There are many reports of medical errors mainly caused by the similarity of drug names in Japanese such as ‘_??__??__??__??__??_’ (Amaryl) and ‘_??__??__??__??__??_’ (Almarl). In order to prevent such errors, one of the authors studied methods of measuring the similarity of drug names and implemented a search engine to find pairs of similar medicine names. However, the existing methods only evaluate sequences of common substrings. In addition, although some methods took the similarity of character (letter) shapes into consideration, they assumed that the similarity of each character pair is given, usually by hand. In order to define the objective similarity index of medicine names, it is also necessary to measure the similarity of character shape objectively. In this study, we propose a method to evaluate character similarity by image processing and apply the method to the medicine name similarity that we propose.
