I tried a couple of different approaches when mapping the face to the head and ended up with something that's probably not the most efficient way but worked for me. In AE, I had the main composition with the raw stop motion, a separate "face" composition and then three separate compositions for eyes, eyebrows and mouths. The face comp contained the time mapped eye, eyebrow and mouth comps, and also a solid yellow fill and a square black border. While I was mapping the cylinderised face in the main composition, I kept the solid yellow fill on at 50% opacity and the black border on at full opacity. As long as the yellow fill exactly covers the whole head area, you know that your head is the right size and orientation. You use the black border to check rotation - it's a lot easier to line up a moving square than it is to line up eyes, mouths, etc. Before the final render, you just hide the yellow solid and black border layers and you're left with the face.

Hmm, not sure how much sense that'll make if you haven't actually looked at how stuff is laid out in my project. You can see an early version of the face comp here before I figured out to use the black border. Notice also the handy facial reference of me singing along.

I didn't do much colour matching at all. I changed the cylinder light source to a light yellow and tried to match the direction to what was used in the scene, and I set the overall face opacity to 85% - that let some of the shadows through and made it look a little bit more natural.

The final lip synch, once I'd discovered this method through lots of trial and error, took about a day. Previous attempts took longer and didn't look anywhere near as good.