| Scope and method of study. The purpose of this study is to develop a prototype on-line handprinted Chinese character recognition system. A new approach to recognize the most frequently used Chinese characters (5,401 categories) is proposed. This new approach includes character representation, primitive stroke extraction, and fuzzy similarity measure from a physical input to the templates in the database. An initial template database is created for the experiments. At the end of the study, the performance of the system is tested based on the 18,000 samples collected from 20 writers. Initial template database and the database after trained are used in the experiments for computing recognition rates, respectively.; Findings and conclusions. Conventional schemes of stroke representation require detailed description of the selected features in order to portrait the characteristics of a stroke. This usually requires a large database file to store the entire Chinese character set. By quantifying selected local features, the proposed representation scheme requires less storage space than the conventional approach for storing similar amount of features. This results in less physical bytes to represent each character. Moreover, the time for retrieving a large and contiguous amount of information from the secondary storage is reduced. However, quantification of features causes slight ambiguity between characters. In the first stage of experiments, 2% of the samples are ambiguous (i.e., out of the first fifteen candidates list) and the recognition rate is 81% on the average. In the second stage of experiments, the recognition rate is improved with selective training from those characters that were not recognized as the first candidate in the first time around. The ambiguity among characters is reduced to 0.4% and the recognition rate is increased to 94.8%. Overall, the average computation time spent in the recognition process is 0.94 seconds/character for 133 reference characters and 1.1 seconds/characters for 163 reference characters. |
