| Presentation | 2022-12-15 DN4C Toshikazu Wada, Koji Kamma, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Color/texture based image segmentation can be widely applied to the images for product and/or medical inspection, remote-sensing, and so on. The accuracy and the robustness against the image noise are drastically improved recently by the introduction of Deep Neural Networks: DNNs. However, depending on the products, lesions, ground surface or vegetation to be examined, training images are quite different, and hence, people cannot enjoy the accurate segmentation system without training the system by themselves. Most of the previous training framework is batch style: providing huge amount of image and annotation pairs and training to minimize the loss function defined by these pairs. On the other hand, if we provide incomplete annotations, like scribbles on small number of images, the system still can learn incomplete segmentation rules. By adding new annotations, the system can learn better rules. By iterating annotation, learning, and segmentation, we can realize “interactive segmentation”, which reduces the annotation tasks. This report proposes a system for interactive segmentation system DN4C by combining DNN and Nearest Neighbor Classifier: NNC. In the pure DNN image segmentation system, the feature distributions at the layer before the output must be linearly separable for producing compatible result with annotations. DN4C have no such limitations. It implies shallower network layers are sufficient, and hence, the faster learning is possible. In the experiment, we examined interactive segmentation is possible using the prototype system. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Image Segmentation / Deep Neural Network / Nearest Neighbor Classifier / Human In the Loop |
| Paper # | PRMU2022-35 |
| Date of Issue | 2022-12-08 (PRMU) |
| Conference Information | |
| Committee | PRMU |
|---|---|
| Conference Date | 2022/12/15(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Toyama International Conference Center |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Seiichi Uchida(Kyushu Univ.) |
| Vice Chair | Takuya Funatomi(NAIST) / Mitsuru Anpai(Denso IT Lab.) |
| Secretary | Takuya Funatomi(CyberAgent) / Mitsuru Anpai(Univ. of Tokyo) |
| Assistant | Nakamasa Inoue(Tokyo Inst. of Tech.) / Yasutomo Kawanishi(Riken) |
| 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) | DN4C |
| Sub Title (in English) | An Interactive Image Segmentation System Combining DNN And Nearest Neighbor Classifier |
| Keyword(1) | Image Segmentation |
| Keyword(2) | Deep Neural Network |
| Keyword(3) | Nearest Neighbor Classifier |
| Keyword(4) | Human In the Loop |
| 1st Author's Name | Toshikazu Wada |
| 1st Author's Affiliation | Wakayama University(Wakayama University) |
| 2nd Author's Name | Koji Kamma |
| 2nd Author's Affiliation | Wakayama University(Wakayama University) |
| Date | 2022-12-15 |
| Paper # | PRMU2022-35 |
| Volume (vol) | vol.122 |
| Number (no) | PRMU-314 |
| Page | pp.pp.19-24(PRMU), |
| #Pages | 6 |
| Date of Issue | 2022-12-08 (PRMU) |