[gjm11]

You'd think they could be smarter about combining the R,G,B images. E.g., allow for the thing under each pixel to have moved a little from one image to the next, look at all their narrowband images (which will typically be strongly correlated, especially for nearby wavelengths) to estimate how fast each part of the image is moving, and then to get a pixel in the final image combine its G with slightly offset (maybe interpolated) R and B.

Fiddly but hardly, er, rocket science.

(Of course for some purposes you really need the unaltered R,G,B images. But if you're producing a single RGB JPEG image for public consumption, that's not one of those purposes.)

[EDITED to add:] More difficult, I guess, is dealing with things that appear in only some of the planes. E.g., if G is taken last then there will be bits of Earth at the trailing edge of the moon in the G frame that have no counterparts in the R,B frames because the moon occluded them in those frames. So there will still be artefacts in the image. But I'd have thought they'd be less objectionable than the ones you get from just naively stacking the R, G, and B.

[mturmon]

It seems like you have re-discovered why segmenting the lunar image and shifting it backwards in time, to align the three color planes, was not done. They don't have the information to fill in the resulting gap, because it was occluded. In a very real sense, they have nothing to put in those pixels.

If they had wanted to do it, they could have. The ephemeris (i.e., basic imaging geometry relating to the location of Earth, Moon, and camera) will be very well-known. That bread-and-butter image processing is no problem.

The historical context here is that NASA centers, prominently JPL, were among the inventors of what became digital image processing (https://en.wikipedia.org/wiki/Digital_image_processing#Histo...).