A technique for reducing the amount of work you have to do when rendering
3D computer graphics with
polygons. If done right, on average it reduces the number of polygons to be drawn by half.
First, though, you need to understand something about the way polygons are used by computer graphics people. As far as CG people are concerned, polygons only have one side. The other side is considered "see-through" or non-existent. The side which is drawn depends on (a) the order in which the vertices are specified and (b) the "handedness" of the system (usually right handed). So you use the right-hand rule, curled in the order of the vertices. An example:
A
|\
| \
| \
| \
+----\
B C
is facing towards you, so would be rendered, but
A
|\
| \
| \
| \
+----\
C B
is facing away from you (into the computer) so wouldn't be rendered.
Having one-sided polygons makes a lot of sense, because otherwise you would have to render both sides of a polygon, and since many of the things you are modelling are solid, you can never see both sides of a given polygon anyway. Solid objects tend to be self-occluding. Consider a cube. A computer graphics person would label the corners in such a way that all the faces were facing outwards (i.e. away from the centre of the cube). But at least three faces of the cube are going to be obscured anyway.
Backface culling just takes this a step further. Work out if a polygon is facing away from us, and if so, just drop it right away. In an average scene, half the polygons will be facing away from us. For example, only three faces of the cube will ever be facing towards us anyway. And hence you get a good speedup.