| Presentation | 2023-03-14 Generation and inpainting of Kuzushiji image data using Boltzmann Machines Hiroki Ikoma, Mauricio Bermudez, Minho Lee, Kazushi Ikeda, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | It is almost impossible for the average person to read Kuzushiji today. For this reason, there is a need to develop tools for deciphering and translating collapsed characters using machine learning and other methods. On the other hand, the frequency of occurrence of characters is biased and character images are sometimes missing, making it difficult to collect training data and requiring data augmentation methods. We proposed data augmentation by generating and repairing corrupted characters using a Boltzmann machine (BM) and conducted computational experiments. Although the repair process was not stable, we were able to generate the Kuzushiji "no" using a BM that had learned about 450 images. Because generating the same Kuzushiji "no" using GAN required about 4700 training data per character, the use of BM has the potential to dramatically reduce the number of images required for data augmentation for Kuzushiji. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Kuzushiji / Data Augmentation / Boltzmann Machines / Generative Models / Image Inpainting |
| Paper # | NC2022-108 |
| Date of Issue | 2023-03-06 (NC) |
| Conference Information | |
| Committee | NC / MBE |
|---|---|
| Conference Date | 2023/3/13(3days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | The Univ. of Electro-Communications |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Brain architecture, General |
| Chair | Hiroshi Yamakawa(Univ of Tokyo) / Junichi Hori(Niigata Univ.) |
| Vice Chair | Hirokazu Tanaka(Tokyo City Univ.) / Hisashi Yoshida(Kinki Univ.) |
| Secretary | Hirokazu Tanaka(NTT) / Hisashi Yoshida(NICT) |
| Assistant | Yoshimasa Tawatsuji(Waseda Univ.) / Tomoki Kurikawa(KMU) / Emi Yuda(Tohoku Univ) / Miki Kaneko(Osaka Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Neurocomputing / Technical Committee on ME and Bio Cybernetics |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Generation and inpainting of Kuzushiji image data using Boltzmann Machines |
| Sub Title (in English) | |
| Keyword(1) | Kuzushiji |
| Keyword(2) | Data Augmentation |
| Keyword(3) | Boltzmann Machines |
| Keyword(4) | Generative Models |
| Keyword(5) | Image Inpainting |
| 1st Author's Name | Hiroki Ikoma |
| 1st Author's Affiliation | Nara Institute of Science and Technology(NAIST) |
| 2nd Author's Name | Mauricio Bermudez |
| 2nd Author's Affiliation | Kyungpook National University(KNU) |
| 3rd Author's Name | Minho Lee |
| 3rd Author's Affiliation | Kyungpook National University(KNU) |
| 4th Author's Name | Kazushi Ikeda |
| 4th Author's Affiliation | Nara Institute of Science and Technology(NAIST) |
| Date | 2023-03-14 |
| Paper # | NC2022-108 |
| Volume (vol) | vol.122 |
| Number (no) | NC-425 |
| Page | pp.pp.94-98(NC), |
| #Pages | 5 |
| Date of Issue | 2023-03-06 (NC) |