Paper Abstract and Keywords |
Presentation |
2008-03-06 14:45
Introduce Batch-Learning Algorithm into the Contour Extraction Method DCDAM using Self-Organizing Map Fukuya Namba (Tottori Univ. of Environmental Studies), Takuya Ueta (GALAXY Inc.), Yasuaki Sumi (Tottori Univ. of Environmental Studies), Noboru Yabuki (Tsuyama National College of Tech.), Takao Tsukutani (Matsue National College of Tech.) CAS2007-121 SIP2007-196 CS2007-86 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
In this paper, we proposed the new contour line extraction method BL-DCDAM introduced the Batch-Learning SOM algorithm into DCDAM we have proposed. DCDAM can extract the contour line from the gray scale image by using self-organizing map (Self-Organizing Map: SOM). DCDAM training has adopted Sequential-Learning SOM algorithm. So in a case of polygonal target object it can extract contour line in some degree it is not so good result than we have expected depending on the sequence of presenting the input vector at the training process. To solve this problem, we have examined Batch-Learning SOM algorithm as a method which does not depend on the sequence of presenting the input vector. When the reference vector is updated, the trainings can be achieved both by updating the reference vector on the competitive layer and by getting the mean vector of the input vector group that exists in the reference vector included in neighborhood in the Voronoi region at Batch-Learning SOM algorithm. Then, in our research to introduce Batch-Learning SOM algorithm into the learning algorithm of DCDAM, we have developed BL-DCDAM, that is, the new method to adjust the degree of the update by putting the weight to each input vector (coordinate value of the pixel) according to the weighting function when this mean vector is calculated. As one example, we have experiments on the road sign as an extraction target. We have obtained an especially excellent result about a polygonal target object (road sign) compared with the Sequential-Learning SOM algorithm. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Contour Extraction / DCDAM / Self-Organizing Map(SOM) / Batch-Learning SOM Algorithm / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 107, no. 528, SIP2007-196, pp. 75-80, March 2008. |
Paper # |
SIP2007-196 |
Date of Issue |
2008-02-28 (CAS, SIP, CS) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
CAS2007-121 SIP2007-196 CS2007-86 |
|