Presentation 2013-12-13
A Study on Generation and Recognition of Dot Matrix Characters
Koji ENDO, Wataru OHYAMA, Tetsushi WAKABAYASHI, Fumitaka KIMURA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) We propose a camera-based OCR technique for dot-matrix characters which are widely used for printing information, such as expire date and production number, on products. The difficulty of recognition of dot-matrix character is due to variety of dot alignment and distance between dots even in the same character class. Furthermore, appearance of camera-captured dot-matrix character is easily changed when the characters are printed on rounded surface such as side of bottles. This appearance change sometimes affects the performance of recognition. In this report, we propose a generative learning method for dot-matrix characters. Generated dataset consists of camera captured actual dot-matrix characters, two types of dot-matrix character generated by our method and usual (non-dot-matrix) characters corrected for printed OCR. To address variety of character appearance due to characters are printed on round surface, three-dimensional rotation is applied on each character in the generated training dataset. We compared each condition of generating training dataset for recognition performance. In the experiment, grayscale gradient features and pseudo Bayes discriminant function are employed. Recognition accuracy on a dataset consists of 38 classes (2030 samples) actual camera-captured dot-matrix characters achieves 97.88%.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) dot matrix character recognition / generative learning / 3d-rotation
Paper # PRMU2013-90
Date of Issue

Conference Information
Committee PRMU
Conference Date 2013/12/5(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) A Study on Generation and Recognition of Dot Matrix Characters
Sub Title (in English)
Keyword(1) dot matrix character recognition
Keyword(2) generative learning
Keyword(3) 3d-rotation
1st Author's Name Koji ENDO
1st Author's Affiliation Information and Computer Science, Graduate School of Engineering, Mie University()
2nd Author's Name Wataru OHYAMA
2nd Author's Affiliation Information and Computer Science, Graduate School of Engineering, Mie University
3rd Author's Name Tetsushi WAKABAYASHI
3rd Author's Affiliation Information and Computer Science, Graduate School of Engineering, Mie University
4th Author's Name Fumitaka KIMURA
4th Author's Affiliation Information and Computer Science, Graduate School of Engineering, Mie University
Date 2013-12-13
Paper # PRMU2013-90
Volume (vol) vol.113
Number (no) 346
Page pp.pp.-
#Pages 5
Date of Issue