Integration of Recognition Algorithms for Self-Learning Recognition System
| Project/Area Number |
15500097
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Perception information processing/Intelligent robotics
|
|---|
| Research Institution | Tohoku University |
|---|
Principal Investigator |
ASO Hirotomo Tohoku University, Graduate School of Engineering, Professor, 大学院・工学研究科, 教授 (10005522)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
OMACHI Shinichiro Tohoku University, Graduate School of Engineering, Associate Professor, 大学院・工学研究科, 助教授 (30250856)
GOTO Hideaki Tohoku University, Information Synergy Center, Associate Professor, 情報シナジーセンター, 助教授 (40271879)
IWAMURA Masakazu Osaka Prefecture University, Graduate School of Engineering, Research Associate, 大学院・工学研究科, 助手 (80361129)
|
|---|
| Project Period (FY) |
2003 – 2005
|
| Project Status |
Completed (Fiscal Year 2005)
|
| Budget Amount *help |
¥3,700,000 (Direct Cost: ¥3,700,000)
Fiscal Year 2005: ¥1,300,000 (Direct Cost: ¥1,300,000)
Fiscal Year 2004: ¥1,200,000 (Direct Cost: ¥1,200,000)
Fiscal Year 2003: ¥1,200,000 (Direct Cost: ¥1,200,000)
|
|---|
| Keywords | character recognition / document recognition system / learning recognition system / recognition algorithm integration / reliability / figure table recognition / statistical pattern recognition / feature vector |
|---|
| Research Abstract |
To construct a document recognition system with brushing-up ability by self-learning, we have basic results.
1.Integration of recognition algorithms : (1)To describe category clusters in the feature space, clusters of hypersphere, hyperellipse, parallelepiped and convex polyhedron are introduced and the base of analysis of cluster structure is given. By proposing a reliability measure, a method of integrating recognition results is given. A self-learning recognition system is proposed. (2)Some basic properties on integration methods using ensemble learning and on multiple classifiers system are investigated.
2.Recognition theory and system : (1)It is known that ICA is effective for feature extraction and PCA is effective for recognition. (2)The effect of positioning the origin for subspace method is investigated. (3)A method to analyze a bias of predictive distribution in pattern recognition and a method correcting the bias are proposed. (4)A platform for web-based OCR systems are constructed with function to search servers.
3.Layout analysis : (1)A recognition algorithm of line graph images in documents by tracing connected components and a speedup method of tracing are proposed. (2)A screen pattern removal method is proposed which is useful for character pattern extraction from high-resolution color document images. (3)String extraction methods from low-resolution scene images by digital camera or video are proposed. (4) A principle of voting to local feature points is proposed and it is applied to isolated character recognition.
|
Report
(4 results)
Research Products
(55 results)
