| Project/Area Number |
16360199
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
System engineering
|
|---|
| Research Institution | Kobe University |
|---|
Principal Investigator |
ABE Shigeo Kobe University, Gradaute School of Science and Technology, Professor, 大学院・自然科学研究科, 教授 (50294195)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
OZAWA Seiichi Kobe University, Gradaute School of Science and Technology, Associate Professor, 大学院・自然科学研究科, 助教授 (70214129)
YOSHIMURA Motohide Siebold University of Nagasaki, Department of Info-Media, Lecturer, 国際情報学部, 講師 (60335461)
|
|---|
| Project Period (FY) |
2004 – 2005
|
| Project Status |
Completed (Fiscal Year 2005)
|
| Budget Amount *help |
¥15,100,000 (Direct Cost: ¥15,100,000)
Fiscal Year 2005: ¥7,600,000 (Direct Cost: ¥7,600,000)
Fiscal Year 2004: ¥7,500,000 (Direct Cost: ¥7,500,000)
|
|---|
| Keywords | Data Mining / Knowledge Acquisition / Support Vector Machines / Image Processing / Face Recognition / Feature Selection / パターン認識 / 音響診断 / クラスタリング / 特徴抽出 |
|---|
| Research Abstract |
We conducted research on knowledge acquisition and system development and got the following results :
1. Knowledge acquisition and system development under steady state environments
(1) We have developed a clustering method using support vector machines (SVMs) by dividing the image data in segments and feature selection method using SVM ensembles.
(2) We have developed an incremental training method that keeps only support vector candidates and demonstrated that the generalization ability was maintained while deleting training data.
2. Knowledge acquisition by data mining and system development under dynamic environments
(1) Dynamic feature space learning : we have developed incremental learning algorithms of Principal Component Analysis (PCA) and Kernel PCA (KPCA), and demonstrated that the feature selection was successfully carried out by adapting to the variation of data distributions.
(2) System development : We proposed a method to integrate the developed dynamic feature selection algor
… More
ithm into a classifier model, in which the k-nearest neighbor method was combined with a dynamic clustering algorithm, and a neural network model. We demonstrated that the proposed method enabled the classifier to conduct stable incremental learning, and that the developed system had excellent performance for not only bench-mark datasets but also facial recognition datasets. Although we tried to develop an incremental SVM system, it has not been completed yet. This is reserved as our future work.
3. Development of image segmentation system by data mining
(1)Development of clustering method for image segmentation : we have developed a basic system to detect and classify the uncertain object imagery and demonstrated that the detection and classification of crystal imagery was properly carried out using image database of protein crystallizations.
(2) Development of feature extraction method for image segmentation : we have developed a feature extraction method based on the multiresolution spectral histograms by wavelet transformation and demonstrated that the proposed feature is available for the image retrieval. Less
|
