First Learning Steps to Recognize Faces in the Noise

Lukas Lamminger; Heinz Hofbauer; Andreas Uhl

doi:10.1145/3577163.3595105

First Learning Steps to Recognize Faces in the Noise

Lukas Lamminger, Heinz Hofbauer, Andreas Uhl

Artificial Intelligence and Human Interfaces

Research output: Chapter in Book/Report/Conference proceeding/Legal commentary › Chapter in Book › Research › peer-review

Abstract

A UNet-type encoder-decoder inpainting network is applied to weaken the protection strength of selectively encrypted face samples. Based on visual assessment, FaceQNet quality, and ArcFace recognition accuracy the strategy is shown to be successful, however, to a different extent depending on the original protection strength. For almost cryptographic strength, inpainting does not cause a practically relevant protection weakening, while for lower original protection strength inpainting almost removes the protection entirely.

Original language	English
Title of host publication	IH and MMSec 2023 - Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security
Pages	139-144
Number of pages	6
ISBN (Electronic)	9798400700545
DOIs	https://doi.org/10.1145/3577163.3595105
Publication status	Published - Jun 2023

Publication series

Name	IH and MMSec 2023 - Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security

Bibliographical note

Publisher Copyright:
© 2023 Owner/Author.

Keywords

deep learning
denoising
face recognition
inpaiting
selective encryption

Fields of Science and Technology Classification 2012

102 Computer Sciences

Access to Document

10.1145/3577163.3595105

Cite this

@inbook{05ec7318580548f39f36dc937cf05584,

title = "First Learning Steps to Recognize Faces in the Noise",

abstract = "A UNet-type encoder-decoder inpainting network is applied to weaken the protection strength of selectively encrypted face samples. Based on visual assessment, FaceQNet quality, and ArcFace recognition accuracy the strategy is shown to be successful, however, to a different extent depending on the original protection strength. For almost cryptographic strength, inpainting does not cause a practically relevant protection weakening, while for lower original protection strength inpainting almost removes the protection entirely.",

keywords = "deep learning, denoising, face recognition, inpaiting, selective encryption",

author = "Lukas Lamminger and Heinz Hofbauer and Andreas Uhl",

note = "Publisher Copyright: {\textcopyright} 2023 Owner/Author.",

year = "2023",

month = jun,

doi = "10.1145/3577163.3595105",

language = "English",

isbn = "9798400700545",

series = "IH and MMSec 2023 - Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security",

pages = "139--144",

booktitle = "IH and MMSec 2023 - Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security",

}

First Learning Steps to Recognize Faces in the Noise. / Lamminger, Lukas; Hofbauer, Heinz ; Uhl, Andreas.
IH and MMSec 2023 - Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security. 2023. p. 139-144 (IH and MMSec 2023 - Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security).

Research output: Chapter in Book/Report/Conference proceeding/Legal commentary › Chapter in Book › Research › peer-review

TY - CHAP

T1 - First Learning Steps to Recognize Faces in the Noise

AU - Lamminger, Lukas

AU - Hofbauer, Heinz

AU - Uhl, Andreas

PY - 2023/6

Y1 - 2023/6

N2 - A UNet-type encoder-decoder inpainting network is applied to weaken the protection strength of selectively encrypted face samples. Based on visual assessment, FaceQNet quality, and ArcFace recognition accuracy the strategy is shown to be successful, however, to a different extent depending on the original protection strength. For almost cryptographic strength, inpainting does not cause a practically relevant protection weakening, while for lower original protection strength inpainting almost removes the protection entirely.

AB - A UNet-type encoder-decoder inpainting network is applied to weaken the protection strength of selectively encrypted face samples. Based on visual assessment, FaceQNet quality, and ArcFace recognition accuracy the strategy is shown to be successful, however, to a different extent depending on the original protection strength. For almost cryptographic strength, inpainting does not cause a practically relevant protection weakening, while for lower original protection strength inpainting almost removes the protection entirely.

KW - deep learning

KW - denoising

KW - face recognition

KW - inpaiting

KW - selective encryption

UR - http://www.scopus.com/inward/record.url?scp=85165784242&partnerID=8YFLogxK

U2 - 10.1145/3577163.3595105

DO - 10.1145/3577163.3595105

M3 - Chapter in Book

SN - 9798400700545

T3 - IH and MMSec 2023 - Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security

SP - 139

EP - 144

BT - IH and MMSec 2023 - Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security

ER -

First Learning Steps to Recognize Faces in the Noise

Abstract

Publication series

Bibliographical note

Keywords

Fields of Science and Technology Classification 2012

Access to Document

Other files and links

Tools for the Generation of Synthetic Biometric Sample Data

Cite this

First Learning Steps to Recognize Faces in the Noise

Abstract

Publication series

Bibliographical note

Keywords

Fields of Science and Technology Classification 2012

Access to Document

Other files and links

Projects

Tools for the Generation of Synthetic Biometric Sample Data

Cite this