| Presentation | 2018-12-13 Candidate Reduction Method Using Hierarchical Overlapping Clustering and Convolutional Neural Network for Fast Chinese Character Recognition Soichi Tashima, Hideaki Goto, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Along with the widespread of the mobile devices equipped with cameras, many applications using the camera function have appeared and the demands for optical character recognition is growing. Although the nearest neighbor search based on the linear search with feature vectors can be easily implemented, its problem is that the computational cost increases linearly by the dimensionality of feature vectors and the number of classes. It is necessary to reduce the computational cost in handwritten character recognition, especially in case of Chinese characters that have complex structures and consist of thousands of classes, on such devices with limited performance. In this paper, we discuss the parameter settings for the candidate reduction system based on the hierarchical overlapping clustering and CNN-based feature extractor. The experimental results show that CNN-based feature extractor enables the system to reduce more candidates and contributes to 2.56 times faster recognition with 0.51% accuracy drop compared with our former method. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Optical Character Recognition / Handwritten Chinese Character Recognition / Hierarchical Overlapping Clustering / Convolutional Neural Network / Nearest Neighbor Search |
| Paper # | PRMU2018-77 |
| Date of Issue | 2018-12-06 (PRMU) |
| Conference Information | |
| Committee | PRMU |
|---|---|
| Conference Date | 2018/12/13(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Shinichi Sato(NII) |
| Vice Chair | Yoshihisa Ijiri(Omron) / Toru Tamaki(Hiroshima Univ.) |
| Secretary | Yoshihisa Ijiri(NEC) / Toru Tamaki(Osaka Univ.) |
| Assistant | Go Irie(NTT) / Yoshitaka Ushiku(OSX) |
| Paper Information | |
| Registration To | Technical Committee on Pattern Recognition and Media Understanding |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Candidate Reduction Method Using Hierarchical Overlapping Clustering and Convolutional Neural Network for Fast Chinese Character Recognition |
| Sub Title (in English) | |
| Keyword(1) | Optical Character Recognition |
| Keyword(2) | Handwritten Chinese Character Recognition |
| Keyword(3) | Hierarchical Overlapping Clustering |
| Keyword(4) | Convolutional Neural Network |
| Keyword(5) | Nearest Neighbor Search |
| 1st Author's Name | Soichi Tashima |
| 1st Author's Affiliation | Tohoku University(Tohoku Univ.) |
| 2nd Author's Name | Hideaki Goto |
| 2nd Author's Affiliation | Tohoku University(Tohoku Univ.) |
| Date | 2018-12-13 |
| Paper # | PRMU2018-77 |
| Volume (vol) | vol.118 |
| Number (no) | PRMU-362 |
| Page | pp.pp.13-18(PRMU), |
| #Pages | 6 |
| Date of Issue | 2018-12-06 (PRMU) |