A color picture can be identified as an element in S ^{ G }, where
S is the set of colors (in our case, a 256x256x256 color cube) and where
the group
G = Z _{ n } x Z _{ m } represents the set of pixels
(if the picture has width n and height m). We think of a picture
as a discrete analogue of a vector field because
at each point is attached a 'rgb vector' (red(x,y),green(x,y),blue(x,y)).
A discrete version of the heat flow e ^{  L t } with
Laplacian L acting on vector fields is a special kind
of cellular automata map. (In general cellular automata can be seen
as discretisations of partial differential equations. Not only space
and time are discrete but also the target space).
The 'heat flow cellular automaton' averages in each time step
in each color coordinate the 5 color values of its 4 nearest
neighbors and its own color.
We have applied this cellular automaton on a picture. What
you see is the movie, when time is reversed.
