When is scene identification just texture recognition?