Gipsy: Knowledge Based Surface Inspection

Despite of a great demand of automatic image analysis systems in the production industry, the spread of installed systems is still limited. Reasons for this are, among others, the high requirements of image analysis systems with respect to computational effort and accuracy, the high engineering effort for the development of image analysis applications, and thus high investment costs for image analysis systems. Another problem, which frequently occurs, is the lack of samples for the adaptation of classificators, which leads to unstable classificators. This paper describes the concept for a generic inspection system which follows the goals fast processing speed, stable and accurate classification, and low development effort. Essential points are the concept of ,,Abstract Sensors" and the integration of numerical adaptation and explicit knowledge about the task domain and the inspection system. The modular structure of the system allows easy adaptation to new image analysis applications.