Original Abstract of :
A Geometrical-Transformation-Invariant Pattern Recognition Concept Incorporating Elementary Properties of Neural Circuits

We introduce a rather general representation of pictorial data that allows one to derive features which are invariant under geometrical transformations. These features are based on intrinsic measures of given signals. Beside these absolute inner geometrical relations, it delivers information about a signal’s situation with respect to external coordinates.
The processing concept was developed for parallel computing structures and we try to give some design hints for suitable parallel computing mechanisms. This allows us, to a certain degree, to compare its capabilities with those demonstrated by biological pattern processing systems, e.g. by the human visual system. Thus, our approach is less suited for implementations on “von Neumann”-type computers.
To illustrate our concept, we show results obtained from an early coherent-optical simulation and from a more extensive experiment performed on a digital general purpose computer.