https://jackw01.github.io/scanlight/
After reading this, it feels like RGB light is the main reason why professional scanners are using it for rich colors but I dont think that there are any commercial grade RGB light sources available even from Negative Supply.
But I wonder if there are any RGB light source that I can buy in order to modify it as a scanning light source?
As far as I understand it, RGB lights really show their useful advantage when:
Someone absolutely correct me if I am wrong here. It is not to say you cannot use RGB with a full visible spectrum sensor, which this shows you can, but the benefits really come into their own when done with monochrome. Seems like a lot of extra steps that can often be worked around perhaps? I don’t know I have never tried it and I would love to see more!
Cheers y’all
If I’d venture into building my own scanner, I’d go for separate R, G and B lights and a relatively simple sensor in order to keep hardware cost low. The sensor wouldn’t need to be panchromatic because colour dependent sensitivity could be corrected in the electronics or software that assembles the scanned image. Early scanners also had e.g. a set of rotating polygonal cylinders that directed the view of a single sensor cell to read the specimen to be scanned. Similar to how point of sales scanners work even today.
With technological advances and the related shift of costs, the rotating polygons, one-eyed sensors and multicolour lighting, have gone in favour of line sensors and white light, both traveling along the specimen in the case of the typical flatbed scanner. Take that one step further with a two-dimensional sensor…and we get what most of us do here: camera scanning.
While getting away from moving parts can make things more robust and lower cost, other compromises creep in like a Bayer-Mosaic and the necessary de-mosaicing algorithms and the imperfections they add geometrically and colourwise. The typical 24 Megapixel Bayer sensor is similar to a 6 Megapixel sensor from a resolution point of view. A 24 Megapixel sensor has to fabricate 24 RGB Megapixels, each from 12M green pixels and 6M pixels of red and blue respectively. The better the algorithms get, the less obvious will the introduced errors be.
Anyways, we can scan with high-CRI light or discrete R, G and B lights and all we get are different starting points and therefore different ways to get what we want as final image.
(And we can only discuss which way is “better”, once we have defined a set of criteria that we use as a measuring standard. If that standard is based on personal taste, the scope of “better” shrinks to a single person space.)
Any RGB light source available? Yes, e.g. an OLED display of a mobile device or as spare “bulb” of a RGB beamer … and/or some DIY. Lowest cost phones are around 150 USD (Xiaomi, Samsung etc.) …and I have no idea about the spectral qualities of such devices.
Looks like $$$$$
@yunhao.jiang , can you post a link that gear?
It’s like around 200 USD, still on presale now. I’m not sure how to get them ship to US, link here:
There is actually such a thing and not even one:
classic Philips PCS 140/150 enlarger . It has RGB light source - under the hood there are three halogen lamps each fronted with dichroic RGB filters. With the use of control block you can have any mix of colors or single channel. I have this , it works, though it is still experimental stage as I need to put all mechanics together.
The second option is very expensive , but if you can justify the purchase as the home cinema, you can have it tomorrow. There is budget 
$900 three laser video projector Optoma ML1080. It has tree narrow band lasers. I wrote the app RGBGenerator available on GitHub which gives you ability to just dial in RGB numbers and have the light of proper color. The RGB peaks are suitable to our needs and close to Status M illumination. Again I did preliminary runs , it works , but again it needs some diy effort to make it usable. Here is spectral distribution of this projector
So that’s is what readily available, of course proper application of color sciences is needed to make this work with camera sensors.
Another idea: RGB Party lights. Available around USD 50.-
Tried it with a dichroic colour head of a Durst M600c and found that the results weren’t better or worse than letting the light pass unfiltered…
I did try this too, but stopped quickly, as I found that the dichroic filters in CMY color head create really weird wide spectral distribution because those filters work by allowing two bands of RGB to pass simultaneously. I cannot say for sure if RGB any better - my first impression was that contrast was higher than usual. But again this is something I plan to explore just for fun of it.
Cannot comment on this because I don’t have any device to measure spectral properties of the light. Moreover, spectral sensitivities of camera sensors aren’t that precise as one can see in the camera sensor data provided by dxomark.com. How the necessary mathematical corrections deal with narrowband lighting is another question I can’t answer today.
Bottom line: Due to what I found testing different ways of lighting, I conclude that a high CRI white light provides good conversions at relatively low cost or effort. Managing stray light was more important.
I agree with you that properly exposed film can be inverted with proper light and produce good output . I though doubt that high CRI is the best, but let’s leave this aside. Typically every person has to spent certain time to find the specific workflow in terms of scanning exposure and NLP settings. Persistence typically pays off. My quest to find better light is sort of play for the sake of play. I also need a good process to scan slides as of now before NLP comes with slide feature I am not satisfied with how my slides come up. I really appreciate discussing things with you.
Hopefully you might write about that some time! Colour, contrast control, linear profiles, IT8 targets…?
I found that NLP does a fairly good job on e.g. Tiffs that I exported from simply inverted captures.
(I’ve opened a respective thread, but posted the steps here for ease of use.)
You know ,everything in me would protest such a tortured process. I understand how it works and why it works, but I cannot possibly force myself to do it ;;-))
Be the master of your own happiness … and wait for NLP version 3.1 
…that does not seem to cover more easy to do things, apparently 
Been there, done that as they say. Read this thread:
In short: I couldn’t find a commercial one and build my own, based on the theoretical argument that individual R, G, and B was better. Going the extra mile to use a monochromatic sensor and combining in software seemed too much, but I get “good” and reproducible color with everything in software linear and neutral. Do I get better colors than with broad spectrum daylight? Don’t know since I didn’t objectively compare, but at least I’m using something that’s theoretically the best way
Anyhow, it seems that commercial versions would become available, and yes, professional built scanner use trichromatic light sources.
Professional film scanners back in the day, and not too far back in fact, also used special fluorescent tubes in conjunction with tri-linear RGB sensors, my Imacon Precision II does, as did the later Hasselblad X1 & X5 though they were special (very expensive) cold cathode light units I believe, my old Microtek Artixscan 4000tf does also. In theory I would have thought a 4 shot pixel-shift Bayer-sensor camera could get similar results from a full spectrum light source with the right software. I suppose the difference might be that camera RGB camera sensors are not narrowband, so too much overlap.
According to dxomark.com, sensitivities vary greatly. Two examples:
I just spoke with him and the price will be around $300 tho it’s $200 in Chinese currency. Cant wait to try.
