Theoretical foundations of synergetic image processing

In this paper the theoretical foundations for applying Synergetic Computers (SCs) to signal and image processing problems are developed. Among the numerous algorithms, which are usually described by the term "Synergetic Computer" only those are considerd, which have been tested already in real world applications. We address issues on classification, rejection, generalization, generation of invariance properties, automated feature extraction, background separation as well as supervised learning. Special emphasis is given to performance measures, which can be derived from theoretical considerations and can serve as criterions for qualitative as well as quantitative comparison between SCs and more established classification algorithms. It is shown that SCs can separate arbitrary pattern distributions in feature space provided that the number of pixels is larger than the number of learned patterns.