TY - GEN
T1 - Risk Analysis of One-Pixel Image Defects in Safety-Critical Deep Neural Networks
AU - Radlak, Krystian
AU - Popowicz, Adam
AU - Szczepankiewicz, Michal
AU - Zawistowski, Pawel
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.
PY - 2026
Y1 - 2026
N2 - Deep neural networks (DNNs) are widely considered essential for developing perception systems in autonomous applications. These models are often vulnerable to small perturbations in input data, even if the changes appear negligible to a human observer. This vulnerability introduces an additional risk of failure in safety-critical systems during normal operation. Unfortunately, there is currently no quantitative risk analysis addressing such image defects. In contrast, this work examines the risk that one-pixel defects may occur naturally within image data, and evaluates how frequently such seemingly minor defects can lead to incorrect decisions by neural networks. Extensive experiments reveal that the number of impactful image defects may be relatively high, depending on both the DNN architecture and the dataset used. These findings establish that image defects require significant attention and it might not be sufficient to argue for an acceptable level of safety based solely on the low probability of occurrence these defects.
AB - Deep neural networks (DNNs) are widely considered essential for developing perception systems in autonomous applications. These models are often vulnerable to small perturbations in input data, even if the changes appear negligible to a human observer. This vulnerability introduces an additional risk of failure in safety-critical systems during normal operation. Unfortunately, there is currently no quantitative risk analysis addressing such image defects. In contrast, this work examines the risk that one-pixel defects may occur naturally within image data, and evaluates how frequently such seemingly minor defects can lead to incorrect decisions by neural networks. Extensive experiments reveal that the number of impactful image defects may be relatively high, depending on both the DNN architecture and the dataset used. These findings establish that image defects require significant attention and it might not be sufficient to argue for an acceptable level of safety based solely on the low probability of occurrence these defects.
KW - deep neural networks
KW - image defects
KW - safety
KW - security
UR - https://www.scopus.com/pages/publications/105014764502
U2 - 10.1007/978-3-032-02018-5_40
DO - 10.1007/978-3-032-02018-5_40
M3 - Conference contribution
AN - SCOPUS:105014764502
SN - 9783032020178
T3 - Lecture Notes in Computer Science
SP - 566
EP - 578
BT - Computer Safety, Reliability, and Security. SAFECOMP 2025 Workshops - CoC3CPS, DECSoS, SASSUR, SENSEI, SRToITS, and WAISE, 2025, Proceedings
A2 - Törngren, Martin
A2 - Gallina, Barbara
A2 - Schoitsch, Erwin
A2 - Troubitsyna, Elena
A2 - Bitsch, Friedemann
PB - Springer Science and Business Media Deutschland GmbH
T2 - Co-Design of Communication, Computing and Control in Cyber-Physical Systems, CoC3CPS 2025, 20th Workshop on Dependable Smart Embedded and Cyber-Physical Systems and Systems-of-Systems, DECSoS 2025, 12th International Workshop on Next Generation of System Assurance Approaches for Critical Systems, SASSUR 2025, 4th International Workshop on Safety and Security Interaction, SENSEI 2025, 2nd International Workshop on Safety/Reliability/Trustworthiness of Intelligent Transportation Systems, SRToITS 2025 and 8th International Workshop on Artificial Intelligence Safety Engineering, WAISE 2025 held in conjunction with the 44th International Conference on Computer Safety, Reliability, and Security, SAFECOMP 2025
Y2 - 9 September 2025 through 9 September 2025
ER -