TY - JOUR
T1 - PrivacyGuard
T2 - Collaborative Edge-Cloud Computing Architecture for Attribute-Preserving Face Anonymization in CCTV Networks
AU - Jeremiah, Sekione Reward
AU - Ha, Jimin
AU - Singh, Sushil Kumar
AU - Park, Jong Hyuk
N1 - Publisher Copyright:
© This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
PY - 2024
Y1 - 2024
N2 - Closed-circuit television (CCTV) footage, a vital component of security and law enforcement, also poses a potential privacy threat. To mitigate this, it’s crucial to anonymize personal identifiers in the footage, preventing their unauthorized or unintended use or disclosure. An effective anonymization system should remove sensitive identity information while preserving attributes necessary for other identity-agnostic tasks. Among several biometric personal identifiers, the face is the primary one. However, anonymizing facial images while retaining facial attributes is a challenging task. Furthermore, neural networks, instrumental in face detection and recognition, often come with significant computational demands, making them unsuitable for resource-constrained edge devices. In a novel approach, our study proposes PrivacyGuard, a modular architecture with two major parts: the image quality assessment and dispatching module, and the attribute-preserving anonymization module. The former categorizes images by quality, sending low-quality images to cloud servers and processing high-quality ones on local edge servers. Simultaneously, the anonymization module leverages lightweight NN models on the edge to generate obfuscated images while preserving the relevant attributes for downstream machine-learning tasks. Experimental results on a diverse dataset (i.e., ethnicities, genders, etc.) demonstrate the model’s effectiveness in quality assessment and anonymization while retaining the relevant attributes. The anonymization model achieves an attribute preservation rate of 98%, along with obfuscated identities.
AB - Closed-circuit television (CCTV) footage, a vital component of security and law enforcement, also poses a potential privacy threat. To mitigate this, it’s crucial to anonymize personal identifiers in the footage, preventing their unauthorized or unintended use or disclosure. An effective anonymization system should remove sensitive identity information while preserving attributes necessary for other identity-agnostic tasks. Among several biometric personal identifiers, the face is the primary one. However, anonymizing facial images while retaining facial attributes is a challenging task. Furthermore, neural networks, instrumental in face detection and recognition, often come with significant computational demands, making them unsuitable for resource-constrained edge devices. In a novel approach, our study proposes PrivacyGuard, a modular architecture with two major parts: the image quality assessment and dispatching module, and the attribute-preserving anonymization module. The former categorizes images by quality, sending low-quality images to cloud servers and processing high-quality ones on local edge servers. Simultaneously, the anonymization module leverages lightweight NN models on the edge to generate obfuscated images while preserving the relevant attributes for downstream machine-learning tasks. Experimental results on a diverse dataset (i.e., ethnicities, genders, etc.) demonstrate the model’s effectiveness in quality assessment and anonymization while retaining the relevant attributes. The anonymization model achieves an attribute preservation rate of 98%, along with obfuscated identities.
KW - Face Anonymization
KW - Privacy
KW - Privacy Protection
KW - Security and Forensics
KW - Usability
KW - Video De-identification
UR - http://www.scopus.com/inward/record.url?scp=85202298085&partnerID=8YFLogxK
U2 - 10.22967/HCIS.2024.14.043
DO - 10.22967/HCIS.2024.14.043
M3 - Article
AN - SCOPUS:85202298085
SN - 2192-1962
VL - 14
JO - Human-centric Computing and Information Sciences
JF - Human-centric Computing and Information Sciences
M1 - 43
ER -