Pattern recognition")?> In the past years automatic recognition of handwritten and typewritten documents lead to a high reduction of expenses in a vast number of industrial applications. Examples can be found in the analysis of credit transfer forms, credit card receipts and checks or in automatic sorting of posted items. Important features usually are high performance rates (several items per second) and at the same time low error rates. Thus both high-end computers and efficient, reliable character recognition algorithms are needed. In our institute we developed a flexible system for the analysis of forms by combining innovative character recognition with dynamic segmentation methods and the alignment with a given data base of possible input strings to yield high reliability.

The exploding development of personal computers now allows the application of pattern recognition programs even on standard PCs. Therefore innovative methods that were restricted to pure resaerch up to now are ready to be introduced into industrial application. The interaction of user requirements and basic research is expected to produce a new generation of pattern recognition software within the next years, that will fulfill the high needs of real-life systems.

Current work to improve the feasibility of such systems aim to increase their modularity and flexibility to allow adaptions to certain working conditions. Parameter adjustments are then necessary which require complex algorithmic knowledge and therefore increase costs significantly. Hence the development of self-learning systems is an important issue.

The pattern recognition group of the ZPR is involved in the following fields: