Visually Undetectable Real-World Attacks on CMOS Sensors for Deep Neural Networks

Adir Hazan; Zvi Stein; Adrian Stern

doi:10.1364/isa.2025.iw4d.3

Visually Undetectable Real-World Attacks on CMOS Sensors for Deep Neural Networks

Adir Hazan
, Zvi Stein
, Adrian Stern

School of Electrical and Computer Engineering

Ben-Gurion University of the Negev

Research output: Contribution to conference › Paper › peer-review

Abstract

We overview a recent invisible optical-based adversarial attack that exploits the rolling shutter mechanism in CMOS cameras. We degrade DNN performance using designed light pulse sequences while satisfying photopic invisibility constraints in real-world physical environments.

Original language	English
DOIs	https://doi.org/10.1364/isa.2025.iw4d.3
State	Published - 1 Jan 2025
Event	2025 Imaging Systems and Applications, IS 2025 - Seattle, United States Duration: 18 Aug 2025 → 21 Aug 2025

Conference

Conference	2025 Imaging Systems and Applications, IS 2025
Country/Territory	United States
City	Seattle
Period	18/08/25 → 21/08/25

ASJC Scopus subject areas

General Computer Science
Space and Planetary Science
Control and Systems Engineering
Electrical and Electronic Engineering
Electronic, Optical and Magnetic Materials
Instrumentation

Access to Document

10.1364/isa.2025.iw4d.3

Cite this

@conference{78d3fc26342441e9bb3dcad9623d5f53,

title = "Visually Undetectable Real-World Attacks on CMOS Sensors for Deep Neural Networks",

abstract = "We overview a recent invisible optical-based adversarial attack that exploits the rolling shutter mechanism in CMOS cameras. We degrade DNN performance using designed light pulse sequences while satisfying photopic invisibility constraints in real-world physical environments.",

author = "Adir Hazan and Zvi Stein and Adrian Stern",

note = "Publisher Copyright: {\textcopyright} 2025 The Author(s); 2025 Imaging Systems and Applications, IS 2025 ; Conference date: 18-08-2025 Through 21-08-2025",

year = "2025",

month = jan,

day = "1",

doi = "10.1364/isa.2025.iw4d.3",

language = "English",

}

TY - CONF

T1 - Visually Undetectable Real-World Attacks on CMOS Sensors for Deep Neural Networks

AU - Hazan, Adir

AU - Stein, Zvi

AU - Stern, Adrian

PY - 2025/1/1

Y1 - 2025/1/1

N2 - We overview a recent invisible optical-based adversarial attack that exploits the rolling shutter mechanism in CMOS cameras. We degrade DNN performance using designed light pulse sequences while satisfying photopic invisibility constraints in real-world physical environments.

AB - We overview a recent invisible optical-based adversarial attack that exploits the rolling shutter mechanism in CMOS cameras. We degrade DNN performance using designed light pulse sequences while satisfying photopic invisibility constraints in real-world physical environments.

UR - https://www.scopus.com/pages/publications/105032186075

U2 - 10.1364/isa.2025.iw4d.3

DO - 10.1364/isa.2025.iw4d.3

M3 - Paper

T2 - 2025 Imaging Systems and Applications, IS 2025

Y2 - 18 August 2025 through 21 August 2025

ER -

Visually Undetectable Real-World Attacks on CMOS Sensors for Deep Neural Networks

Abstract

Conference

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this