Research On Selective Video Encryption Algorithm Based On Region Of Interest

The importance of content security is being promoted by the various video applications. The main content of the video is usually considered as the areas with particular meaning or impression which are called region of interest(ROI), such as the moving people in surveillance video, the interviewee in news report or the main character in television programs.Scenes with region of interest may not only be related with personal privacy but also involve security and benefits of companies or even countries. So the research on encryption of ROI has long become a hot spot in the research field of video encryption. By only encrypting the content of ROIs, the most important part of video can be protected with less cost and operation. But existing ROI encryption schemes are normally restricted to a fixed area through the whole video and streams with specific functionalities are also required to locate the encryption area in the process of decoding.To solve the problems in existing ROI encryption schemes, a improved selective encryption scheme is firstly proposed in this paper to protect the content of variable ROIs in H.264 video. In this scheme, frames are divided into grid-like slicegroups. For independent slicegroup encoding is supported with the mechanism of flexible macroblock ordering(FMO),encryption on these selected slicegroups will not affect other ones and variable ROI can be formed by different slicegroups in different frames.With encryption on the selected slicegroups in each frames, the security of variable ROI can be guaranteed.In some videos, the region of human face is usually far more important that the background and often needs special protection. A scheme of encryption and decryption on human face ROI is proposed in this paper. In the scheme, through the color-based facial detection, the ROI of a video isautomatically located, and grid-like slicegroups are also used to form and protect the ROI area. In encryption, the mechanisms of facial detection and H.264 coding are both taken advantage of to make the encrypted area characterized enough to be detected in decryption. In this way, streams with specific positional functionalities are no more needed. This scheme can provide encrypted video streams in accordance with H.264 coding grammars and will not lead to compatible problems.Moving objects in the video often attract more attention than the changeless background. So the scheme of encryption and decryption on moving object ROIs is also included in this paper. In this scheme, the moving object ROIs of the video are firstly detected by Gaussian Mixture Model(GMM). Then encryption algorithms are applied individually on the luminance component and chrominance components to achieve good encryption quality and characterized ROIs. So the encrypted area will be located and decrypted automatically in authorized decoder and streams in this scheme are also of compatibility with unauthorized ones.All the schemes in this paper are realized with the H.264 official JM18.4 coding platform, and standard sequences provided in JM manual are used to perform the simulation experiment. The results present the accuracy, compatibility and effectiveness of our scheme in encrypting and decrypting variable ROIs in H.264 video without the need of specific positional functionalities data in stream.