Japanese Character Recognition with Microstrip Line Networks

Abstract

Character recognition, a topical field seen from the large number of practical applications, allows the generation of characters in electronic media from printed documents or images. The Japanese language is one of the most difficult language because of the very large number of characters or classes of objects to be identified, and many recognition systems have been proposed. A new way of recognizing Japanese printed characters is presented in this paper. The proposed method is based on the modeling with microstrip lines of the character topological structure and the analysis using microwave engineering methods. The spectral distribution of the resonance frequencies is used as a local descriptor in the recognition process. A deep convolutional recurrent network is used for the classification step. The proposed method ensures good accuracy in the Japanese character recognition process using the lowest number of descriptors among the methods described in the literature. The character recognition method has been tested by analyzing the first ten resonant frequencies of the five Japanese vowels, and 99.9% character recognition has been provided for 20% dispersion of the microstrip line segment lengths.