The grey level profiles of adjacent image regions tend to be different
, whilst the 'hidden' physical parameters associated with these region
s (e.g. surface depth, edge orientation) tend to have similar values.
We demonstrate that a network in which adjacent units receive inputs f
rom adjacent image regions learns to code for hidden parameters. The l
earning rule takes advantage of the spatial smoothness of physical par
ameters in general to discover particular parameters embedded in grey
level profiles which vary rapidly across an input image. We provide ex
amples in which networks discover stereo disparity and feature orienta
tion as invariances underlying image data.