Rapid advances in computer vision have enabled powerful visual perception systems in areas such as surveillance,autonomous driving,healthcare,and augmented reality.However,these systems often raise serious privacy con...Rapid advances in computer vision have enabled powerful visual perception systems in areas such as surveillance,autonomous driving,healthcare,and augmented reality.However,these systems often raise serious privacy concerns due to their ability to identify and track individuals without consent.This paper explores the emerging field of identity-hiding visual perception,which aims to protect personal identity within visual data through techniques such as anonymization,obfuscation,and privacy-aware modeling.We provide a system-level overview of current technologies,categorize application scenarios,and analyzemajor challenges-particularly the trade-off between privacy and utility,technical complexity,and ethical risks.Furthermore,we examine regulatory trends and propose future research directions,including model-level privacy mechanisms such as federated learning and machine unlearning.By synthesizing insights across technical,ethical,and policy dimensions,this work offers a conceptual roadmap for developing responsible,privacy-preserving visual perception systems.展开更多
基金funded by the Guangxi Natural Science Foundation(Grant No.2024GXNSFAA010484)the National Natural Science Foundation of China(No.62466013).
文摘Rapid advances in computer vision have enabled powerful visual perception systems in areas such as surveillance,autonomous driving,healthcare,and augmented reality.However,these systems often raise serious privacy concerns due to their ability to identify and track individuals without consent.This paper explores the emerging field of identity-hiding visual perception,which aims to protect personal identity within visual data through techniques such as anonymization,obfuscation,and privacy-aware modeling.We provide a system-level overview of current technologies,categorize application scenarios,and analyzemajor challenges-particularly the trade-off between privacy and utility,technical complexity,and ethical risks.Furthermore,we examine regulatory trends and propose future research directions,including model-level privacy mechanisms such as federated learning and machine unlearning.By synthesizing insights across technical,ethical,and policy dimensions,this work offers a conceptual roadmap for developing responsible,privacy-preserving visual perception systems.
