties.
The other option, if you are in the US, is to use Glidden Diamond 450 Titanium White paint, it reflects more of the spectrum than any other paint we tested, and would be a good choice for coating the inside of a DIY sphere.
I prefer to use an array of Red, Green and Blue LEDs, as this allows you to set the colour balance to match the spectral response of whichever sensor you use. It also allows you to adjust the light for better scanning of faded stocks.
High CRI LEDs are better in theory than low CRI LEDs, but there are problems, typically in the skintones which can make some not a good choice as a light source for a scanner.
Thanks for the info Peter! So would a scan of, say, a faded magenta print turn out better after adjusting the individual colored LEDs? Are all R, G, and B LEDs created equal or is there a better choice of color wavelength to match film stock?
Yes, absolutely. With professional scanners you adjust the light for the particular stock, and also to help compensate for any fade. It makes a huge difference to the amount of colour information you can extract from the scan.
You can also choose the spectral response of particular LEDs to best match that of the sensor you are using.
Every brand LED has a different response curve, so they are not created equal The Arri scanners for instance use two different Red LED types to cover the full red response.
@Peter That’s super helpful, thanks! Curious about spectrum when it comes to the actual light component itself. It’s becoming clearer to me why you would want several LEDs of each color type to create your overall “white” light for control. But how much of an improvement is this, really? Are most white LEDs going to work just fine, as long as color is corrected in software? Are we talking a 40% reduction in color quality? 5%? I’m new to the science of lighting and am unsure how much effectiveness I’d be trading for increased complexity in circuitry with the R + G + B option.
I read this page on Frank’s site after seeing your post and he’s very informative. I learned a lot. Thanks for mentioning it.
LED choice should match the sensor choice (CCD vs. CMOS, etc.)
Turning color channels off at different times to control color balance
Maximum use of the sensor’s color spectrum actually reduces noise.
It is a massive improvement if you need to do a colour grade after capture, (whihc is nearly always) or if you have faded film stock to deal with, I’ll post some examples of heavily faded film with white LEDs vs an array of RGB LEDs.
Especially if you are using a bayer sensor, you end up unable to correct far enough in software if the light source doesn’t match the sensor, or if you have fade.
I’ll dig some images out tonight.
Rolling shutter on DSLR is one of the problems when you use pulse width modulation of RGB LED array. A constant current reduction (CCR) LED driver could avoid the rolling shutter but CCR does have negativ impacts on the color temperature.
Has anybody experience with the color shift due to a variable current?
That is another reason I do not use PWM to drive the LEDs, you get flicker problems even with global shutter setups.
CCR can have colour change issues, so it is important to calibrate your setup correctly to cater for it.
It might be a good idea to try out using lenticular sheets for light diffusion. These are often used in holographic displays or as a filter to turn a point laser into a line focus laser. They only spread light in one direction, but you can make a little sandwich of them, two layers oriented perpendicular to each other.
Take a look here to see how they are used in a (book) scanner:
The issue is that I’m getting scan bars slowly moving down in my scans. It’s like a darker horizontal band taking about 1/4th of the screen and you can see it moving when playing back a scan. This is happening at 1/48th shutter. It seems to be better at 1/60th or 1/30th which makes sense but I need to be able to have more flexibility with the camera shutter speeds.
Any suggestions on how to improve my power so I could use more shutter speeds and not have to worry about scan bars ?
Someone suggested I install a High Speed Dimmer similar to this and said it would improve flickering. What do you guys think? Amazon.com
i honestly think trying to use a balanced light source is the wrong approach - assuming you want colour fidelity. you would be much better off with separate red, blue, and green leds to measure the densities of the coloured dyes in the film.
this with some colour space conversion (a 3x3 matrix) should give you more accurate results, abet less “retro” looking.
I would recommend you use separate R, G, and B lens.
The continuous light spectrum has already been quantised into three bands by the
sensitivities of the of three layers in the film when it was exposed.
This has been translated (via development) into three slightly different colours in
the dyes in the film.
To get accurate colour reproduction you need to measure the densities of the dyes and
then matrix them together to translate from the colour sensitivities to your preferred video
There are several papers which explain this much batter than I can on the BBC’s R&D website:
Thankfully dichroic lenses are not necessary these days, three LEDs will do.