Paper Title
Onomatopoeia Classification using Two-Stage CNN

Abstract
Onomatopoeia is found in comics, which is one of the important pieces of information that make up the manga. Studies are being conducted to automatically retrieve various important information from comics using deep learning and other methods. The purpose of our study is to infer onomatopoeia from the images of comics. Since onomatopoeia is complex, our study adopts 2 stages of CNN to improve the accuracy of onomatopoeia classification. In the first stage, CNN finds objects commonly seen in comics, such as faces, hands, or feet. In the second stage, CNN classifies onomatopoeia based on the object relationships detected in the first stage. We evaluate our proposed method by experimentation. The results showed that the proposed method could classify five types of onomatopoeia with 87% accuracy. It was also found that if the first stage CNN could be optimized, the classification could be done with 96% accuracy. Keywords - Comics, Manga, Cartoon, Convolutional Neural Network, Deep learning, Onomatopoeia