Parallelizing OCR

The authors describe how to parallelize optical character recognition (OCR) in EFECT, a parallel dataflow-oriented graphical programming language. OCR-systems are judged on two criteria: speed and reliability. To obtain both, a simple, reliable, and fast feature extraction method has been chosen to be executed in a parallel environment. As the EFECT environment has not been completed yet, the algorithms have been implemented and tested on a personal computer, ignoring the speed up when using a parallel architecture. The main features described are: feature extraction, storing knowledge in a neural network, and implementing a parallel learning and recognition algorithm in EFECT. Isolating words and symbols from the raw data is only roughly described. Some test runs and their results are presented and discussed. The raw data were produced without a scanner using a pixel-oriented painting program. The character isolation was done interactively