保护监控视频隐私的漂移失真免疫算法研究

Digital Intelligent Video Surveillance has attracted a wide spread attention, and has been widely applied in every walk of life. However, the lack of centralized control and unified management makes the leakage of privacy a very serious problem in current video surveillance. Since magnanimous compressed videos have to be analyzed and responded in real-time, thus, Real-time Privacy Protection become a vital issue in video surveillance. However, existing pixel-domain based privacy protection approaches which have to decode the compressed video, protect the privacy-sensitive region, and then re-compress the video, do not support real-time response for their high complexity and long time delay. This research is based on the H.264 video coding standard which is widely used in current video surveillance. According to the real-time response requirement of magnanimous compressed videos, this research aims to present a novel transform-domain privacy protection approach without the distortion drift in privacy-insensitive region which are introduced by the scrambling of privacy-sensitive region. This novel approach can be applied on compressed video directly, while avoiding the decoding and re-encoding process and without distortion drift, thus can provide real-time privacy protection for video surveillance. The main research contents are: Theory on intra-frame distortion drift and its compensation method in transform domain; Theory on inter-frame distortion drift and its compensation method in transform domain; Controlled scrambling algorithm in transform domain without distortion drift. This research can protect privacy for video surveillance in real time and provide the technical support of real-time response for magnanimous compressed videos.

当前监控视频隐私泄露问题严重。目前相关研究大多集中在像素域隐私保护领域。而我国监控视频广泛采用H.264压缩格式存储和传输，因此，需研究压缩域隐私保护技术。本项目以H.264压缩视频为隐私保护对象，旨在探索压缩域的对漂移失真免疫的隐私保护算法，加快隐私保护速度。拟基于预研成果，探索用本团队创新性提出的帧内漂移失真免疫理论与压缩域补偿算法、帧间漂移失真免疫理论与压缩域补偿算法、帧内帧间漂移失真均免疫的压缩域受控安全扰乱算法，解决压缩域隐私保护的帧内漂移失真、帧间漂移失真、以及无漂移失真的压缩域扰乱等三个方面的关键科学问题。研究成果可为H.264视频快速隐私保护提供理论依据和技术方法，有利于视频监控应用的实时响应。

为了解决监控视频的隐私泄漏问题，本项目研究了压缩视频的隐私保护技术，针对帧内失真漂移问题、帧间失真漂移问题和压缩域扰乱问题进行探索，在调研分析后率先提出了视频的无损隐私区域保护概念，并提出了一系列监控视频的无损隐私保护方案。具体包括具高度视觉安全性的压缩域可逆扰乱算法;帧内失真漂移免疫的监控视频可逆隐私保护方案;无帧内与帧间失真漂移的H.264压缩视频可逆隐私保护方案等。所提出的方案可保护隐私区域安全，同时保持非隐私区域视觉效果不变以支持监控视频异常检测，且授权用户可完全恢复在摄像头端录制的原始视频。研究成果特别适合于云视频监控，以及对数据完整性要求较高的应用领域，如执法、金融和医疗等。