Hiragana and Katakana Minutiae based Recognition System

The Japanese writing system is unique due to the number of characters employed and the methods used to write words.It consists of three different 'alphabets', which may result in the methods used to process Latin script not being sufficient to obtain satisfactory results when attempting to apply them to a recognition of the Japanese script.The authors present an algorithm based on minutiae, i.e., feature points, to recognise the hiragana and katakana characters.A method using image processing algorithms is compared with a method using a neural network for the purpose of automating this process.Based on the distribution and type of minutiae, vectors of features have been created to recognise 96 different characters.The authors conducted a study showing the effect of the chosen segmentation method on the accuracy of the character recognition.The proposed solution has achieved a maximum accuracy at the level of 65.2%.