Adaptive hierarchical structures

Abstract

The construction of a hierarchical system to distinguish classes of patterns can be improved by the combination of a dynamical elastic matching with a time dependent image resolution. Using the elastic matching as a preprocessing for synergetic computers one achieves an invariant perception by means of arbitrary spatial transformations. On the other hand a time dependent resolution, realized with a Gaussian distribution, can be interpreted as a dynamical change of the available image information. The coupling of these two dynamics results in a hierarchy of locality of pattern transformations. Therefore this system utilizes a dynamical feature extraction and implies the definition several classes of patterns.