IEICE Technical Committee Submission System
Conference Paper's Information		 Online Proceedings
[Sign in]
Tech. Rep. Archives
 Go Top Page Go Previous   [Japanese] / [English] 

Paper Abstract and Keywords
Presentation 2023-02-28 10:40
Image reconstruction with a diffusion model for robust image classification against unknown degradation
Teruaki Akazawa (Tokyo Metro. Univ.), Yuma Kinoshita (Tokai Univ.), Hitoshi Kiya (Tokyo Metro. Univ.) EA2022-83 SIP2022-127 SP2022-47
Abstract (in Japanese) (See Japanese page) 
(in English) This paper presents an image reconstruction method with a diffusion model for robust image classification against image degradation due to unknown factors. In general, image classification models based on deep neural networks are not robust against degradation such as rain or blur which is not considered in the training phase. There are two approaches for addressing this problem: including degraded images in training data for classification models, or removing such degradation with image restoration methods. Image restoration is a task that removes the degradation from measurements and restores original clean images without degradation as accurately as possible. However, conventional image restoration methods assume that degradation types such as rain are known, and detailed modeling against the degradation factor is performed. In contrast, by reconstructing degraded images with a diffusion model, the proposed scheme focuses on recovering only important features for image classification, not exactly restoring original images. Therefore, the proposed scheme can maintain the accuracy of image classification even under the challenging constraint where degradation factors are unknown. In experiments with the CIFAR-10C dataset, the effectiveness of the proposed method is shown.
Keyword (in Japanese) (See Japanese page) 
(in English) Diffusion Model / SDEdit / Image Reconstruction / Unknown Degradation / Image Classification / / /  
Reference Info. IEICE Tech. Rep., vol. 122, no. 388, SIP2022-127, pp. 49-54, Feb. 2023.
Paper # SIP2022-127 
Date of Issue 2023-02-21 (EA, SIP, SP) 
ISSN 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 EA2022-83 SIP2022-127 SP2022-47

Conference Information
Committee SP IPSJ-SLP EA SIP  
Conference Date 2023-02-28 - 2023-03-01 
Place (in Japanese) (See Japanese page) 
Place (in English)  
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To SIP 
Conference Code 2023-02-SP-SLP-EA-SIP 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Image reconstruction with a diffusion model for robust image classification against unknown degradation 
Sub Title (in English)  
Keyword(1) Diffusion Model  
Keyword(2) SDEdit  
Keyword(3) Image Reconstruction  
Keyword(4) Unknown Degradation  
Keyword(5) Image Classification  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Teruaki Akazawa  
1st Author's Affiliation Tokyo Metropolitan University (Tokyo Metro. Univ.)
2nd Author's Name Yuma Kinoshita  
2nd Author's Affiliation Tokai University (Tokai Univ.)
3rd Author's Name Hitoshi Kiya  
3rd Author's Affiliation Tokyo Metropolitan University (Tokyo Metro. Univ.)
4th Author's Name  
4th Author's Affiliation ()
5th Author's Name  
5th Author's Affiliation ()
6th Author's Name  
6th Author's Affiliation ()
7th Author's Name  
7th Author's Affiliation ()
8th Author's Name  
8th Author's Affiliation ()
9th Author's Name  
9th Author's Affiliation ()
10th Author's Name  
10th Author's Affiliation ()
11th Author's Name  
11th Author's Affiliation ()
12th Author's Name  
12th Author's Affiliation ()
13th Author's Name  
13th Author's Affiliation ()
14th Author's Name  
14th Author's Affiliation ()
15th Author's Name  
15th Author's Affiliation ()
16th Author's Name  
16th Author's Affiliation ()
17th Author's Name  
17th Author's Affiliation ()
18th Author's Name  
18th Author's Affiliation ()
19th Author's Name  
19th Author's Affiliation ()
20th Author's Name  
20th Author's Affiliation ()
21st Author's Name  
21st Author's Affiliation ()
22nd Author's Name  
22nd Author's Affiliation ()
23rd Author's Name  
23rd Author's Affiliation ()
24th Author's Name  
24th Author's Affiliation ()
25th Author's Name  
25th Author's Affiliation ()
26th Author's Name / /
26th Author's Affiliation ()
()
27th Author's Name / /
27th Author's Affiliation ()
()
28th Author's Name / /
28th Author's Affiliation ()
()
29th Author's Name / /
29th Author's Affiliation ()
()
30th Author's Name / /
30th Author's Affiliation ()
()
31st Author's Name / /
31st Author's Affiliation ()
()
32nd Author's Name / /
32nd Author's Affiliation ()
()
33rd Author's Name / /
33rd Author's Affiliation ()
()
34th Author's Name / /
34th Author's Affiliation ()
()
35th Author's Name / /
35th Author's Affiliation ()
()
36th Author's Name / /
36th Author's Affiliation ()
()
Speaker Author-1 
Date Time 2023-02-28 10:40:00 
Presentation Time 20 minutes 
Registration for SIP 
Paper # EA2022-83, SIP2022-127, SP2022-47 
Volume (vol) vol.122 
Number (no) no.387(EA), no.388(SIP), no.389(SP) 
Page pp.49-54 
#Pages 6 
Date of Issue 2023-02-21 (EA, SIP, SP) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan