Paper Abstract and Keywords |
Presentation |
2010-10-23 15:20
Parametric control in a region-based coupled MRF model for coarse image region segmentation towards its VLSI implementation Haichao Liang, Kenji Matsuzaka, Kazuki Nakada (Kyushu Inst of Tech), Masato Okada (Univ. of Tokyo/RIKEN), Takashi Morie (Kyushu Inst of Tech) NC2010-48 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
For visual image processing, coupled Markov random field (MRF) models have been proposed, and they are classified into boundary-based and region-based models, according to the interactions between the nodes corresponding to the image pixels. The original region-based MRF models have a local-minimum problem because of its discrete hidden variables, but an improved model has been proposed by introducing phase variables as continuous hidden variables to solve this problem. We have already modified this model by introducing piecewise linear (PWL) functions to simplify the model towards its VLSI implementation. In this paper, we investigate the controllability of the parameters of PWL functions, and evaluate the performance in the application of coarse image region segmentation. We have verified that the segmentation tuning and the processing speed can be improved by optimizing of the parameters. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
region-based coupled MRF model / boundary-based coupled MRF model / coarse image region segmentation / phase variables / piecewise linear functions / parametric control / / |
Reference Info. |
IEICE Tech. Rep., vol. 110, no. 246, NC2010-48, pp. 35-40, Oct. 2010. |
Paper # |
NC2010-48 |
Date of Issue |
2010-10-16 (NC) |
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 |
NC2010-48 |
|