Sensor pitches for high-magnification photography

For computational photography that is resolution-oriented, narrow-band R/G/B LEDs might be worth the potential metamerism because broad-spectrum light will prevent sharp focus with a lens like that, even in a single color channel. When I first read of this property of the lens and the fact that you were going to focus-stack, I was thinking of the possibility of actually focus-stacking each color channel independently, and then scale once to combine them, which sounded easier to me than trying to get focused-and-scaled RGB frames to stack, which would require accurate guesstimation of the focus offsets for the multiple wavelengths.

You want to minimize pollution from any unfocused data if you want maximum resolution, so I don't think that it helps to have more than one wavelength used for any frame, unless the separation of the color channels in the CFA is strong enough to lose cross-channel sensitivity to the neighborhood of the noise floor. IIRC, I shot a three-socket lamp with R, G, and B narrow-band LEDs over 15 years ago with one of my first DSLRs, and there was little signal from the "wrong" bulb for each raw color channel, but I think most later cameras overlap more, and you may get some "pollution", mainly in the green channel.

I mean, you could use R,G, and B LEDs in each frame and accept a small amount of pollution, and stack each color channel independently and then scale and combine them, but if you are already doing all this extra work to maximize resolution and focus-stacked DOF, you may as well go all the way and only record one wavelength at a time.

Of course, if you take the approach of one narrow-band LED each exposure, then there is no reason why, in theory, you couldn't use more than 3 wavelengths, and actually reduce the metamerism.

That tradeoff seems to be unavoidable with a lens like that. Any increased breadth of bandpass also decreases focus, even within a single raw color channel. What range of narrow-band LEDs do you (or consumers in general) have access to? Surely just doubling or tripling the number of wavelengths could greatly decrease the metamerism; you just need a software solution for constructing full-spectrum RGB from arbitrary wavelengths.

That would be simple enough to test. Just make tiny white dots on a black matte board and illuminate them with various LEDs and see what their PSFs are like, or if you just want a simple number to compare, do something like MTF Mapper or Imatest with a white target and narrow-band LEDs, with images captured with critical manual focus.

How do you set up to stitch focus-stacks of macros? That sounds pretty messy. Stitching is most suitable when the subject matter is far enough away from the lens that parallax is negligible. It should only work up close when the sensor is the only thing moving, like when using a shift lens and the lens, not the camera, is fixed. A stereoscopic focus stack, however, is easy enough to envision.

At 10:1, I don’t think parallax will be an issue. At much lower magnifications you can move the rear standard.

focus-stacking each color channel independently, and then scale once to combine them
That is the way I could try.

Or, when I start to think this, it might be possible to make a LoCA-free image from focus stacked shots combaining R, G and B channels from different shots.
There is "Split Channels" command in Photoshop. It makes 3 grayscale pictures from one RGB. I don't have PS but if I remember, PS names those as Img_001_red, Img_001_green and Img_001_blue, or something as easy to use. PS can also combain grayscale channel pictures to rgb. I don't remember the command, but it's in the same place as "split channels".

You only have to find in which picture's red channel at the some point is as sharp as that point in another picture's green channel and in third picture's blue one.

The way I envision it, You wouldn't need to be concerned with anything except having the focus distance range the same for all channels, which would mean a slightly different range of lens-to-subject distance for each channel. I'm not sure that the errors from not doing so would be visible, though, so it may be nothing to be concerned about unless the distance ranges were way off. and parts of the subject at the ends of the range would have a big difference in focus across the 3 channels. I'm not sure if we notice artifacts like that very much, as we get most of our sense of focus from greens.

Joe W always recommended moving the camera, www.dpreview.com/forums/post/34497523
He used a Manfrotto focusing rail to move the entire camera assembly. His setup is at the bottom.

Monochrome sensor, one colour plane for each capture.

Someone should write an extension to one of the open source stacking programs to separate the colors in the raw image and then place the best focus at the correct position in the stack.

Given the CFA bandwidths, it might be not as good as starting from a monochrome sensor and narrow-band lights.

The simple beauty of such an approach is that since you already start with known wavelengths and don't have to interpolate them through the CFA, all a software needs to know is the sensor's relative sensitivity to them. No CFA surprises or complexities.

A monochrome sensor is certainly simpler and gives full resolution with equal pixel SNR potential, but narrow-band lights will only see CFAs as 3 different neutral density filters at any given wavelength. You can get tremendous DR when using a single narrow-band LED with a CFA.

this is a fast and dirty test only (120 images should have been 200),first time ive stiched a macro. dont know why you would want to as 30meg image can print huge 5x7 foot print. this stitch could easily print 10 foot wide. it was a very simple stich in photoshop.