I feel you need to advise what the true objective of the exercise is.
You state you wish to know whether a head is over a pillow or not. But for what reason - this relatively simple item of information could actually influence the design of your system. For example, if the objective of the exercise is to determine whether someone has fallen out of bed, then determining whether or not their head is on the pillow is determination of the incorrect variable.
However if it truly is a "head-on-pillow" objective, you should consider using multiple sensors for the following reasons ;
1) the weight of a head is inconsistent (big heads, small heads and all sizes in between
2) some things that weigh the same as a head are not heads (books, cats, dogs etc)
3) head weights are transitory - sleepers might "bang" their head into the pillow or might lift their head periodically whilst being asleep
4) heads (and animals) emit heat, whereas inert objects (books etc) do not
5) heads (and animals) generally make noises (grunts, snores etc)
6) heads (and animals) come in different shapes and sizes
OK it all sounds a bit silly, but the "human head" definition consists many variables and for a pillow sensor to be 100% reliable you need to consider all the variables you can think of - and then a few more
So, you probably need to look at an algorithm that takes into account factors of the above points and "compute" whether or not the "head" actually is a head or is it an inert weight or an animal that is looking for somewhere to sleep. You might also wish to consider the macabre possibility of considering the head actually expiring during sleep - "it was a live head at the start but now it's a dead head"