Ein Beitrag zur effektiven Implementierung adaptiver Spektraltransformationen in applikationsspezifische integrierte Schaltkreise.

In many areas of one- and two-dimensional signal processing, the task is to classify signals or objects independently of their current positions by means of suitable features. With the help of fast nonlinear spectral transformations a position invariant feature extraction is possible. In this work real transformations are presented whose features can be adjusted with respect to different parameters. To mention is the group invariant behavior the computational complexity and the implementability in application specific circuits. By using different computational structures, for example, the separation property can be adapted according to the task. Based on the concept of characteristic matrices, a generalized method for the computation of the transformations is derived. With respect to their characteristics, the transformations to be presented can be said to be equal or even superior to others. In combination with a fuzzy pattern classification (FPC) method, a system-on-programmable-chip pattern recognition system is developed which is implemented on a programmable application specific circuit (FPGA). The system is capable of classifying pixel-based images. In the application of print image inspection, the pattern recognition system proves to be applicable in practice.