I have been experimenting with creating synthetic L4, L5 and L6 images for use when one or more of these color channels is 'missing'. So far I have done this for a six image sequence from Spirit starting with image 2P130265292RAD0700P2543L2C1.img from sol 44.
The images I used were RAD files that were radiometrically corrected and converted to PNGs using my IMG2PNG utility. I then measured the intensity of several representative features (the sky, Columbia hills, dunes, hollows, rocky terrain, several rocks etc.) in each image and then determined which combination of two images yielded the best synthetic image in a least squres sense. This is by no means a 'scientific result' but the results looked good:
http://www.mmedia.is/bjj/misc/mer/s044_L456.jpg (L456)
http://www.mmedia.is/bjj/misc/mer/s044_L257.jpg (L5 and synthetic L4 and L6)
These images look remarkably similar. The processing I did is only rudimentary, for example I'm not happy with the color of the sky.
The results of the calculations I mentioned were:
L4=1.408871 x (0.462 x L2 + 0.538 x L5)
L6=1.153437 x (0.305 x L5 + 0.695 x L7)
L5=1.037111 x (0.204 x L4 + 0.796 x L6)
L4=1.223131 x (0.581 x L3 + 0.419 x L5)
This is for the surface/sky only, I made no attempts to make Spirit itself look nice. In an ideal world you'll want to process the resulting synthetic images exactly as if they were 'real' images.
I now plan to do something similar for Opportunity. The results will probably be somewhat different since the terrain there is very different.