A new character recognition algorithm with a capacity for easy learning has been developed. This algorithm incorporates two different kinds of processes: a triadic decision tree and a pattern matching method. Its significant features are that (1) a user can generate a reliable recognition dictionary by using only a limited number of characters, (2) the user can easily add new categories to the recognition dictionary or modify it for matching to actual patterns, and (3) the recognition speed is fast. We have also developed a unique OCR system on a personal computer, adopting this recognition algorithm. The system has several unique functions: (1) A user does not need to prepare special sample sheets on which all categories of unknown font are printed in alphabetical order. New dictionaries can be generated by using general texts. (2) It provides him with a concurrent spelling correction capability in order to facilitate learning and recognition. We conducted tests on the system to investigate the effect of the number of learning times and the quantity of sample data on the recognition accuracy. In our tests, the dictionary that was generated from eight data per category (the original and the first to third successive copies of two printed sets) attained a recognition accuracy of 99.6%. In another test, the recognition accuracy of the test documents was improved from 92.3% to 99.5% by modifying a different font dictionary four times with the test font data. A recognition speed of 120 char/sec was achieved by our software program on an IBM PS/55 personal computer.

An integrated pen interface system was developed to allow effective Japanese text entry. It consists of sub-systems for handwriting recognition, contextual post-processing, and enhanced Kana-to-Kanji conversion. The recognition sub-system uses a hybrid algorithm consisting of a pattern matcher and a neural network discriminator. Special care was taken to improve the recognition of non-Kanji and simple Kanji characters frequently used in fast data entry. The post-processor predicts consecutive characters on the basis of bigrams modified by the addition of parts of speech and substitution of macro characters for Kanji characters. A Kana-to Kanji conversion method designed for ease of use with a pen interface has also been integrated into the system. In an experiment in which 2,900 samples of Kanji and non-Kanji characters were obtained from 20 subjects, it was observed that the original recognition accuracy of 83.7% (the result obtained by using the pattern matching recognizer) was improved to 90.7% by adding the neural network discriminator, and that it was further improved to 94.4% by adding the post-processor. The improved recognition accuracy for non-Kanji characters was particularly marked.