Deteksi Balon Ucapan Pada Komik Jepang Dengan Convolutional Neural Network, Canny Edge Detection dan Run Length Smooth Algorithm
Keywords:
Website, Web Service, Ecommerce, Object Detection, Object Recognition, Haar Cascade Classifiers, SURF, Feature MatchAbstract
Comic is an entertainment media that is usually used to fill free time. Comics themselves are already very well known in the world, especially comics from Japan. Comics from Japan, commonly called Manga, have a high level of popularity. The proof is a lot of Manga that is translated into each country's language. Examples such as One Piece that has been circulating in 43 countries. Even so the translation process is quite long especially in Japanese translation.
This research can be used to accelerate the translation process by using CNN and Canny Edge Detection to detect speech balloons in Manga. The detection results are segmented and with the help of OCR to digitize Japanese characters. Then use copy-paste techniques in an online dictionary or online translator to find the meaning of letters that are not understood. Because searching for letters from a physical dictionary (book) takes more time.
The results of the research to segment the speech balloon from Manga were successful but to classify the image in the form of a speech balloon or not with CNN was unsuccessful. Researchers assume because the dataset created is small in number or a problem during pre-processing.References
[1] Dubray, D., & Laubrock, J. 2019. Deep CNN-based Speech
Balloon Detection and Segmentation for Comic Books. 2019
International Conference on Document Analysis and
Recognition (ICDAR). Sydney, Australia. 1237-1243.
https://doi.org/10.1109/ICDAR.2019.00200
[2] Fang, W., Ding, Y., Zhang, F., & Sheng, V. 2019. DOG: A
new background removal for object recognition from images.
Neurocomputing, 361, 85-91.
https://doi.org/10.1016/j.neucom.2019.05.095
[3] Kuboi, T. 2014. Element Detection in Japanese Comic Book
Panels. Thesis, California Polytechnic State University,
Computer Science, San Luis Obispo.
doi:10.15368/theses.2014.141
[4] Liliana, Budhi, G. S., & Hendra. 2010. Segmentasi Plat
Nomor Kendaraan Dengan Menggunakan Metode RunLength Smearing Algorithm (RLSA). Retrieved from:
https://www.researchgate.net/publication/277124943_Segme
ntasi_Plat_Nomor_Kendaraan_Dengan_Menggunakan_Meto
de_Run-Length_Smearing_Algorithm_RLSA
[5] Ogawa, T., Otsubo, A., Narita, R., Yusuke, M., Yamasaki,
T., & Aizawa , K. 2018. Object Detection for Comics using
Manga109 Annotations. arXiv:1803.08670v2. Retrieved
from https://arxiv.org/abs/1803.08670
[6] Rigaud, C., Burie, J.-C., & Ogier, J.-M. 2017. TextIndependent Speech Balloon Segmentation for Comics and
Manga. International Workshop on Graphics Recognition,
133-147. https://doi.org/10.1007/978-3-319-52159-6_10
[7] Simonyan, K., & Zisserman, A. 2015. Very Deep
Convolutional Networks for Large-scale Image Recognition.
arXiv:1409.1556v6. Retrieved from
https://arxiv.org/abs/1409.1556
[8] C. Szegedy et al., Going deeper with convolutions, 2015
IEEE Conference on Computer Vision and Pattern
Recognition (CVPR), 2015, Boston, MA. 1-9,
