SVM Based Hiragana and Katakana Recognition Algorithm with Neural Network Based Segmentation

Abstract

A Japanese writing system, unlike the European system, is complex. It contains three types of signs: hiragana, katakana and Kanji. For daily use, more than 2000 characters are used, and each symbol can consist of 6 or more strokes. That is why it seems possible to recognise each sign by using a similar approach to fingerprint recognition. Authors are using the minutiae-finding algorithm to find three types of characteristic points. For preprocessing and classification, machine learning algorithms were used. The presented system uses the image of a single sign as an input.