A Prior Study on the Improvement of the Recognition Rate of Medieval Korean Using Class Compression and Division in Object Detection
Yeongseo Ha, Hoseok Hwang, Min-Jun Kim, Changjun Lee, Jaechang Shim · Journal of Korea Multimedia Society · 2023
Text recognition is an important method for converting documents into digital format by allowing the extraction of text from images. While optical character recognition (OCR) technology has been developed extensively over the years, it has lower performance when it comes to recognizing Hangeul (the Korean alphabet) due to its larger number of character types compared to other languages. In this study, we propose a technique for object detection that incorporates class compression and division to improve the accuracy of Hangeul text recognition. We conducted experiments comparing models with and without compression and division, as well as OCR, a well-established text recognition method. The results showed that the model utilizing compression and division achieved the highest accuracy, even with the smallest amount of data.