2018 Fiscal Year Final Research Report
Study on mechanism of feature extraction based on GDS projection
Project/Area Number |
16H02842
|
Research Category |
Grant-in-Aid for Scientific Research (B)
|
Allocation Type | Single-year Grants |
Section | 一般 |
Research Field |
Perceptual information processing
|
Research Institution | University of Tsukuba |
Principal Investigator |
Fukui Kazuhiro 筑波大学, システム情報系, 教授 (40375423)
|
Co-Investigator(Kenkyū-buntansha) |
日野 英逸 統計数理研究所, モデリング研究系, 准教授 (10580079)
小林 匠 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 主任研究員 (30443188)
|
Research Collaborator |
Maki Atsuto スウェーデン王立工科大学
Xue Jing-Hao ユニバーシティ・カレッジ・ロンドン, 統計科学科
|
Project Period (FY) |
2016-04-01 – 2019-03-31
|
Keywords | 画像認識 / 部分空間法 / 差分部分空間 / 一般化差分部分空間 / 特徴抽出 |
Outline of Final Research Achievements |
This research discusses a new type of discriminant analysis based on the orthogonal projection of data onto a generalized difference subspace (GDS). GDS projection also works as a discriminant feature extraction through a similar mechanism as the Fisher discriminant analysis (FDA) only by changing the dimension of the GDS. The direct consideration of the connection between GDS projection and FDA is difficult due to the significant difference in their formulations. To avoid the difficulty, we first introduce geometrical Fisher discriminant analysis (gFDA), which is a discriminant analysis based on a simplified Fisher criteria. Next, we prove that gFDA is equivalent to GDS projection with a small correction term. This equivalence ensures GDS projection to inherit the discriminant ability via gFDA from FDA. Extensive experiments show that gFDA and GDS projection have equivalent or higher performance than the original FDA and its extensions.
|
Free Research Field |
パターン認識
|
Academic Significance and Societal Importance of the Research Achievements |
GDS射影の有効性は,商用顔認識システム(東芝FacePass)等での実運用を通して十分に検証済みである.しかしながら,その有効性の根源であるGDSの幾何構造については未解明の部分が多く,標準的な基盤理論として認知されるまでには至っていなかった.本研究において,GDS射影の幾何構造が完全に解明され,その有効性が多角的に検証された.これによりGDS射影は安心して使える特徴抽出理論として世界的に認知されるようになると期待される.
|