| Presentation | 2011-12-16 Image Annotation Using Adapted Gaussian Mixture Model Yukihiro TSUBOSHITA, Noriji KATO, Masato OKADA, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In the present study, we focus on learning based automatic image annotation method using Gaussian mixture model (GMM) as a probabilistic model. In Supervised Multiclass Labeling (SML), which is a conventional image annotation method to use GMM, training samples assigned to each semantic label are collected separately, and each probabilistic model of semantic labels is trained independently. The number of training samples therefore varies a great deal according to the labels. Consequently, there is a problem of low performances of semantic labels that have a few training samples because of over fitting. In the present study, we propose to introduce a penalty term using training samples that is not confined by a particular semantic label when training each semantic label model. According to the proposed method, while each probabilistic model to semantic labels is trained independently, optimization of whole annotation system is achieved, and the over fitting of models of labels that have a few samples is suppressed. As the result of evaluation tests using a standard test collection for image annotation, we found the proposed method exhibited higher performance than the conventional SML, especially recall and N+. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Automatic Image Annotation / Machine Learning / Gaussian Mixture Model / Penalty Term |
| Paper # | PRMU2011-144 |
| Date of Issue |
| Conference Information | |
| Committee | PRMU |
|---|---|
| Conference Date | 2011/12/8(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Pattern Recognition and Media Understanding (PRMU) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Image Annotation Using Adapted Gaussian Mixture Model |
| Sub Title (in English) | |
| Keyword(1) | Automatic Image Annotation |
| Keyword(2) | Machine Learning |
| Keyword(3) | Gaussian Mixture Model |
| Keyword(4) | Penalty Term |
| 1st Author's Name | Yukihiro TSUBOSHITA |
| 1st Author's Affiliation | Research & Technology Group, Fuji Xerox Co., Ltd.() |
| 2nd Author's Name | Noriji KATO |
| 2nd Author's Affiliation | Research & Technology Group, Fuji Xerox Co., Ltd. |
| 3rd Author's Name | Masato OKADA |
| 3rd Author's Affiliation | Graduate School of Frontier Sciences, The University of Tokyo |
| Date | 2011-12-16 |
| Paper # | PRMU2011-144 |
| Volume (vol) | vol.111 |
| Number (no) | 353 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |