Measurement of static and dynamic calligraphic features

The application of optical character recognition techniques to calligraphic text is examined. Calligraphic material was scanned with a conventional image scanner and the resulting images were segmented into lines of text consisting of individual symbols or characters. A novel segmentation process yielded identifying features which were fed to a neural network processor. After a training period. the system is capable of recognizing individual symbols when a new sample of text is presented. The input to the neural network consists of features such as the number of changes of direction and junctions or crossings in a segment of writing. The segmentation process also provides information on the complexity and line width involved, as well as on the presence of islands, such as dots and isolated strokes which characterize the individual symbols. As an example, Hebrew calligraphic text was segmented. The segmentation process yielded a large number of identifying features, 11 of which were used as an input to the neural network