How do we recognize a face despite changes in viewing angle, image size, facial expression, hair style, etc.? As an example, look at the pictures below. Our ability to recognize a person despite these changes is an example of "constancy."

Can face recognition be explained by template representation? There would have to be a lot of stored templates to handle all the different images we can recognize.

Can face recognition be explained by feature representation? Consider the following rearrangement of features:

The face on the right has the same features as the face on the left, but doesn't look like the same person. So face recognition is more than just having the right features; they need to be in the right places.

Can face recognition be explained by structural representation, in which the features have to be not only present, but in the right place? Consider the upside-down pictures below. They both have the features in the right places.

Now turn your head upside down and look again. Ugh! So it's more than getting the features in the right place, it's getting a whole variety of featural relationships correct.