Character recognition by context-dependent transformations

This paper records an investigation of a character-recognition method applicable to hand-printed numerals. The technique employed may be characterised as a feature-detection method which utilises a decision tree to reduce the data to a decodable form. The data, consisting of lines and enclosures, are classified into labelled categories by a process which results in the formation of a one-dimensional symbol string. This string is then reduced to a decodable form by sets of context-dependent transformations. Data obtained by computer simulation are presented, indicating the ability of the method to recognise machine- and handprinted numerals.