I don't think I've *ever* seen a stereo image where they literally used an R filter for the left eye image and a B filter for the right eye. This is normally used to *present* monochrome images to each eye. I think you still misunderstand conventional anaglyphs (I'm prepared to be wrong, of course).

The problem is that for "true" colour anaglyphs we need to present each eye with a full RGB image, or to find a compromise that "nearly" works. The shutter-glass or polarisation method is one way.

Another way is the mirror stereogram. You have two images, photographed with a small separation. You put them side by side, but one is printed in reversed L-R mirror symmetry. Then, you get a mirror and place it vertically between the images, and touch your nose to it. One eye sees the normal image. The other sees the inverted image *in the mirror*. It is thus re-inverted and positioned over the top of the normal image as seen by the other eye.

This works really well.

But, you need *two* RGB images...

I think there was a design limit on the number of filters. The option was to have RGB on both eyes, or to have as wide a range of filters as possible. The setup is to have one camera as full-range visible light camera with near IR and UV. The other camera is a multispectral IR camera. AND, they have a set of filters in common for key tasks. This gives them several different science options by combining images/filters.

The accumulated experience with imaging Mars suggests that one can get away with two filters - one reddish and one bluish. It isn't perfect, but it'll do.

You should be able to mix L2 & L7 as R & B with an appropriate synthetic G derived from them to give a crude colour RGB image. Then do the same for the right "eye" with R2 & R1.

Another way is to find two sets of RGB filtered images taken when the rover was in different physical positions, but these will likely involve too large a shift. Maybe some of the earlier and later El Capitan images? (But you could never automate this, it's too tricky about geometry - you'd need to rescale some of the images to match)

And yes, a trapezoidal distortion would help, making the horizons match and reducing the near-field offset.