Adversarial Attacks against Blind Image Quality Assessment Models

Jari Korhonen; Junyong  You

doi:10.1145/3552469.3555715

Adversarial Attacks against Blind Image Quality Assessment Models

Jari Korhonen, Junyong You

Research output: Chapter in Book/Report/Conference proceeding › Published conference contribution

8 Citations (Scopus)

Abstract

Several deep models for blind image quality assessment (BIQA) have been proposed during the past few years, with promising results on standard image quality datasets. However, generalization of BIQA models beyond the standard content remains a challenge. In this paper, we study basic adversarial attack techniques to assess the robustness of representative deep BIQA models. Our results show that adversarial images created for a simple substitute BIQA model (i.e. white-box scenario) are transferable as such and able to deceive also several other more complex BIQA models (i.e. black-box scenario). We also investigated some basic defense mechanisms. Our results indicate that re-training BIQA models with a dataset augmented with adversarial images improves robustness of several models, but at the cost of decreased quality prediction accuracy on genuine images.

Original language	English
Title of host publication	QoEVMA '22
Subtitle of host publication	Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications
Publisher	Association for Computing Machinery
Number of pages	9
ISBN (Electronic)	978-1-4503-9499-4
DOIs	https://doi.org/10.1145/3552469.3555715
Publication status	Published - 14 Oct 2022
Event	2nd Workshop on Quality of Experience in Visual Multimedia Applications (QoEVMA) at ACM Multimedia - Lisbon, Portugal Duration: 10 Oct 2022 → 14 Oct 2022 https://2022.acmmm.org/

Publication series

Name	QoEVMA 2022 - Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications

Conference

Conference	2nd Workshop on Quality of Experience in Visual Multimedia Applications (QoEVMA) at ACM Multimedia
Country/Territory	Portugal
City	Lisbon
Period	10/10/22 → 14/10/22
Internet address	https://2022.acmmm.org/

Bibliographical note

Funding Information:
This work was supported in part by Natural Science Foundation of China under grant 61772348, Guangdong ”Pearl River Talent Recruitment Program” under Grant 2019ZT08X603, and Shenzhen Technology R&D Fund under Grant 202008121558110.

Keywords

Image quality assessment
Quality of Experience
Adversarial attacks
Adversarial defenses

Access to Document

10.1145/3552469.3555715Licence: Unspecified

Cite this

Korhonen, J., & You, J. (2022). Adversarial Attacks against Blind Image Quality Assessment Models. In QoEVMA '22: Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications (QoEVMA 2022 - Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications). Association for Computing Machinery. https://doi.org/10.1145/3552469.3555715

Adversarial Attacks against Blind Image Quality Assessment Models. / Korhonen, Jari; You, Junyong .
QoEVMA '22: Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications. Association for Computing Machinery, 2022. (QoEVMA 2022 - Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications).

Research output: Chapter in Book/Report/Conference proceeding › Published conference contribution

Korhonen, J & You, J 2022, Adversarial Attacks against Blind Image Quality Assessment Models. in QoEVMA '22: Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications. QoEVMA 2022 - Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications, Association for Computing Machinery, 2nd Workshop on Quality of Experience in Visual Multimedia Applications (QoEVMA) at ACM Multimedia , Lisbon, Portugal, 10/10/22. https://doi.org/10.1145/3552469.3555715

Korhonen J, You J. Adversarial Attacks against Blind Image Quality Assessment Models. In QoEVMA '22: Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications. Association for Computing Machinery. 2022. (QoEVMA 2022 - Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications). Epub 2022 Oct 10. doi: 10.1145/3552469.3555715

@inproceedings{db84cce857ca4c5692143c8f2be8445b,

title = "Adversarial Attacks against Blind Image Quality Assessment Models",

abstract = "Several deep models for blind image quality assessment (BIQA) have been proposed during the past few years, with promising results on standard image quality datasets. However, generalization of BIQA models beyond the standard content remains a challenge. In this paper, we study basic adversarial attack techniques to assess the robustness of representative deep BIQA models. Our results show that adversarial images created for a simple substitute BIQA model (i.e. white-box scenario) are transferable as such and able to deceive also several other more complex BIQA models (i.e. black-box scenario). We also investigated some basic defense mechanisms. Our results indicate that re-training BIQA models with a dataset augmented with adversarial images improves robustness of several models, but at the cost of decreased quality prediction accuracy on genuine images.",

keywords = "Image quality assessment, Quality of Experience, Adversarial attacks, Adversarial defenses",

author = "Jari Korhonen and Junyong You",

note = "Funding Information: This work was supported in part by Natural Science Foundation of China under grant 61772348, Guangdong ”Pearl River Talent Recruitment Program” under Grant 2019ZT08X603, and Shenzhen Technology R&D Fund under Grant 202008121558110. ; 2nd Workshop on Quality of Experience in Visual Multimedia Applications (QoEVMA) at ACM Multimedia ; Conference date: 10-10-2022 Through 14-10-2022",

year = "2022",

month = oct,

day = "14",

doi = "10.1145/3552469.3555715",

language = "English",

series = "QoEVMA 2022 - Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications",

publisher = "Association for Computing Machinery",

booktitle = "QoEVMA '22",

url = "https://2022.acmmm.org/",

}

TY - GEN

T1 - Adversarial Attacks against Blind Image Quality Assessment Models

AU - Korhonen, Jari

AU - You, Junyong

N1 - Funding Information: This work was supported in part by Natural Science Foundation of China under grant 61772348, Guangdong ”Pearl River Talent Recruitment Program” under Grant 2019ZT08X603, and Shenzhen Technology R&D Fund under Grant 202008121558110.

PY - 2022/10/14

Y1 - 2022/10/14

N2 - Several deep models for blind image quality assessment (BIQA) have been proposed during the past few years, with promising results on standard image quality datasets. However, generalization of BIQA models beyond the standard content remains a challenge. In this paper, we study basic adversarial attack techniques to assess the robustness of representative deep BIQA models. Our results show that adversarial images created for a simple substitute BIQA model (i.e. white-box scenario) are transferable as such and able to deceive also several other more complex BIQA models (i.e. black-box scenario). We also investigated some basic defense mechanisms. Our results indicate that re-training BIQA models with a dataset augmented with adversarial images improves robustness of several models, but at the cost of decreased quality prediction accuracy on genuine images.

AB - Several deep models for blind image quality assessment (BIQA) have been proposed during the past few years, with promising results on standard image quality datasets. However, generalization of BIQA models beyond the standard content remains a challenge. In this paper, we study basic adversarial attack techniques to assess the robustness of representative deep BIQA models. Our results show that adversarial images created for a simple substitute BIQA model (i.e. white-box scenario) are transferable as such and able to deceive also several other more complex BIQA models (i.e. black-box scenario). We also investigated some basic defense mechanisms. Our results indicate that re-training BIQA models with a dataset augmented with adversarial images improves robustness of several models, but at the cost of decreased quality prediction accuracy on genuine images.

KW - Image quality assessment

KW - Quality of Experience

KW - Adversarial attacks

KW - Adversarial defenses

UR - http://www.scopus.com/inward/record.url?scp=85140224173&partnerID=8YFLogxK

U2 - 10.1145/3552469.3555715

DO - 10.1145/3552469.3555715

M3 - Published conference contribution

T3 - QoEVMA 2022 - Proceedings of the 2nd Workshop on Quality of Experience in Visual Multimedia Applications

BT - QoEVMA '22

PB - Association for Computing Machinery

T2 - 2nd Workshop on Quality of Experience in Visual Multimedia Applications (QoEVMA) at ACM Multimedia

Y2 - 10 October 2022 through 14 October 2022

ER -

Adversarial Attacks against Blind Image Quality Assessment Models

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this