Presentation 2003/3/7
Training Data Selection and Recognition Accuracy for Handwritten Characters using SVM
Sanae SHINOHARA, Yuji WAIZUMI, Nei KATO, Yoshiaki NEMOTO,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In this paper, we use SVM for investigating training data selection and recognition accuracy for handwritten characters. For the case of two-class pattern recognition, learning in SVM is formulated in Quadratic Programing as selecting classification boundary which maxxnizes a distance between two classes. Only training data lying near the classification boundary is used to decide the boundary. For such occasions it is difficult to classify test data, if data deviating from main training data cluster exist. In this paper, we propose a method for selecting and removing of data deviating from main distribution. We show an increase in recognition accuracy by experiment using the database ETL9B(similar charcters).
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Support Vector Machine / Unsupervised Support Vector Machine / handprinted character recognition / similar character / ETL9B
Paper # PRMU2002-256
Date of Issue

Conference Information
Committee PRMU
Conference Date 2003/3/7(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) Training Data Selection and Recognition Accuracy for Handwritten Characters using SVM
Sub Title (in English)
Keyword(1) Support Vector Machine
Keyword(2) Unsupervised Support Vector Machine
Keyword(3) handprinted character recognition
Keyword(4) similar character
Keyword(5) ETL9B
1st Author's Name Sanae SHINOHARA
1st Author's Affiliation Graduate School of Information Sciences, TOHOKU University()
2nd Author's Name Yuji WAIZUMI
2nd Author's Affiliation Graduate School of Information Sciences, TOHOKU University
3rd Author's Name Nei KATO
3rd Author's Affiliation Graduate School of Information Sciences, TOHOKU University
4th Author's Name Yoshiaki NEMOTO
4th Author's Affiliation Graduate School of Information Sciences, TOHOKU University
Date 2003/3/7
Paper # PRMU2002-256
Volume (vol) vol.102
Number (no) 708
Page pp.pp.-
#Pages 6
Date of Issue