Privacy Information Protection And Hiding For Surveillance Video

Video surveillance system plays a important role in preventing crime and creating a safe city/community. However, it brings a problem that personal privacy information may be leaked. Thus, the privacy protection of video information is not only gaining public’s considerable concerns, but also attracting extensive research in the field of video information security. In this paper, we focus on the privacy protection problem in surveillance areas with some relatively fixed personnel, such as companies, hospitals, etc, rather than squares or other public places.In this paper, an improved privacy information protection framework is proposed. It discriminates fixed persons from those ordinary persons. The fixed persons are removed from surveillance video, and its information is embedded into the inpainted video as a watermark. The privacy information is embedded into surveillance video as a reversible watermark, and only legal user with key can extract it. The main issues of this scheme are two-fold:firstly, the reconstruction of the damage area within ordinary persons after protecting the privacy; secondly, the hiding of large amount of privacy data.In order to inpaint the hole left by the removal of the privacy object in the video, a novel object-based video inpainting algorithm is presented. It applies different strategies to handle static and dynamic portions of the hole. To inpaint the dynamic moving foreground regions, it introduces a novel sliding-window based dissimilarity measure,and transforms the inpainting problem into a minimization problem under a dynamic programming framework. Its inpainting strategy is based on the semantic object rather than the patch, providing a significant computational advantage by avoiding exhaustive search. Its inpainting process is based on a sliding window rather than a single video frame, providing a continuous movement. This technique can effectively inpaint large regions of occlusions, objects that are completely missing for several frames, or changing in size and pose.For watermark embedding, it exploits the fact that the quantized DCT coefficients of H.264/AVC coded video follows Cauchy distribution. After a classification of DCT coefficients, histogram shifting algorithm is utilized to realize high-capacity watermark embedding. Aim at the serious bitrate increase caused by the embedding process in the zero coefficients, it analyzes the H.264/AVC coding theory, and creates a function which takes a comprehensive consideration of the increase in bitrate and visual distortion. Under certain constraints, this function is minimized in order to obtain optimal rate-distortion. Experimental results show that the proposed reversible watermarking-based privacy information protection framework can hide the privacy information without significant distortion and bitrate increase. Moreover, the authorized users can lossless recover the original video by watermark extraction.