Efficient Privacy-Preserving Anomaly Detection and Localization in Bitstream Video

In cloud computing, videos may be in an encrypted format to protect privacy. Therefore, encrypted video processing is an important application in secure cloud computing. In this paper, we focus on parameter estimation and anomaly detection in an encrypted video bitstream. By analyzing the common properties of video encoding frameworks and the format-compliant encryption schemes, we propose an anomaly detection scheme for encrypted video bitstream with format-compliant encryption. From the encrypted bitstream, we extract three types of complementary features, i.e., the macroblock sizes, the macroblock partitions, and the motion vector difference magnitude, and then propose a method to combine these three features. The proposed detection and localization scheme does not involve video decryption, full decompression, or an interactive protocol, which makes it efficient. Our scheme is also compatible with different video encryption methods. To accelerate the running time, we develop a parallel implementation for our scheme. The experimental results show that our method achieves good running time and detection rate performance.