| 引用本文： |
-
李英龙,李星星,陈红,陈炳元,陈铁明.视觉数据隐私检测研究综述[J].信息安全学报,已采用 [点击复制]
- LI Yinglong,LI Xingxing,CHEN Hong,CHEN Bingyuan,CHEN Tieming.A Research Review on Privacy Detection in Visual Data[J].Journal of Cyber Security,Accept [点击复制]
|
|
| 摘要: |
| 随着社交网络快速发展和便携式拍摄设备的普及，日常生活产生的大量的视觉数据会被公共平台收集和分享。然而，在这些视觉数据得以广泛传播的同时，其固有的隐私安全问题亦日益凸显。由于视觉数据通常包含个人隐私信息，一旦被恶意利用或泄露，极易导致严重的个人隐私安全问题。针对这一问题，视觉数据隐私检测作为识别和缓解此类风险的关键技术环节，通过运用深度学习模型分析视觉内容，旨在识别图像及视频中潜在的隐私敏感区域或信息，从而为后续的隐私保护机制提供必要的基础。因此，视觉数据隐私检测研究已成为计算机视觉和信息安全领域一个重要的研究热点。本文对视觉数据隐私检测的研究现状进行了全面综述。首先，本文指出现有视觉数据隐私领域的综述研究普遍缺少对隐私检测工作的系统性探讨。然后，探讨了隐私度量的概念并指出这一体系缺乏统一标准的问题，深入分析了视觉数据隐私检测的特点及目前面临的研究挑战。接着，通过回顾近年来视觉数据隐私检测领域的研究进展，系统分析了现有的视觉隐私数据集，对比了它们在数据规模、标注粒度等多个方面的差异;并基于视觉内容、视觉特征以及多模态信息等不同的技术视角，对现有的视觉数据隐私检测方法进行了细致的总结与分类。此外，还探讨了视觉数据隐私检测技术在社交网络、监控视频、城市交通以及定向广告等典型应用场景中的实际应用情况。最后，本文总结了关于视觉数据隐私检测的研究现状并对该领域未来的发展趋势进行了展望。 |
| 关键词: 视觉隐私 隐私度量 隐私数据集 深度学习 |
| DOI： |
| 投稿时间：2025-05-15修订日期：2025-08-18 |
| 基金项目:国家自然科学基金重点项目 |
|
| A Research Review on Privacy Detection in Visual Data |
|
LI Yinglong1, LI Xingxing1, CHEN Hong2, CHEN Bingyuan1, CHEN Tieming1
|
| (1.Zhejiang University of Technology;2.Renmin University of China) |
| Abstract: |
| With the rapid development of social networks and the widespread adoption of portable image capture devices, a substantial amount of visual data generated in daily life is collected and shared on public platforms. However, while this visual data is broadly disseminated, its inherent privacy and security risks have become increasingly prominent. Since that visual data often contains personal privacy information, malicious use or leakage can easily lead to severe privacy breaches. To address this challenge, visual data privacy detection, serving as a crucial technical step in identifying and mitigating such risks, em-ploys deep learning models to analyze visual content, aiming to recognize potential privacy-sensitive regions or information within images and videos, thereby providing a necessary foundation for subsequent privacy protection mechanisms. As a result, research on visual data privacy detection has become an important research hotspot in the fields of computer vision and information security. This paper presents a comprehensive review of the current research status in visual data privacy detection. First, this paper points out the current situation where existing surveys on visual data privacy research lack cov-erage of privacy detection work. Then, it discusses the concept of privacy metrics and points out the lack of unified standards in this system, and deeply analyzes the characteristics of visual data privacy detection and the current research challenges. Subsequently, by reviewing recent research advances in the field of visual data privacy detection, this paper systematically analyzes existing visual privacy datasets, comparing their differences across multiple dimensions such as data scale and annotation granularity; and provides a systematic classification and detailed analysis of current visual data privacy detection methods, examining three technical approaches: visual content analysis, visual feature extraction, and multimodal infor-mation processing. Additionally, this paper explores the practical applications of visual data privacy detection technology in typical application scenarios such as social networks, surveillance videos, urban transportation, and targeted advertising. Finally, this paper summarizes the current research landscape in visual data privacy detection and outlines future develop-ment trends. |
| Key words: visual privacy privacy measurement privacy dataset deep learning |
