A systematic procedure for the extraction of multiple-invariant shape descriptors is introduced that is based on generalised auto-correlation functions. It can be carried out in parallel processing networks that also serve the extraction of motion parameters by flow-field analysis. It is argued that this dualistic processing principle is neurobiologically relevant. Models of neural networks are presented and their properties in both modes are discussed.

Results of psychophysical experiments are reported that permit first insights to fixation strategies of humans, when confronted with composite motion of extended patterns. It appears as if pursuit eye movements not only minimize the translational component of a moving retinal pattern but also other changes, such as rotations and expansions. This is achieved by pursuit fixation of a geometrically well-defined location that generally differs from the optic center of gravity of the retinal pattern.