Motivated by the impressive human performance for visual perception and by psychophysical and neurobiological findings about the mechanisms in the human visual system, I focused on two aspects of the problem: hierarchy and recurrence. I found that local recurrent interaction in a hierarchy of simple processing elements is an efficient way to incorporate context into visual inference by integrating bottom-up, top-down, and lateral influences.
In November 2002, I defended the dissertation thesis "Hierarchical
Neural Networks for Image Interpretation" at FU Berlin.
The thesis is available as Springer
LNCS volume 2766.
